Contact Explore Products

Tag: cash flow

  • AI Financial Advice Is Better Than Expected—If Ecommerce Founders Ask the Right Questions

    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.