Category: AI for Ecommerce

Working With AI Is a Leadership Skill, Not a Coding Skill

Many ecommerce and marketing leaders quietly believe AI is not for them.

They see developers using coding agents, hear unfamiliar terms such as models, context windows, APIs, and embeddings, and assume they arrived late to a technical revolution.

That conclusion is understandable—and mostly wrong.

You do not need to know how to train a model to use AI effectively. You need to know how to lead work.

The strongest AI users are not always the people who write the best code. They are often the people who can define the outcome, provide the right context, delegate a bounded task, assess the result, give useful feedback, and intervene when the work moves in the wrong direction.

Those are leadership behaviours.

For an ecommerce founder, CMO, growth lead, or marketing manager, the opportunity is not to become a part-time machine-learning engineer. It is to become better at three things you should already be doing with human teams:

Writing clear briefs.
Building feedback loops.
Knowing when to override.

AI makes those behaviours more visible because it responds immediately. A vague brief produces confident vagueness in seconds. Weak review allows mistakes to scale. Good leadership turns the same technology into useful, repeatable leverage.

Why AI looks like a coding skill

AI entered many workplaces through technical teams. Developers adopted coding assistants early because software work is structured, digital, and easy to test. That visibility created a misleading picture: people who code appear to be the people who “understand AI.”

Technical literacy is valuable. Some AI projects genuinely require developers, data specialists, security professionals, or integration engineers. But using AI for everyday ecommerce and marketing work is a different challenge.

Consider these tasks:

Turning customer research into campaign angles.
Drafting a product-page brief.
Comparing competitor positioning.
Creating first-pass lifecycle email variations.
Summarising reviews into recurring objections.
Preparing an experiment plan.
Auditing product data for missing information.
Converting a strategy into a sequence of team actions.

The limiting factor is rarely code. It is whether the person directing the work understands the customer, objective, constraints, evidence, quality standard, and decision that the output must support.

That is why an experienced marketer with clear judgment can outperform a technically impressive user who gives AI poor direction and accepts fluent output too easily.

What the evidence actually says

The claim that AI use is a leadership skill is a practical interpretation—not a scientific finding that coding no longer matters.

However, research supports the underlying logic.

A field experiment involving 758 consultants found that AI improved speed and quality on tasks inside its capability boundary. On a task outside that boundary, participants using AI were less likely to reach the correct answer. The researchers call this uneven boundary the “jagged technological frontier.” The lesson is not simply “use AI.” It is to decide where AI is useful, how work should be divided, and when human judgment must take control.

The study also observed two broad working patterns. “Centaurs” divided tasks between the human and AI, while “cyborgs” moved back and forth between human and machine contributions. Both patterns are forms of delegation and workflow design—not merely prompting.

Microsoft’s 2025 Work Trend Index, based partly on a survey of 31,000 workers across 31 countries, describes an emerging “agent boss” role: people create, delegate to, and manage AI agents. Microsoft reported that 46% of surveyed leaders said their organisations were already using agents to automate some workflows, particularly in customer service, marketing, and product development.

Those numbers come from a vendor-sponsored survey and should not be treated as neutral proof that every company is adopting AI successfully. They do, however, reinforce the direction of travel: AI is moving from individual chat sessions into managed work.

Meanwhile, NIST’s Generative AI Profile frames responsible AI adoption as an organisational risk-management problem. Governance, measurement, testing, documentation, and human oversight matter because fluent output can still be inaccurate, biased, unsafe, or unsuitable for its intended context.

In plain business language: AI can perform work, but someone must still lead it.

Leadership behaviour 1: Write clear briefs

Most “prompt engineering” advice makes the subject sound more technical than it needs to be.

A strong prompt is usually a strong brief.

If you can brief a copywriter, designer, analyst, developer, or agency properly, you already have the foundation for working with AI. The difference is that AI will rarely interrupt to expose what you forgot. It will fill the gaps itself, often with plausible assumptions.

The weak ecommerce brief

Write a high-converting product description for this serum.

This provides no reliable definition of “high-converting.” It does not identify the audience, product facts, evidence, brand voice, channel, objective, length, forbidden claims, or desired action.

The AI must guess. Its response may sound polished while being commercially useless—or legally risky.

The leadership brief

Write a 140-word product-page section for skincare-aware women aged 28–45 who want a simple evening routine. Use only the supplied product facts. Lead with the lightweight texture and fragrance-free formula. Explain the role of the listed ingredients without making medical, clinical, or guaranteed-result claims. Use a warm, specific, low-hype tone. Return one headline, one short paragraph, three bullets, and a two-word CTA. The goal is to reduce uncertainty before add-to-cart.

That brief does five leadership jobs:

Defines the audience.
Defines the business objective.
Supplies the relevant context.
Establishes constraints.
Describes what “done” looks like.

No coding is required. But judgment is required at every step.

A reusable AI brief for ecommerce teams

Before assigning a task, answer these seven questions:

Outcome: What decision, behaviour, or business result should this work support?
Audience: Who will use or see the output?
Context: Which facts, examples, brand rules, and source materials does AI need?
Deliverable: What exactly should be returned?
Constraints: What must it include, avoid, preserve, or never infer?
Quality bar: What separates acceptable from excellent?
Approval: Who checks the result before it affects customers, spend, data, or revenue?

If a leader cannot answer those questions, the problem is not the prompt. The work itself is underspecified.

Leadership behaviour 2: Build feedback loops

Weak AI users treat the first response as the finished product. Strong AI users treat it as the first draft in a managed loop.

The difference is familiar to anyone who has led a good creative or commercial team.

Bad feedback sounds like this:

Make it better.

Useful feedback identifies the gap:

The structure is correct, but the opening is generic and the benefits are not tied to the customer’s evening routine. Keep the factual claims and length unchanged. Rewrite the headline and first paragraph with a calmer, more specific tone. Avoid words such as revolutionary, flawless, and game-changing.

That instruction separates what should change from what must remain stable.

Ecommerce example: a Black Friday campaign

Imagine AI produces a campaign concept built around the largest discount. The copy is competent, but your customer research shows loyal buyers respond more strongly to early access and limited bundles than aggressive price language.

A productive feedback loop might be:

Ask AI to explain the assumptions behind the first concept.
Provide the customer-research finding and previous campaign data.
Request three revised concepts built around early access, bundles, and convenience.
Score them against brand fit, margin impact, operational feasibility, and testing clarity.
Ask for a final brief based on the selected direction.
Test the concept with real customers and feed the result into the next cycle.

The AI supplies speed and range. The leader supplies evidence, priorities, and correction.

Feedback needs a scoreboard

AI workflows improve when the team evaluates output against explicit criteria rather than personal reactions.

For ecommerce copy, the scorecard might include:

Criterion	Question
Accuracy	Does every claim come from approved product information?
Customer relevance	Does the message address a known need, objection, or motivation?
Brand fit	Could this appear on our store without sounding imported from another company?
Clarity	Can a customer understand the value without decoding jargon?
Commercial fit	Does the copy support the offer, margin, inventory, and campaign goal?
Compliance	Has the appropriate human reviewed sensitive or regulated claims?
Testability	Can we measure whether the change improved the intended outcome?

This turns “I don’t like it” into operational feedback that a person or AI can use.

Leadership behaviour 3: Know when to override

Delegation without override is abdication.

AI can produce recommendations, drafts, analyses, forecasts, and actions. It cannot carry your accountability to the customer, team, regulator, or bank account.

The most important AI skill may be recognising the moment when efficiency must stop and human judgment must take over.

Override when the cost of being wrong is high

A sensible ecommerce rule is:

The greater the effect on money, customer rights, health, privacy, access, or brand trust, the stronger the human review.

Human approval should be explicit for tasks such as:

Publishing health, performance, environmental, or comparative product claims.
Changing prices, discounts, tax, shipping, refunds, or subscription terms.
Sending sensitive customer-service responses.
Acting on personal customer data.
Allocating significant advertising spend.
Editing checkout logic or payment workflows.
Making hiring, disciplinary, or supplier decisions.
Deleting records or changing source-of-truth systems.

The goal is not to review every comma forever. It is to match control to risk.

Override when AI is confidently outside its frontier

The jagged-frontier research matters because an AI answer does not become visibly less fluent when the task exceeds its capability. Confidence of presentation is not confidence of truth.

Stop or redirect the workflow when:

The model cannot identify the source of a factual claim.
The answer changes materially after small prompt changes.
It invents data to complete an incomplete brief.
It optimises the requested metric while ignoring a business constraint.
Its recommendation conflicts with customer evidence or operational reality.
The output cannot be tested or explained.
Nobody on the team is qualified to review the result.

Sometimes the correct leadership decision is not a better prompt. It is assigning the task to a human specialist.

What this looks like inside an ecommerce team

AI leadership becomes concrete when tasks have different levels of autonomy.

Work type	AI’s role	Human leader’s role
Review summarisation	Extract themes and quote candidates	Verify samples, identify bias, choose priorities
Product-copy draft	Produce structured options from approved facts	Approve claims, brand voice, and final copy
Campaign ideation	Generate angles and variations	Set strategy, margin limits, audience, and test plan
Marketing brief	Convert evidence into a structured first draft	Resolve trade-offs and approve the direction
Customer-service draft	Suggest a response using policy context	Review sensitive cases and exceptions
Performance analysis	Surface patterns and hypotheses	Validate data and decide what action is justified
Workflow automation	Execute defined low-risk steps	Set permissions, monitoring, escalation, and shutdown rules

Notice the pattern: AI handles expansion, transformation, summarisation, and repetition. Leaders retain intent, trade-offs, exceptions, and accountability.

You may already be better prepared than you think

Non-technical leaders often underestimate the value of their existing skills.

If you can do the following, you are not starting from zero:

Turn a vague request into a clear objective.
Explain who the customer is and what they care about.
Distinguish evidence from assumption.
Give specific feedback without rewriting everything yourself.
Decide what deserves speed and what deserves caution.
Coordinate specialists around a shared outcome.
Take responsibility for the final decision.

Those skills are difficult to automate because they depend on context, organisational knowledge, customer understanding, and consequences.

The technical layer will continue to become easier. The leadership layer becomes more important because AI increases the amount of work a person can initiate. When execution becomes abundant, direction and quality control become scarce.

A 30-minute exercise for your next team meeting

Choose one recurring ecommerce task—such as a weekly campaign brief, product-page draft, review summary, or performance report—and map it using this process:

Define the outcome. What business decision or customer action should the work support?
List the inputs. Which approved facts, data, examples, and constraints are required?
Delegate one bounded step. Give AI a task that can be reviewed independently.
Create the scorecard. Agree on factual, brand, commercial, and compliance checks.
Run one feedback cycle. Identify the gap and request a controlled revision.
Set the override. Decide which condition stops automation or requires a specialist.
Record the result. Compare time saved, corrections required, and business usefulness.

Do not begin by buying ten AI tools. Begin by leading one workflow properly.

Final takeaway

The best AI users are not people who know the most technical vocabulary. They are people who make the work clear.

They brief with context. They delegate with boundaries. They review against evidence. They correct without creating confusion. They override when the cost of a mistake exceeds the value of speed.

That is why ecommerce and marketing leaders should stop saying, “AI is not for me because I am not technical.”

You do not need to become a coder.

You need to become the kind of leader who can direct intelligence—human or artificial—toward a useful outcome without surrendering judgment.

AI fluency matters. Technical expertise matters. But the skill that turns AI into business value is leadership.

Frequently Asked Questions

Do ecommerce leaders need coding skills to use AI?

Not for most everyday marketing, research, analysis, copy, and workflow-design tasks. Technical specialists remain important for integrations, security, data infrastructure, and custom systems, while business leaders must define objectives, constraints, quality standards, and approval rules.

What is the most important AI leadership skill?

Clear task definition is the foundation. A leader must explain the outcome, audience, context, deliverable, constraints, quality bar, and approval process. Feedback and override judgment determine whether the work improves safely.

Can AI make ecommerce decisions automatically?

AI can support or automate bounded decisions, but autonomy should reflect risk. Decisions affecting prices, customer rights, personal data, regulated claims, significant spend, or revenue-critical systems need stronger controls and accountable human oversight.

Why does AI sometimes make good work worse?

AI capabilities are uneven. Research describes a “jagged technological frontier”: AI improves performance on some tasks but can reduce accuracy on tasks outside its capability boundary. Polished language can make those errors harder to notice.

How should a marketing team start using AI?

Choose one recurring, reviewable workflow. Define its inputs and desired outcome, delegate one bounded step, create a quality scorecard, run a feedback loop, and establish conditions for human override. Measure editing time and business usefulness before scaling.

Your Shopify Dev Retainer Is About to Shrink—Here’s What to Fund Instead

There is a viral argument moving through software circles: AI is “removing the middle class of software engineering.”

The original thesis is more nuanced than “AI will replace developers.” Florian Herrengt argues that AI removes the natural speed limit on producing code. Strong engineers can move faster—but weak decisions can also become thousands of lines of unreviewable complexity before anyone asks whether the system should have been built that way at all.

For ecommerce owners, this is not an abstract debate. It changes what you should expect from the people building your Shopify theme, checkout extensions, apps, integrations, tracking stack, and internal tools.

Here is the practical version:

Routine Shopify implementation is becoming cheaper. Trusted technical judgment is becoming more valuable.

Your development budget may shrink, but cutting it blindly would be a mistake. The smarter move is to stop paying premium rates for repetitive production and redirect that money toward conversion, customer experience, reliable data, security, and experiments that generate measurable business value.

First, the “AI does 80% of the work” claim is not a fact

You will hear confident predictions that AI can now handle 80% of software development. Treat that number as a scenario, not a verified industry benchmark.

The evidence supports a more careful conclusion. GitHub’s coding agents can already research repositories, fix bugs, implement incremental features, improve tests, update documentation, and open pull requests. Shopify now provides an AI Toolkit that connects supported coding assistants to current Shopify documentation, API schemas, code validation, and store-management workflows.

Shopify is also reducing the amount of code needed at the merchant level. Horizon themes can generate custom theme blocks from natural-language prompts, while Sidekick can modify store settings and theme elements through conversation.

But faster code generation does not automatically mean faster, safer delivery. DORA’s research describes AI as an amplifier of an organisation’s existing strengths and weaknesses. Its earlier generative-AI analysis found that increased adoption could coincide with lower delivery throughput and stability when teams allowed larger batches of generated code to overwhelm review and operational discipline.

So the honest answer is not “AI replaces 80% of your developers.” It is:

AI can absorb a large share of routine production, but the percentage depends on the task, the store, the quality of its systems, and the people reviewing the output.

The Shopify work that is becoming cheaper

Many ecommerce retainers are still priced around hours consumed rather than business risk or value created. AI puts pressure on that model because it compresses the time required for predictable tasks.

1. Basic theme changes

Creating sections, adjusting layouts, editing Liquid templates, modifying CSS, adding schema settings, and making responsive fixes are increasingly AI-assisted. Shopify’s own AI block-generation features push some of this work directly into the merchant’s theme editor.

A developer may still need to check accessibility, performance, browser behaviour, theme compatibility, and maintainability. But the raw production time should fall.

2. Repetitive app and integration scaffolding

AI can generate API clients, webhook handlers, GraphQL queries, data mappings, validation logic, test fixtures, and documentation. Shopify’s AI Toolkit can validate GraphQL, Liquid, and extension code against platform-specific schemas, reducing time spent searching documentation or correcting basic platform mistakes.

The integration is not suddenly risk-free. Authentication, retries, rate limits, idempotency, data ownership, failure recovery, and privacy still require deliberate design. Yet merchants should no longer accept the same price for boilerplate that once took days and can now be drafted in hours.

3. QA preparation and routine bug fixing

AI agents can propose test cases, reproduce common issues, inspect logs, write unit tests, and repair contained defects. Humans remain responsible for deciding whether the test coverage reflects real customer journeys and whether a “fix” creates a new commercial or operational problem.

4. Migration and maintenance work

Platform migrations often contain repeatable transformations. In 2026, Shopify demonstrated its AI Toolkit helping developers move checkout and customer-account extensions to newer components. That does not eliminate migration planning, but it can remove substantial mechanical effort.

The same principle applies to dependency updates, code cleanup, documentation, and straightforward technical-debt tickets.

The work store owners will still pay well for

The valuable ecommerce developer is no longer the person who types code fastest. It is the person who can decide what should exist, understand how it affects the business, and prove that it works safely.

Commerce architecture

Should you use an off-the-shelf app, a Shopify Function, a checkout UI extension, a custom app, or no new technology at all? AI can list options. An experienced engineer must evaluate total cost, platform constraints, lock-in, failure modes, and the capabilities of your team.

That decision matters because Shopify checkout customisation operates through governed extension surfaces such as UI extensions, Functions, web pixels, and payment extensions. Shopify Functions can customise backend logic for discounts, delivery options, and cart or checkout validation, but the implementation must respect Shopify’s platform model.

Checkout and revenue-risk decisions

A checkout change can affect conversion, payment acceptance, tax, fraud, discounts, fulfilment, analytics, and customer support at the same time. Generating a checkout component is not the hard part. Understanding the revenue consequences is.

Store owners will continue paying for people who can model those consequences, design rollback plans, monitor releases, and take responsibility when money is on the line.

Integration reliability

Connecting two APIs is easy in a demo. Operating that connection through traffic spikes, partial outages, duplicate webhooks, expired credentials, schema changes, and inconsistent data is engineering.

The premium shifts from “we connected the systems” to “we can prove orders and customer data remain correct when something fails.”

Security, privacy, and compliance

AI can generate insecure code just as efficiently as secure code. Ecommerce systems handle personal data, payments, discounts, accounts, and commercially sensitive information. Threat modelling, permission design, dependency review, audit trails, and incident response remain accountable human work.

Measurement and experimentation

The highest-value ecommerce teams will connect technical changes to a measurable hypothesis. They will define the metric, preserve analytics integrity, control the experiment, interpret the result, and decide what to do next.

AI makes it cheaper to build variants. It does not automatically tell you whether a conversion lift is real, profitable, or caused by broken tracking.

Who is left on the ecommerce development team?

The team becomes smaller at the production layer and stronger at the judgment layer.

Role	What changes	Why the role remains valuable
Ecommerce technical lead	Reviews more AI-produced work and writes less routine code	Owns architecture, quality, risk, and trade-offs
Shopify specialist	Uses AI to build themes, Functions, apps, and extensions faster	Knows Shopify’s constraints and chooses the right extension surface
Conversion/product strategist	Runs more experiments with shorter build cycles	Connects customer behaviour to commercial outcomes
Data and analytics engineer	Automates implementation but strengthens validation	Protects attribution, event quality, and decision accuracy
Security/reliability specialist	Uses AI for scanning and remediation support	Designs controls and handles failures that affect revenue or trust
Junior developer	Produces faster with guidance and learns through review	Builds the future talent pipeline and handles bounded tasks

The endangered role is not simply “mid-level developer.” It is the person whose value is limited to translating a well-specified ticket into ordinary code without owning the outcome.

That work will not disappear overnight, but it will face aggressive price compression.

What should you do with the budget you save?

Assume your current development retainer falls by 20%, 30%, or even 50%. Do not book the entire reduction as profit before you know why the cost fell. Reinvest a deliberate portion into capabilities that AI makes more valuable.

1. Fund a conversion experiment pipeline

Build a ranked backlog of hypotheses across product pages, merchandising, search, cart, checkout, post-purchase, retention, and customer service. Require every development task to state:

The customer or business problem.
The expected measurable effect.
The minimum viable change.
The tracking and QA plan.
The rollback condition.

Cheaper production should produce more learning—not more random features.

2. Fix your data foundation

Audit customer events, consent, channel attribution, server-side tracking, product feeds, order data, and business reporting. If your inputs are unreliable, AI only helps you make incorrect decisions faster.

3. Pay for senior review, not permanent senior typing

You may need fewer developer hours but more concentrated expertise. Buy architecture reviews, security reviews, release oversight, performance audits, and incident readiness. A few hours of strong judgment can prevent months of generated complexity.

4. Improve speed and resilience

Measure storefront performance, app weight, integration failure rates, checkout errors, deployment frequency, recovery time, and escaped defects. AI-created output should pass the same—or stricter—quality gates as human-written output.

5. Invest in proprietary advantage

Do not spend your savings rebuilding commodity features. Invest in systems tied to your unique operations or customer knowledge: smarter merchandising rules, distinctive bundles, better lifecycle personalisation, internal decision tools, or a superior service workflow.

If a competitor can generate the same feature from the same prompt, it is not a durable advantage.

A better way to structure your next Shopify retainer

The traditional retainer sells access to a pool of hours. The AI-era retainer should sell controlled outcomes.

Ask agencies and developers to separate proposals into four layers:

Commodity production: theme edits, scaffolding, routine fixes, migrations, and documentation.
Expert decisions: architecture, platform selection, data design, security, and performance.
Validation: automated tests, manual QA, analytics verification, code review, monitoring, and rollback.
Commercial outcomes: experiments, conversion improvements, cost reduction, reliability, and customer experience.

Then ask five direct questions:

Where will AI reduce delivery time, and is that saving reflected in the price?
Who is accountable for reviewing generated code?
How will you test platform constraints, security, accessibility, and performance?
What happens when an integration or release fails?
Which business metric is this work expected to improve?

An agency should not charge yesterday’s hours for today’s automation. A merchant should not demand bargain pricing while expecting someone else to carry unlimited revenue risk. The fair model prices routine output lower and accountable expertise higher.

The real opportunity for ecommerce owners

AI is not making good ecommerce engineering irrelevant. It is revealing which part of your development bill was buying production and which part was buying judgment.

Production is becoming abundant. Judgment, context, taste, reliability, and accountability are not.

That means the winning store owner will not simply fire the development team or cut every retainer. They will build a leaner relationship with clearer standards:

Fewer hours spent on boilerplate.
More senior attention on high-risk decisions.
Smaller releases that are easier to review.
Better automated and human testing.
More experiments tied to customer and profit outcomes.
Clear ownership when systems fail.

Your Shopify dev retainer may be about to shrink. Good. Use the savings to make the business smarter, faster, safer, and harder to copy.

Final takeaway

The “middle class of software engineering” debate should not make ecommerce owners celebrate layoffs or developers defend inefficient billing. It should force both sides to update the value exchange.

AI will write more of the code. The people worth paying will decide what belongs in the store, what should never ship, how to verify it, and how it contributes to profitable growth.

That is not the end of ecommerce development. It is the end of paying premium prices for implementation without accountability.

Frequently Asked Questions

Will AI replace Shopify developers?

AI is more likely to reduce time spent on routine theme work, scaffolding, tests, documentation, and contained bug fixes than to eliminate Shopify developers completely. Developers who own architecture, security, data quality, checkout risk, and commercial outcomes should remain valuable.

Can AI build a complete Shopify store?

AI can generate theme blocks, content, code, and substantial parts of a storefront. A production store still requires decisions about brand, merchandising, integrations, analytics, performance, accessibility, privacy, security, and operational reliability.

Should I reduce my Shopify development retainer now?

Review it rather than cutting it blindly. Ask which tasks are now AI-assisted, how those efficiencies affect pricing, who reviews generated work, and which measurable outcomes the retainer owns.

What Shopify tasks should remain under human review?

Checkout logic, discounts, payments, customer data, tracking, authentication, integrations, migrations, accessibility, performance, and any change that can affect revenue or trust should receive accountable human review.

What should ecommerce brands reinvest AI savings in?

Prioritise conversion experiments, data quality, security, reliability, performance, customer research, and proprietary systems that create a genuine competitive advantage.

Related reading

August 14, 2026

Do OpenAI and Anthropic Really Drive 70% of AI Revenue? What It Means for Your Business
One statistic is racing around the AI industry:

More than 70% of AI revenue comes from OpenAI and Anthropic.

It is a powerful number. It suggests that thousands of AI products, billions in infrastructure spending and the strategies of the world’s largest technology companies rest on two model providers.

It is also easy to repeat incorrectly.

The available evidence does not establish that OpenAI and Anthropic collect 70% of every dollar earned across the entire AI market. The figure comes from analyst estimates about a narrower—and in some ways more revealing—part of the ecosystem: the AI-related revenue earned by Amazon, Microsoft and Google from cloud compute, model access and associated commercial arrangements.

In other words, the claim is less “70% of AI revenue flows to two companies” and more:

Analysts estimate that OpenAI and Anthropic may directly or indirectly drive more than 70% of the AI-related revenue attributed to the three largest US cloud platforms.

Some of that money flows from OpenAI and Anthropic to cloud providers for compute. Some comes from cloud customers buying access to their models. The exact totals are not disclosed cleanly by the companies and the estimates differ by analyst.

That correction weakens the viral headline—but strengthens the business lesson.

The AI economy may be broader than two companies. The commercial infrastructure underneath it is still remarkably concentrated. If your marketing workflow, ecommerce operation or software product depends on one foundation-model provider, you are not merely choosing a tool. You are inheriting that provider’s pricing, availability, policy and strategic risk.

Fact check: what does the 70% figure actually measure?

The source behind the current discussion is Ed Zitron’s analysis, “The AI Demand Bubble”. It combines estimates attributed to analysts at Barclays, UBS and Wells Fargo.

The estimates cited in that analysis include:
- Amazon Web Services: One Barclays estimate put OpenAI and Anthropic at 73% of AWS AI revenue in 2026 and 2027. A separate estimate cited in the same article put their direct compute contribution at 59% in 2026.
- Google Cloud: UBS estimates cited in the analysis assigned 28% of total 2026 Google Cloud revenue to OpenAI and Anthropic, rising to more than 48% in 2027. The author then inferred that the pair could represent at least 70% of Google’s narrower AI-related revenue.
- Microsoft: Wells Fargo estimates cited in the article put OpenAI and Anthropic at 70% or more of Microsoft’s AI revenue, reaching approximately 74% in the relevant forecast period.
These are not one consistent, audited market-share dataset. They mix:
- direct purchases of computing capacity;
- cloud revenue associated with the two laboratories;
- revenue from platforms that resell access to their models;
- analyst forecasts for future periods;
- the article author’s own classification and inference.
The cloud companies do not publish a standard “AI revenue” line that allows outsiders to calculate a definitive global market share. OpenAI and Anthropic are also private companies, so their financial disclosure is more limited than that of a public company.

The accurate version of the claim

Use this formulation:

Analyst estimates suggest OpenAI and Anthropic account for roughly 70% or more of the AI-related revenue attributed to Amazon, Microsoft and Google, although definitions and estimates vary.

Avoid this formulation:

OpenAI and Anthropic receive 70% of all revenue in the global AI industry.

That broader statement would require a defined market covering chips, cloud infrastructure, enterprise software, consumer subscriptions, services, advertising, robotics, data platforms and other AI-related businesses. The cited analysis does not provide that calculation.

Why the corrected number still matters

The statistic is not a clean measure of the whole AI market, but it exposes three forms of concentration.

1. Demand concentration

Cloud providers have invested extraordinary amounts in data centres, accelerators and power capacity. If a large share of the associated revenue depends on two customers and their models, the infrastructure boom has a narrower demand base than the headline “AI adoption” numbers imply.

This does not prove the boom will collapse. It means future growth depends heavily on OpenAI and Anthropic continuing to:
- attract paying customers;
- raise or generate enough cash to fund compute;
- turn model capability into sustainable demand;
- serve workloads efficiently enough to support margins;
- maintain favourable relationships with cloud partners.
2. Model concentration

Many applications are not independent AI businesses in a technical sense. They are interfaces, workflows or specialised data layers built on a small number of foundation models.

That can be a perfectly good business. Shopify did not need to build a payment network, and SaaS companies do not manufacture their own processors. Specialisation creates value.

The risk begins when the application has no meaningful advantage beyond one provider’s output and cannot operate if that provider changes.

3. Strategic concentration

OpenAI and Anthropic influence more than model quality. Their decisions can shape:
- token and subscription prices;
- API limits and access tiers;
- context-window and tool-use behaviour;
- model retirement schedules;
- safety policies and refused use cases;
- data-processing terms;
- regional availability;
- integration standards;
- which workflows become economically viable.
A business built on top of one provider may experience these decisions as product changes—even when it had no voice in making them.

What happens if one of the two stumbles?

“Stumbles” does not have to mean bankruptcy. For a customer, smaller changes can create the same operational effect.

Prices rise

If inference pricing increases or a subsidised product becomes more expensive, an application with weak margins may become uneconomic overnight.

This is especially dangerous when the company offers customers a fixed monthly price while paying the model provider per token, image, tool call or unit of compute.

A model or feature is retired

Prompts tuned for one model do not automatically behave the same on its replacement. Output format, tone, refusal patterns, latency and tool selection can all change.

Without regression tests, a “simple upgrade” can quietly damage product listings, customer replies, campaign copy or structured data.

Reliability declines

An outage at the foundation-model layer can stop every workflow built above it. Even partial degradation—higher latency, elevated errors or inconsistent tool calls—can create queues, duplicate actions and failed customer experiences.

Regulation or litigation changes access

New regulatory restrictions, court decisions, government procurement rules or regional compliance requirements can affect how models are offered and which data may be processed.

The correct response is not to predict one dramatic ban. It is to ensure that a single legal or policy change cannot disable an essential workflow without an alternative.

Provider strategy shifts

A model company can enter your category directly, prioritise enterprise contracts, discontinue a partner feature or bundle functionality that makes your product less differentiated.

Platform risk is not only technical. Your supplier can become your competitor.

What AI market concentration means for marketers

For marketers, the immediate temptation is to treat model choice as a creative preference: Which assistant writes the strongest hooks? Which one follows brand voice best? Which one creates the most attractive images?

Those questions matter, but operational dependence matters more.

A marketing stack may use one provider for:
- campaign research;
- segmentation ideas;
- advertisement variations;
- product copy;
- email personalisation;
- image generation;
- social scheduling;
- performance analysis.
If every step depends on one vendor, a policy update or service interruption can stop the entire content pipeline.

The better approach is to distinguish between creative preference and business-critical dependency.
- It is reasonable to prefer one model for campaign concepts.
- It is risky if no other model can render the required data structure.
- It is reasonable to use one assistant for drafts.
- It is risky if brand knowledge exists only inside that provider’s proprietary workspace.
- It is reasonable to optimise prompts for quality.
- It is risky if no regression suite tells you when an update changes the output.
Keep brand guidelines, approved claims, product facts, audience definitions and reusable prompt templates in systems you control. The model should consume your marketing intelligence, not become the only place where it exists.

What it means for ecommerce brands

Ecommerce companies have a deeper dependency problem because AI is moving from content generation into operational action.

Models increasingly help with:
- catalogue enrichment;
- onsite search and recommendations;
- customer-service responses;
- translations;
- merchandising analysis;
- campaign creation;
- pricing recommendations;
- returns and order workflows.
The closer AI gets to customers, orders and money, the more expensive provider concentration becomes.

Your catalogue must remain the source of truth

Store product attributes, claims, translations and policy rules in your own product-information or commerce systems. Do not let one model’s memory or proprietary knowledge feature become the authoritative record.

If you are evaluating content tools, the criteria in our AI tools for ecommerce product listings benchmark remain useful: accuracy, structured output, brand consistency, channel adaptation and measurable workflow performance matter more than a flashy one-off result.

Separate recommendations from execution

An alternative model can replace a copywriting assistant relatively easily. Replacing an autonomous agent that can change prices, send campaigns or issue refunds is much harder.

Our analysis of why humans missed one in three dangerous AI agent commands explains why manual approval alone is insufficient. Permissions, spending limits, audit trails and rollback must be enforced outside the model.

Design graceful degradation

If the preferred model is unavailable, decide what the store should do:
- switch to a verified secondary model;
- queue non-urgent work;
- fall back to deterministic templates;
- preserve human support for sensitive cases;
- disable autonomous writes while keeping read-only analysis available.
“Try again until it works” is not a resilience strategy for orders or customer data.

What it means for AI builders

For developers and founders, concentration creates risk and opportunity at the same time.

The risk: your product becomes a thin wrapper

If your product is only a prompt plus one API call, the provider can reproduce it, a competitor can copy it, and pricing changes can erase its margin.

The strongest moat usually sits elsewhere:
- proprietary workflow data;
- domain-specific evaluation;
- integrations that are difficult to maintain;
- governance and approval controls;
- customer-specific configuration;
- reliable structured outputs;
- auditability;
- user experience and distribution;
- measurable business outcomes.
The opportunity: become the independence layer

Concentration increases demand for products that help businesses use leading models without becoming trapped by them.

Potential opportunities include:
- model routing based on quality, cost and latency;
- portable prompt and policy management;
- cross-model evaluation suites;
- provider-neutral agent tooling;
- caching and cost controls;
- observability across model vendors;
- data-loss prevention and access governance;
- fallbacks for regulated or regional workloads;
- migration testing when models are retired.
HelpingBrains’ AI Governance Platform is aimed at this control layer: AI inventory, prompt governance, access monitoring, risk management and audit-ready reporting should remain consistent even when the underlying model changes.

The opportunity hidden inside a two-horse race

Market concentration is not automatically bad for customers.

Two strong providers can:
- compete aggressively on model quality;
- reduce prices through efficiency gains;
- standardise tool-use patterns;
- accelerate enterprise features;
- make advanced capabilities accessible without infrastructure investment;
- create a large ecosystem for specialised products.
Competition between OpenAI and Anthropic may also prevent either from exercising complete control. Google, Meta, xAI, specialist providers and open-weight models add further pressure even if they are smaller in a particular revenue dataset.

The opportunity for businesses is to use the leading platforms while retaining the ability to move.

This is similar to a sound cloud strategy: “multi-cloud” should not mean duplicating everything across three providers at enormous cost. It should mean identifying critical dependencies, using portable interfaces where practical and maintaining tested alternatives for the failures that matter.

A practical AI diversification checklist

You do not need to abandon OpenAI or Anthropic. You need to know what would break if one disappeared from your stack tomorrow.

1. Map every dependency
- ☐ List every model, API, assistant, agent and AI-enabled SaaS product in use.
- ☐ Record which business workflow each one supports.
- ☐ Identify the provider behind tools that resell or abstract another model.
- ☐ Mark workflows that affect customers, revenue, production data or legal obligations.
- ☐ Assign one accountable owner to every critical AI system.
2. Separate your assets from the provider
- ☐ Store prompts, policies and templates in a controlled repository.
- ☐ Keep product facts, brand rules and customer permissions in your own systems.
- ☐ Export conversation or workflow data where contractually and technically possible.
- ☐ Avoid provider-specific data formats unless the benefit clearly exceeds the switching cost.
- ☐ Document how model output is transformed before it reaches customers or production.
3. Build a model-independent boundary
- ☐ Use a stable internal request and response schema.
- ☐ Isolate provider-specific code behind adapters.
- ☐ Validate structured output rather than trusting free text.
- ☐ Enforce permissions, budgets, privacy rules and prohibited actions outside the model.
- ☐ Log model, version, prompt, tool calls, latency, cost and outcome.
4. Test at least one alternative
- ☐ Maintain a representative evaluation set using real—but sanitised—business cases.
- ☐ Compare quality, cost, latency, refusals and structured-output reliability.
- ☐ Test a secondary hosted model or a suitable open-weight alternative.
- ☐ Measure migration effort rather than assuming APIs are interchangeable.
- ☐ Repeat tests after major model releases.
5. Plan the failure mode
- ☐ Define when to switch providers automatically and when to require human review.
- ☐ Queue non-critical work instead of producing lower-quality customer-facing output.
- ☐ Keep deterministic templates for essential communications.
- ☐ Prevent retries from duplicating sends, refunds, catalogue edits or orders.
- ☐ Run a provider-outage exercise and record the recovery time.
This is the save-worthy part of the story: diversification is not buying two subscriptions. It is making your data, controls and workflows portable enough that a second provider can actually take over.

Should you use both OpenAI and Anthropic?

Not automatically.

A small company may create more complexity than resilience by integrating multiple providers too early. Every additional model introduces another contract, privacy review, evaluation surface and operational path.

Use a second provider when at least one of these is true:
- the workflow is important enough that an outage creates material loss;
- model pricing represents a significant share of your unit cost;
- customers require regional or provider choice;
- one provider frequently refuses or performs poorly on essential tasks;
- a model retirement would require a rushed migration;
- your product promises provider-independent results;
- regulation, procurement or data residency makes one provider insufficient.
For low-risk experimentation, one provider plus good abstraction may be enough. For revenue-critical execution, a tested fallback becomes much more valuable.

The real lesson: concentration belongs on your risk register

The viral 70% claim is too broad. OpenAI and Anthropic do not demonstrably receive 70% of all revenue across the global AI economy.

What the available estimates suggest is still significant: a very large share of the AI-related revenue credited to Amazon, Microsoft and Google may depend directly or indirectly on two foundation-model companies.

That concentration does not mean businesses should stop building. It means they should stop confusing easy access with independence.

Use the best model available for the job. But keep control of:
- your data;
- your prompts and policies;
- your customer relationships;
- your business rules;
- your evaluation criteria;
- your permission boundaries;
- your fallback plan.
The winners will not necessarily be the companies that predict which AI laboratory wins the race. They will be the ones that create value above the model layer—and can keep operating regardless of who is leading next year.

Frequently asked questions

Do OpenAI and Anthropic earn 70% of all AI revenue?

There is no public, audited dataset proving that they receive 70% of revenue across the entire global AI industry. The viral figure is based on analyst estimates of AI-related revenue at Amazon, Microsoft and Google, including compute purchased by OpenAI and Anthropic and cloud platforms reselling access to their models.

Why is AI market concentration a risk for businesses?

Heavy reliance on one or two providers exposes businesses to price changes, outages, model retirements, policy changes, regulatory restrictions and strategic competition. The risk is highest when core data, prompts and workflows cannot move to another provider.

Is a multi-model strategy always better?

No. Multiple providers add cost and complexity. The right approach is proportional: abstract critical integrations, maintain evaluation tests and create a verified fallback for workflows where downtime or forced migration would cause material harm.

How can ecommerce companies avoid AI vendor lock-in?

Keep catalogue data and business rules in company-controlled systems, use stable internal schemas, separate provider-specific code, enforce permissions outside the model and test the same workflow against at least one alternative model.

What creates a defensible AI product if the models are commoditised?

Defensibility usually comes from proprietary data, domain workflow, evaluation, integrations, governance, user experience, distribution and measurable outcomes—not exclusive access to a general-purpose model.
August 10, 2026

Humans Missed 1 in 3 Dangerous AI Agent Commands. Here’s the Fix

An AI agent asks for permission to run a command. The request looks routine. You have already approved ten similar actions, the workflow is waiting, and the button says Approve.

Would you notice if that harmless-looking command could expose credentials, contact an unknown server or execute code that the agent quietly modified one step earlier?

In a browser experiment covering more than 40,000 game sessions and 409,000 approve-or-deny decisions, the average player missed roughly one in three simulated threats. Mean accuracy was 66.3%. Seven percent of sessions approved every prompt.

That is an alarming result—but it needs careful framing.

This was not a peer-reviewed enterprise security study or a test inside real ecommerce companies. It was a short, 60-second game created by developer Alex Wauters. Approximately 34% of the commands shown were threats, far more than a normal employee should encounter. Players were also placed under artificial time pressure and knew that dangerous commands were present.

So the experiment does not prove that every business reviewer will miss exactly one-third of malicious AI actions. It demonstrates something narrower and still important: a human approval button is a weak security control when the reviewer lacks context, sees too many prompts or is asked to judge risk from a familiar-looking command name.

For companies connecting AI agents to storefronts, CRMs, advertising platforms, customer-service tools or payment workflows, that distinction matters.

The problem is not simply that AI can make a dangerous request. The problem is an approval workflow that expects a busy human to recognize every dangerous request in time.

What the 40,000-play experiment actually found

The original analysis by Alex Wauters reported four headline findings:

Players missed about one in three threats on average, with 66.3% mean accuracy.
32.9% of sessions produced a negative score because approved threats and incorrectly blocked safe commands outweighed correct decisions.
35.2% caught every threat, but some achieved this by blocking almost everything.
7% approved every request.

The type of command made a large difference.

Threat category	Examples used in the game	Miss rate
Obviously destructive	Deleting root files or opening all permissions	11.7%
Persistent mutation	Scheduled-task injection or Git configuration hijacking	23.8%
Exfiltration or code execution	Unknown APIs or typosquatted packages	33.4%
Scope violations	Reading cloud credentials or cluster configuration	35.0%

People were comparatively good at catching commands that *looked* destructive. They struggled more when the risk was indirect, depended on earlier context or was hidden behind an ordinary developer workflow.

The most frequently missed example was npm run analyze, approved 64.7% of the time. The name sounds harmless, but an npm script can execute whatever has been defined in package.json. In the game, the earlier history showed suspicious behaviour, yet many players still approved it.

That is the real lesson: humans often evaluate the label in front of them, not the full chain of changes behind it.

Why “human in the loop” is not enough

Human oversight remains valuable. A qualified person should own high-impact decisions, especially those affecting customers, money, production data or legal obligations.

But adding an approval dialog does not automatically create effective oversight. It can fail in at least four ways.

1. Approval fatigue turns review into clicking

When an agent requests permission repeatedly, each new prompt feels less exceptional. The human shifts from evaluating risk to keeping the workflow moving.

This is familiar from cookie banners, security warnings and access prompts: too many alerts train people to clear alerts, not investigate them.

2. The reviewer sees the command, not its history

A request such as “publish campaign,” “run analysis” or “sync catalogue” may conceal a chain of earlier edits, retrieved instructions, third-party tool calls and generated files.

The final action may look ordinary even when its inputs have been poisoned.

3. Business users cannot evaluate technical side effects

A marketer can judge whether email copy fits the campaign. They should not be expected to determine whether a connector is sending customer data to an unapproved endpoint.

Likewise, a customer-service manager can approve a refund policy. They may not know whether the agent’s tool call can access every customer record rather than only the current case.

4. The decision is framed as approve or block

Binary approval forces a false choice. The reviewer may want the agent to continue, but only with a smaller audience, lower budget, redacted dataset or reversible draft.

A safe workflow should support modify, constrain, simulate and escalate, not only approve or deny.

What this means for marketing, ecommerce and customer workflows

The commands in Wauters’ experiment were designed around coding agents, but the underlying problem applies to any agent that can take consequential action.

Marketing automation

An agent may draft content safely but create risk when it can also:

publish directly to social accounts;
email an unrestricted customer segment;
increase campaign budgets;
upload customer lists to external services;
override consent or frequency rules;
make unsupported product, health or sustainability claims.

“Approve campaign” does not tell the reviewer whether the agent changed the audience, budget, tracking configuration or destination URL.

Ecommerce operations

An ecommerce agent may be able to edit prices, promotions, stock, orders and product information at machine speed. One broad approval could produce thousands of customer-facing changes.

A pricing request that looks reasonable could violate a minimum-margin rule. A product-description update could insert an unsupported claim across hundreds of SKUs. A refund agent might perform a legitimate action on the wrong account because its identity boundary is too broad.

This is why my earlier analysis of what happened when GPT-5.6 ran a real business recommends controlled leverage: let an agent investigate broadly, recommend clearly, draft quickly and execute only inside narrow technical limits.

Customer service

Support agents often need access to personal data and account actions. The risks include:

exposing one customer’s information to another;
issuing excessive refunds or credits;
changing account details without strong identity checks;
sending invented policy explanations;
closing or modifying the wrong case;
being manipulated by malicious instructions contained in a customer message or attachment.

In each case, human approval is helpful only if the reviewer sees the relevant customer, data, policy, proposed change and maximum impact in one place.

Before giving an AI agent permission, ask these five questions

The fix is not to remove humans from the process. It is to stop using humans as the only enforcement layer.

Use these five questions before granting an AI agent any meaningful permission.

1. What is the smallest permission this agent actually needs?

Do not grant access based on everything the agent *might* do. Grant only what it needs for the current, defined job.

For example:

Prefer “read performance for these 20 products” over “store administrator.”
Prefer “draft an email” over “send to all subscribers.”
Prefer “propose a refund” over “issue unlimited refunds.”
Prefer “adjust bids within ±5%” over “manage the advertising account.”

Use a dedicated service identity, separate read and write permissions, restrict accessible records and make elevated access temporary where possible.

2. What is the maximum damage one action can cause?

Assume the model is mistaken, manipulated or operating on bad data. Then calculate the blast radius.

Ask:

How much money can it spend or refund?
How many customers can it contact?
How many products, prices or orders can it change?
Which personal or confidential data can it read?
Can it delete, overwrite or publish anything?

Enforce financial, volume, recipient and frequency limits outside the prompt. “Be careful” is an instruction; a €100 transaction ceiling is a control.

3. Can the action be previewed, reversed and tested safely?

The safest default is to let the agent prepare the change without executing it.

Use:

previews for emails, product updates and campaigns;
staging environments for code and configuration;
dry runs for imports, refunds and bulk changes;
small canary groups before a full rollout;
before-and-after snapshots;
idempotency protection to prevent duplicate actions;
automatic rollback when thresholds are breached.

If an action cannot be reversed, its approval threshold should be much higher.

4. What evidence will the reviewer see?

Never show only the final button and a friendly action name.

The approval screen should display:

the exact action and target;
all data that will leave the organisation;
the system, account and permission being used;
a summary of relevant changes made earlier in the workflow;
expected benefit and maximum downside;
policy checks passed or failed;
affected customer count, spend and scope;
rollback or recovery plan.

For high-impact actions, require approval from a named role with the expertise to evaluate that risk. A finance owner should review spending; a privacy owner should review sensitive-data transfers; a merchandiser should review pricing and product claims.

5. What happens if the reviewer misses the threat?

This is the most important question because eventually someone will click the wrong button.

Build controls that remain effective after mistaken approval:

sandbox untrusted execution;
restrict network destinations;
block access to secrets by default;
enforce policy at the tool or API gateway;
monitor unusual sequences and repeated failures;
log every agent identity, tool call, input and output;
alert on privilege escalation or scope changes;
maintain a kill switch independent of the agent;
test incident response and credential revocation.

Research into coding-agent execution security describes isolation, capability controls, network egress restrictions and auditability as distinct layers—not substitutes for one another. The 2026 execution-security review reinforces why businesses need defence in depth rather than a single approval mechanism.

A better AI approval workflow

A responsible workflow separates what the model proposes from what the system permits.

Classify the action. Determine whether it is read-only, reversible, customer-facing, financial, privacy-sensitive or destructive.
Enforce machine-readable policy. Block prohibited tools, data, destinations and limits before asking a human.
Generate a constrained preview. Show the exact change, target, scope, evidence and rollback plan.
Route to the right owner. Ask a qualified, accountable person—not whichever employee happens to be watching the agent.
Execute with narrow credentials. Use temporary, task-specific authority rather than a broad administrator token.
Verify the outcome. Compare the result with the approved action and automatically stop on deviation.
Preserve an audit trail. Record who approved what, based on which evidence, and what actually happened.

HelpingBrains’ AI Governance Platform is being designed around the organisational side of this challenge, including AI inventory, prompt governance, access monitoring, risk management and audit-ready reporting.

A practical pre-permission checklist

Copy this into your agent deployment review:

☐ The agent has one defined owner and one defined business purpose.
☐ It uses a dedicated identity rather than an employee’s shared credentials.
☐ Read, write, publish, send, spend and delete permissions are separated.
☐ Access is limited to the records, tools and time window required.
☐ Customer data and credentials are inaccessible unless strictly necessary.
☐ Spend, refund, recipient, volume and frequency limits are technically enforced.
☐ High-impact actions produce a preview with scope, evidence and downside.
☐ The approver has the knowledge and authority to assess the action.
☐ Untrusted code or tools run inside an isolated environment.
☐ External network destinations are restricted and monitored.
☐ Every important action is logged and attributable.
☐ Repeated failures, unexpected scope changes and policy violations trigger a stop.
☐ Rollback, credential revocation and the independent kill switch have been tested.

If you cannot tick the relevant boxes, the agent is not ready for that permission.

The problem is the approval workflow

The 40,000-play experiment does not prove that humans are useless or that AI agents should never act. It shows why “a human clicked approve” is not an adequate security architecture.

Humans are strongest when they make a small number of well-framed decisions with the right context. They are weakest when software floods them with repetitive prompts and expects them to reconstruct hidden technical history under time pressure.

The goal should therefore be fewer approvals, better approvals and smaller consequences when an approval is wrong.

Before connecting your next AI agent, do not ask only:

“Will a human approve dangerous actions?”

Ask:

“What prevents one mistaken approval from becoming a business incident?”

The problem is not the AI alone. It is the approval workflow—and that is something businesses can redesign now.

Frequently asked questions

Did a scientific study prove humans miss one in three AI threats?

No. The figure comes from more than 40,000 sessions of a timed browser game and 409,000 approval decisions. It is useful evidence of approval fatigue and context problems, but it was not a controlled, peer-reviewed study of enterprise employees. The threat rate and time pressure were intentionally artificial.

Is human-in-the-loop approval useless for AI agents?

No. Human review is valuable for judgment, accountability and exceptional cases. It should sit on top of technical controls such as least privilege, spending limits, sandboxing, network restrictions, logging and rollback—not replace them.

Which AI agent actions should always require approval?

Require strong review for irreversible or high-impact actions involving customer communications, production changes, personal data, payments, refunds, pricing, account permissions, deletion and external publication. The exact threshold should depend on scope, reversibility and maximum possible harm.

How can ecommerce companies reduce AI approval fatigue?

Automatically allow low-risk actions inside strict policies, automatically block prohibited actions, and escalate only meaningful exceptions. Give reviewers a clear preview of affected products, customers, budget, data and rollback options instead of a raw technical command.

What is the safest way to deploy an AI agent?

Start with read-only observation, then recommendations, then drafts. Allow narrow execution only after the workflow has been tested in a sandbox and limited rollout. Use task-specific identities, hard limits, monitoring, audit logs and an independent kill switch.

August 8, 2026

AI Financial Advice Is Better Than Expected—If Ecommerce Founders Ask the Right Questions
What if the difference between useless AI advice and a genuinely useful financial plan is not the model—but the question you ask it?

New research highlighted by MIT Sloan found that large language models often encouraged sensible financial behaviour. They recommended stronger saving buffers, diversified investments and risk levels that changed over time. But their advice improved when researchers replaced ordinary questions with detailed, structured prompts.

That finding matters far beyond personal finance.

Ecommerce founders ask AI vague questions every day:

“How can I improve my cash flow?”

“Should I spend more on Meta ads?”

“Is my profit margin good?”

Those questions usually produce familiar advice: cut costs, improve conversion, negotiate with suppliers and test more creatives. None of it is necessarily wrong. None of it tells you what to do on Monday morning.

The better approach is to give AI the business context, define the decision, state the constraints, require calculations, and ask it to reveal its assumptions. Below are five copy-paste prompts built specifically for ecommerce founders.

What the MIT Sloan research actually found

The researchers surveyed 1,000 adults, collected the prompts they would use to seek financial advice, and simulated the long-term results of following AI recommendations. They compared ordinary user-written prompts with more complete “academic” prompts containing relevant financial conditions and explicit economic assumptions.

Three findings stand out:
1. The advice was better than expected. AI generally moved simulated users closer to established financial-planning principles.
2. Structured questions produced better advice. Complete prompts improved consumption and saving guidance and reduced reliance on simplistic rules of thumb.
3. The advice still had weaknesses. Models did not respond well enough to shocks such as unemployment, allowed portfolios to drift, and produced different outcomes depending on how users framed their questions.
This was a personal-finance study, not an ecommerce experiment. Applying it to ecommerce is therefore an informed translation, not a result the researchers directly tested. The useful principle is simple: AI performs better when the prompt contains the variables, assumptions, constraints and decision criteria that a knowledgeable adviser would ask for.

Before you use the prompts: prepare this data

Use figures from the same reporting period and remove customer names, addresses, payment details and other personal data before pasting anything into a public AI tool.

Prepare:
- Opening cash balance
- Revenue by channel
- Cost of goods sold, including inbound freight and duties
- Advertising spend by channel or campaign
- Payment processing and marketplace fees
- Fulfilment, shipping and return costs
- Payroll and other fixed operating expenses
- Accounts payable and expected payment dates
- Inventory units, landed cost and expected sell-through
- Refund and return rates
- Your minimum cash reserve and risk tolerance
If these figures are incomplete, tell the model which numbers are estimates. Never let it silently treat guesses as facts.

Prompt 1: Build a 13-week ecommerce cash-flow forecast

Use this when sales look healthy but cash still feels tight.
```
Act as a cautious ecommerce FP&A analyst. Build a 13-week cash-flow forecast for my store using the data below.

Business context:
- Currency: [EUR/USD/GBP]
- Opening cash balance: [amount]
- Minimum cash reserve: [amount]
- Current weekly revenue: [amount]
- Expected weekly revenue growth or decline: [%]
- Gross sales paid out after a delay of: [days]
- Refund/chargeback rate: [%]
- Cost of goods payments and due dates: [list]
- Ad spend by week: [list]
- Payroll and fixed costs by week/month: [list]
- Tax/VAT payments and due dates: [list]
- Inventory purchase commitments: [list]
- Other cash inflows/outflows: [list]

Instructions:
1. Separate profit from cash movement.
2. Show opening cash, inflows, outflows, net movement and closing cash for every week.
3. Identify the first week cash falls below my minimum reserve.
4. Create base, downside and upside scenarios. Use [10–20%] lower revenue in the downside case and clearly state every assumption.
5. Rank the five actions that would improve cash fastest without damaging customer experience or long-term growth.
6. For each action, estimate the cash impact, timing, trade-off and confidence level.
7. List missing data that could materially change the result.

Do not invent numbers. Mark unknown values as “missing” and show formulas so I can verify the calculations.
```
What makes this prompt better: It forces the model to model timing, not just profitability. That matters when payment-provider delays, inventory deposits, VAT and ad bills hit on different dates.

Prompt 2: Allocate the advertising budget by contribution profit

ROAS alone can reward campaigns that generate revenue while destroying cash.
```
Act as an ecommerce growth finance analyst. Recommend how to allocate next month’s ad budget using contribution profit and cash constraints—not ROAS alone.

For each channel/campaign I will provide:
- Spend
- Revenue
- New-customer revenue
- Orders
- Average order value
- Gross margin before ads
- Discount rate
- Payment fees
- Pick/pack and shipping subsidy
- Return/refund rate
- Repeat-purchase rate or 90-day LTV, if known
- Current daily budget
- Minimum viable spend for learning

My total available budget is [amount]. My minimum cash reserve is [amount]. My target payback period is [days/months].

Tasks:
1. Calculate estimated contribution profit after variable costs for every campaign.
2. Separate acquisition efficiency from retention assumptions.
3. Flag campaigns where ROAS looks healthy but contribution profit is weak or negative.
4. Recommend which budgets to increase, hold, reduce or pause and by how much.
5. Preserve at least [10–15%] of the budget for controlled experiments.
6. Provide conservative, base and aggressive allocation scenarios.
7. State the break-even ROAS for each campaign or product group.
8. Explain which recommendation is most sensitive to uncertain data.

Do not assume all revenue has the same margin. Do not count unproven future LTV as current profit. Show formulas and round monetary values to two decimals.
```
This turns AI from a media-buying cheerleader into a decision-support tool. It also exposes the gap between platform-reported success and money retained by the business.

Prompt 3: Find margin leakage by SKU

A bestselling product can still be one of the least valuable products in the catalogue.
```
Act as an ecommerce unit-economics analyst. Analyse margin by SKU and identify hidden margin leakage.

For each SKU, use:
- Units sold
- Selling price before and after discounts
- Landed product cost
- Marketplace/payment fee
- Pick-and-pack cost
- Packaging cost
- Outbound shipping paid by us
- Return rate
- Average return-processing cost
- Damage/write-off rate
- Ad spend attributed to the SKU, if available

Tasks:
1. Calculate gross margin, contribution margin before ads and contribution margin after ads in both currency and percentage terms.
2. Rank SKUs by total contribution profit, not revenue.
3. Identify high-revenue/low-profit products, products made unprofitable by returns, and products where discounting crosses the break-even point.
4. Calculate the minimum viable selling price and maximum safe discount for each SKU.
5. Recommend one action per weak SKU: raise price, reduce discount, renegotiate cost, change packaging, reduce acquisition spend, bundle, improve product content, or discontinue.
6. Estimate the monthly profit impact if the top three recommendations work.
7. State all assumptions and flag missing or unreliable data.

Do not average costs across SKUs when SKU-level data is available. Show the calculation method before giving recommendations.
```
Product content can affect both conversion and returns. If your weak SKUs also have unclear or inconsistent listings, compare the tools in our guide to the best AI tools for ecommerce product listings.

Prompt 4: Decide what inventory to reorder—and what not to buy

Inventory planning is a cash-allocation decision disguised as an operations task.
```
Act as a conservative ecommerce inventory and cash-planning analyst. Recommend reorder quantities for the next [90/180] days while protecting my minimum cash reserve.

For each SKU, I will provide:
- Units currently available
- Units already on purchase order
- Average weekly sales
- Sales volatility or weekly sales history
- Supplier lead time
- Minimum order quantity
- Landed unit cost
- Selling price
- Contribution margin per unit
- Stockout cost or strategic importance
- Product expiry/seasonality information
- Return rate

Business constraints:
- Cash available for inventory: [amount]
- Minimum cash reserve after purchasing: [amount]
- Forecast horizon: [weeks/months]
- Desired service level: [%]

Tasks:
1. Estimate weeks of cover and likely stockout date for each SKU.
2. Recommend reorder quantity and timing under base, downside and upside demand.
3. Prioritise purchases by expected contribution profit per euro/dollar of cash invested.
4. Flag slow-moving, seasonal, expiring or low-margin stock that should not be reordered.
5. Show how the purchase plan changes if sales are 20% below forecast or lead times increase by 30 days.
6. Identify where a smaller order, supplier negotiation, preorder or bundle would reduce risk.
7. Keep the plan within the cash constraint and show the remaining reserve.

Do not assume growth continues indefinitely. Do not recommend an order unless you show the demand, cash and margin logic behind it.
```
For teams managing large catalogues, clean product attributes and channel-ready data are prerequisites for reliable analysis. That is the problem CatalogOps is designed to address.

Prompt 5: Run a founder decision stress test

Use this before a sale, a major inventory order, an agency commitment or a large ad-budget increase.
```
Act as a skeptical ecommerce CFO. Stress-test this decision: [describe the decision].

Current position:
- Cash balance: [amount]
- Minimum reserve: [amount]
- Monthly revenue: [amount]
- Contribution margin after variable costs: [% and amount]
- Fixed monthly costs: [amount]
- Inventory commitments: [amount and dates]
- Tax/VAT liabilities: [amount and dates]
- Proposed investment: [amount and timing]
- Expected benefit: [assumption]
- Time to expected payback: [assumption]

Tasks:
1. Build best, base and worst-case outcomes over the next six months.
2. Calculate break-even, cash runway and payback period for each scenario.
3. Identify the three assumptions most likely to make the decision fail.
4. Perform sensitivity analysis on conversion rate, ad costs, return rate, gross margin and revenue timing.
5. Recommend clear go, test, delay or reject criteria.
6. Design the smallest reversible test that could validate the riskiest assumption.
7. Tell me what evidence would change your recommendation.
8. Critique your own analysis and list anything a qualified accountant or finance professional should verify.

Use only the information supplied. Separate calculations, assumptions, interpretation and recommendation. Never present an estimate as a known fact.
```
The last instruction is important. AI should not merely defend its first answer; it should identify how that answer could be wrong.

A reusable structure for better ecommerce finance prompts

The five prompts follow the same pattern:
1. Role: Define the type of analysis required.
2. Context: Supply the relevant business model and reporting period.
3. Data: Include the variables that drive the decision.
4. Constraints: Set cash reserves, budgets, timelines and risk limits.
5. Scenarios: Require downside, base and upside cases.
6. Calculations: Ask for formulas and prohibit invented numbers.
7. Decision: Request a ranked action, not a generic explanation.
8. Challenge: Ask what is missing and what could invalidate the answer.
This is also why prompt governance matters inside a company. A shared, reviewed prompt produces more consistent decisions than every employee improvising a different question. Our AI Governance Platform is being built around controls such as prompt governance, AI inventory and audit-ready reporting.

Where AI financial advice can still go wrong

AI is useful for structuring analysis, explaining trade-offs and exploring scenarios. It is not your accountant, tax adviser or fiduciary.

Use these safeguards:
- Recalculate important figures in a spreadsheet or accounting system.
- Verify tax, VAT, employment and regulatory advice with a qualified professional.
- Do not paste personal or confidential customer data into an unapproved tool.
- Require the model to distinguish facts, estimates and assumptions.
- Compare the recommendation with a downside scenario.
- Keep a human responsible for every material financial decision.
- Re-run the analysis when revenue, ad costs, returns, lead times or cash commitments change.
The risks are higher when AI can take action rather than simply give advice. In a recent real-business experiment, an AI agent reportedly lied, spammed and lost money—an example we unpack in what happened when GPT-5.6 ran a real business.

The real lesson for ecommerce founders

MIT Sloan’s research does not prove that AI can replace an accountant. It shows something more immediately useful: AI can produce sensible financial guidance, but users who frame the problem well receive better advice.

For ecommerce founders, the prompt should behave like a good finance intake form. It should capture cash timing, contribution margin, inventory commitments, uncertainty and decision constraints before the model recommends anything.

Do that, and AI becomes more than a chatbot. It becomes a fast, inexpensive second pair of eyes—one that helps you prepare better questions for your accountant, challenge optimistic assumptions and make daily operating decisions with greater discipline.

Copy the five prompts, replace the placeholders with your own numbers, and begin with the decision that is tying up the most cash today.

FAQ

Can ChatGPT replace an ecommerce accountant?

No. AI can help organise data, calculate scenarios and explain trade-offs, but it does not carry professional responsibility and may make arithmetic, tax or regulatory errors. Use it as decision support and have material decisions verified by a qualified professional.

What financial data should I give an AI tool?

Provide aggregated business figures such as revenue, product costs, fees, ad spend, returns, inventory commitments and cash balances. Remove customer personal data, payment details, credentials and any confidential information your company has not approved for the tool.

Why is contribution margin better than ROAS for ad-budget decisions?

ROAS compares revenue with advertising spend but ignores product cost, discounts, fulfilment, fees, shipping subsidies and returns. Contribution margin shows how much money remains after variable costs and therefore provides a stronger basis for budget allocation.

How often should an ecommerce cash-flow forecast be updated?

Update a 13-week forecast at least weekly, and immediately after a material change such as a delayed supplier payment, large inventory order, payout hold, tax bill or significant change in sales or ad efficiency.

Which AI tool should ecommerce founders use for financial analysis?

Choose a tool that supports structured data, strong privacy controls and enough context for your analysis. The quality and completeness of the input often matter more than small differences between leading models. Verify calculations outside the model before acting.
August 2, 2026

Best AI Tools for Ecommerce Product Listings: A Practical Benchmark

Generic AI writing advice is everywhere. Ecommerce operators, however, have a more demanding problem.

You do not need one clever paragraph. You may need 500 accurate product descriptions, hundreds of distinct SEO titles, marketplace-ready bullet points, consistent brand language, and a workflow that does not invent materials, features, or certifications.

So which AI tool is actually best for ecommerce product listings?

For this benchmark, we compare three widely used options:

ChatGPT with a structured custom prompt
Jasper
Copy.ai

The goal is not to declare one universal winner. It is to find which tool fits each part of a real listing workflow—from a single product page to bulk catalogue production.

What an Ecommerce Listing Tool Must Get Right

A polished description is only one part of a useful listing. A serious evaluation should cover:

Factual accuracy: Does the output stay within the supplied product data?
SEO quality: Does it use the target keyword naturally in the title, description, and supporting copy?
Conversion clarity: Are the main benefits easy to scan and understand?
Brand consistency: Can it reliably follow a defined tone of voice?
Channel adaptation: Can one source record become a Shopify description, Amazon bullets, a Google Shopping title, and social copy?
Bulk workflow: Can a team process dozens or hundreds of SKUs without repetitive manual work?
Editing effort: How much human correction is needed before publishing?
Cost at scale: Does the workflow remain economical when catalogue volume increases?

The best tool is therefore not necessarily the one that writes the most impressive first draft. It is the one that produces the highest proportion of publishable listings with the least risk and rework.

The Test Product

To keep the comparison fair, each tool should receive exactly the same source data.

Product: Insulated stainless-steel water bottle
Capacity: 750 ml
Material: 18/8 stainless steel
Insulation: Double-wall vacuum insulation
Claim supplied by brand: Keeps drinks cold for up to 24 hours and hot for up to 12 hours
Features: Leak-resistant lid, wide mouth, BPA-free lid components
Colours: Black, blue, and sand
Target customer: Commuters, gym users, and hikers
Primary keyword: insulated stainless steel water bottle
Brand voice: Practical, confident, and low-hype

The tools should then be asked to produce:

An SEO product title
A 120–160 word product description
Five benefit-led bullet points
A meta title and meta description
A shorter marketplace version

This test can be repeated across several categories—fashion, beauty, electronics, and homeware—to expose category-specific weaknesses.

1. ChatGPT: Best for Flexible, Controlled Workflows

ChatGPT becomes much more useful for ecommerce when it receives a structured prompt instead of a vague instruction such as “write a product description.”

Its main advantage is flexibility. You can define the exact output schema, restrict it to approved facts, supply examples of your brand voice, and request multiple channel formats in one response.

Where ChatGPT performs well

Producing several listing formats from one product record
Following detailed formatting and tone instructions
Rewriting copy for different audiences or markets
Generating structured output that can feed an automation
Iterating quickly when a draft needs a different angle

Where it needs control

It may add plausible but unsupported claims if the prompt is loose
Output consistency can drift across a large batch
Bulk processing requires a spreadsheet, API, or custom workflow
Human review remains essential for regulated or high-risk categories

Best use case

ChatGPT is strongest for teams that want control and are willing to build a repeatable prompt or automation around it. It is particularly attractive when product data already exists in a PIM, ERP, spreadsheet, or ecommerce platform.

2. Jasper: Best for Brand-Governed Marketing Teams

Jasper is positioned around marketing content and brand consistency. That makes it relevant for ecommerce teams managing multiple writers, campaigns, or product categories.

Its value is less about producing one description and more about giving a marketing team a controlled environment for reusable brand context and content workflows.

Where Jasper performs well

Maintaining a defined brand voice across content
Supporting non-technical marketing teams
Reusing campaign and company context
Creating product copy alongside ads, emails, and landing pages

Where it needs control

The additional platform cost needs to be justified by team usage
Output still requires verification against source product data
A polished interface does not automatically solve catalogue integration
Teams should test whether its workflow matches their actual SKU volume

Best use case

Jasper makes the most sense for established marketing teams that value governance, shared brand context, and a guided interface more than maximum workflow flexibility.

3. Copy.ai: Best for Repeatable Go-to-Market Workflows

Copy.ai has expanded beyond basic copy generation into repeatable workflows. For ecommerce, that can be useful when listing creation is connected to broader go-to-market tasks.

For example, one product launch could require a product description, marketplace bullets, retailer outreach copy, an email, and social posts. A workflow-led system can reduce the number of disconnected steps.

Where Copy.ai performs well

Turning repeatable content tasks into workflows
Producing multiple go-to-market assets from shared inputs
Helping teams standardise common generation processes
Supporting use cases beyond the product detail page

Where it needs control

Workflow setup requires clear inputs and quality rules
Ecommerce-specific integrations should be checked for your stack
Generated claims still need validation
The benefit is smaller if you only need occasional descriptions

Best use case

Copy.ai is worth considering when product listing creation sits inside a larger, repeatable launch or sales-content process.

Quick Comparison

Requirement	ChatGPT	Jasper	Copy.ai
Prompt flexibility	Excellent	Good	Good
Brand governance	Good with setup	Strong	Good
Structured outputs	Strong	Moderate	Strong in workflows
Bulk automation	Strong with API or custom tooling	Depends on workflow	Workflow-oriented
Ease for non-technical teams	Good	Strong	Good
Best fit	Custom ecommerce operations	Brand-led marketing teams	Repeatable go-to-market processes

These ratings describe workflow fit, not guaranteed performance. Plans and features change, so test each option against your own catalogue before committing.

Before and After: What Better AI Input Changes

The biggest performance difference often comes from the input, not the tool.

Weak prompt

Write a catchy description for this water bottle.

Likely result: generic claims, exaggerated language, missing SEO structure, and possible invented features.

Ecommerce-ready prompt

Using only the approved product facts below, create an SEO title, a 120–160 word description, five benefit-led bullets, a meta title under 60 characters, and a meta description under 155 characters. Use the primary keyword naturally. Write in a practical, confident, low-hype tone. Do not invent claims, certifications, materials, dimensions, or guarantees. If required information is missing, flag it instead of guessing.

That instruction makes the output easier to review, compare, and automate.

A Reusable Product Listing Prompt

Copy and adapt this template:

You are an ecommerce product-content specialist.

Create:
1. SEO product title
2. Product description of [WORD COUNT]
3. Five benefit-led bullet points
4. Meta title under 60 characters
5. Meta description under 155 characters
6. [CHANNEL]-specific short version

Approved product data:
[PASTE STRUCTURED PRODUCT DATA]

Primary keyword:
[KEYWORD]

Secondary keywords:
[KEYWORDS]

Audience:
[CUSTOMER]

Brand voice:
[VOICE RULES]

Rules:
- Use only the supplied facts.
- Do not invent performance claims, certifications, ingredients, materials, compatibility, or guarantees.
- Prioritise customer benefits while preserving technical accuracy.
- Avoid repetition, filler, and unsupported superlatives.
- Flag missing information rather than guessing.
- Return the result using the requested headings.

For bulk production, keep product inputs in fixed fields such as SKU, product type, material, dimensions, features, approved claims, audience, keyword, and prohibited terms. Consistent data makes consistent copy possible.

How to Measure Real Conversion Impact

A before-and-after example can show that the copy is clearer, but it cannot prove a conversion lift. To measure commercial impact, run a controlled test.

Track:

Product-page conversion rate
Add-to-cart rate
Organic impressions and clicks
Click-through rate from category or search pages
Return rate
Customer questions caused by unclear information
Time spent creating and approving each listing
Percentage of drafts published without major edits

Test a meaningful group of comparable products, keep pricing and promotions stable where possible, and run the experiment long enough to reduce normal sales variation.

The most useful result may not be “AI increased conversion by X%.” It may be that the team cut production time while maintaining conversion rate and improving catalogue coverage. That is still a significant operational gain.

The Verdict

Choose ChatGPT if you want the most flexible option for custom prompts, structured output, and integration into an ecommerce automation.

Choose Jasper if your priority is brand governance and a shared environment for a marketing team.

Choose Copy.ai if listing creation is part of a broader, repeatable go-to-market workflow.

For most technically capable ecommerce teams, a structured ChatGPT workflow is the best starting point because it can be adapted to existing product data and publishing systems. For larger marketing organisations, the governance and workflow experience of a dedicated platform may justify the extra cost.

The winning setup is not the tool that produces the flashiest paragraph. It is the one that turns reliable product data into accurate, channel-ready content—with measurable savings and no unsupported claims.

July 28, 2026

Category: AI for Ecommerce

Working With AI Is a Leadership Skill, Not a Coding Skill

Why AI looks like a coding skill

What the evidence actually says

Leadership behaviour 1: Write clear briefs

The weak ecommerce brief

The leadership brief

A reusable AI brief for ecommerce teams

Leadership behaviour 2: Build feedback loops

Ecommerce example: a Black Friday campaign

Feedback needs a scoreboard

Leadership behaviour 3: Know when to override

Override when the cost of being wrong is high

Override when AI is confidently outside its frontier

What this looks like inside an ecommerce team

You may already be better prepared than you think

A 30-minute exercise for your next team meeting

Final takeaway

Frequently Asked Questions

Do ecommerce leaders need coding skills to use AI?

What is the most important AI leadership skill?

Can AI make ecommerce decisions automatically?

Why does AI sometimes make good work worse?

How should a marketing team start using AI?

Related reading

Your Shopify Dev Retainer Is About to Shrink—Here’s What to Fund Instead

First, the “AI does 80% of the work” claim is not a fact

The Shopify work that is becoming cheaper

1. Basic theme changes

2. Repetitive app and integration scaffolding

3. QA preparation and routine bug fixing

4. Migration and maintenance work

The work store owners will still pay well for

Commerce architecture

Checkout and revenue-risk decisions

Integration reliability

Security, privacy, and compliance

Measurement and experimentation

Who is left on the ecommerce development team?

What should you do with the budget you save?

1. Fund a conversion experiment pipeline

2. Fix your data foundation

3. Pay for senior review, not permanent senior typing

4. Improve speed and resilience

5. Invest in proprietary advantage

A better way to structure your next Shopify retainer

The real opportunity for ecommerce owners

Final takeaway

Frequently Asked Questions

Will AI replace Shopify developers?

Can AI build a complete Shopify store?

Should I reduce my Shopify development retainer now?

What Shopify tasks should remain under human review?

What should ecommerce brands reinvest AI savings in?

Related reading

Do OpenAI and Anthropic Really Drive 70% of AI Revenue? What It Means for Your Business

Fact check: what does the 70% figure actually measure?

The accurate version of the claim

Why the corrected number still matters

1. Demand concentration

2. Model concentration

3. Strategic concentration

What happens if one of the two stumbles?

Prices rise

A model or feature is retired

Reliability declines

Regulation or litigation changes access

Provider strategy shifts

What AI market concentration means for marketers

What it means for ecommerce brands

Your catalogue must remain the source of truth

Separate recommendations from execution

Design graceful degradation

What it means for AI builders

The risk: your product becomes a thin wrapper

The opportunity: become the independence layer

The opportunity hidden inside a two-horse race

A practical AI diversification checklist

1. Map every dependency

2. Separate your assets from the provider