Notes from the Field · November 2026 · 5 min read

Ask the AI, Then the Client: The Five-Question Check

Five questions to ask a small business owner after any AI diagnostic, with what each one catches and what to do next. Save it for your next client session.

An advisor enters a client’s revenue, costs and loan request into a chatbot. The diagnosis comes back in seconds: thin margins, weak marketing, prices to adjust. It reads well. It also misses that sales drop by half every summer, that a family loan is due in March, and that the main supplier stopped giving credit last year.

AI diagnostics are now part of everyday work in Business Development Services (BDS). The tool reads the numbers well. Only the client can explain them. This check is the ten minutes between the AI’s answer and the client’s next decision.

Why the Check Matters

The evidence comes from a five-month randomized trial in Kenya by researchers at UC Berkeley’s Haas School of Business and Harvard Business School: Nicholas Otis, Rowan Clarke, Solène Delecourt, David Holtz and Rembrand Koning. They gave 640 small business owners, running fast-food stalls, poultry farms, cybershops and other small firms, an AI business mentor built on GPT-4 and delivered on WhatsApp. A comparison group received a standard business guide.

On average, the mentor made no difference. Owners who were already doing well improved their results by about 15 percent. Owners who were struggling saw revenue fall by about 8 percent. Both groups asked about as many questions, of similar quality. The struggling owners asked about harder problems, such as competition, drought and raising capital, and were more likely to act on ‘one-size-fits-all’ advice, such as cutting prices or spending more on advertising. The stronger owners kept only the advice that fit their own business.

So what does the advisor add when the machine writes the diagnosis? The judgment to tell which advice applies to this business. That is what I train BDS providers for: not the best template, but the next question.

The Five-Question Check

Ask the client these five questions after every AI diagnostic and before any advice is acted on.

  1. Which of these numbers was unusual this year?
    It catches a lost customer, a one-off order or a season that came early.
    If yes: ask the AI to diagnose a normal year, not this one.
  2. What money moves outside these accounts?
    It catches family loans, savings groups and household spending paid from the business.
    If yes: add these flows before any loan or cash-flow advice.
  3. What do your suppliers and buyers allow you?
    It catches credit terms, minimum orders and how late customers pay.
    If yes: test every pricing or stock recommendation against these terms.
  4. Have you tried this before?
    It catches advice the owner already tested, such as advertising or lowering prices.
    If yes: ask what happened, and drop or adjust the advice.
  5. What would you lose if this advice is wrong?
    It catches risk the business cannot carry.
    If the answer is “a lot”: start with a small test, not the full change.

How to Use It in a Client Session

  1. Run the AI diagnostic as usual and read it with the client.
  2. Ask the five questions and write the answers in the client’s own words.
  3. Give the answers to the AI and ask again. The second answer is built on this business, not an average one.
  4. Note in the client file which advice changed and why.

The form captures the numbers. The client explains them. The interview after the prompt is the new diagnostic.

For Program Managers

Programs that count diagnostics completed will see that number climb fast with AI. Count how often the advice changed after the check, and train advisors on the interview after the prompt, not on the prompt itself. The language side of the same problem is in AI That Does Not Speak the Client’s Language.

Work With Me on This

If your program trains business advisors or brings AI into advisory work, I help teams build checks like this one into their training.

See the Service →

Common Questions

Does AI business advice help small businesses?

The evidence so far is mixed. In a five-month randomized trial with 640 Kenyan entrepreneurs, an AI mentor had no average effect. It improved results for high performers by about 15 percent and reduced revenue for struggling firms by about 8 percent.

How should business advisors use AI diagnostics?

As a first draft. Ask the client five questions about unusual numbers, money outside the accounts, supplier and buyer terms, what they have tried before and what they would lose, then run the AI again with the answers.

Why did struggling firms do worse with AI advice?

In the Kenya trial, they asked about harder problems and were more likely to follow generic advice, such as cutting prices or spending more on advertising, that did not fit their business.

Sources

← All Perspectives

Organizations I Have Worked With
USAIDGlobal Affairs CanadaWorld BankCentral Bank of EgyptCommercial International BankIFCKfWEuropean UnionHabitat for Humanity InternationalAga Khan Agency for MicrofinanceDrosos FoundationUN WomenUN-HabitatLFS AdvisoryGoss GilroyEnvironment and Development GroupAECOMFINBI