Is the AI Startup Solving a Budgeted Problem or Selling a Demo?

A polished AI demo can win a room without winning a budget. If you invest in an application startup, ask what work customers already pay to do, who can approve a new purchase, and what happens after the pilot ends. The answers show whether the product belongs in a recurring workflow or only in a presentation.

Start with the last real job

Ask a prospective customer to describe the last time the problem occurred. Who noticed it, what records did they open, how long did it take, and what happened when they got it wrong? Request a recent redacted case. If the customer can't name a recent instance, the problem may be too rare or too minor to sustain a purchase.

Keep the product out of the first part of the interview. “Would you use this AI assistant?” invites a polite prediction. “Show me how your team completed the last contract review” gives you a workflow to test. In Y Combinator's guide to talking with users, Eric Migicovsky recommends asking about a specific past problem and what the person did to solve it. The point is to learn about behaviour before asking for an opinion on a proposed product.

Interview the person who does the work, not only the executive who likes the idea. An operator can show the exceptions, approvals, and records the demo skipped. If several customers independently describe the same expensive step, ask how often it occurs and whether the product removes that step after review and correction.

Find the current spend and workaround

Ask what the organisation uses now. The answer may be an incumbent licence, a contractor, overtime, an internal tool, or a manual process that consumes staff time. Record the cost and the owner of that cost. A team that pays people to reconcile claims every Friday has a different buying signal from a team that says an AI summary “would be nice.”

There may be no named software budget yet. A new product can still win if a buyer owns the outcome and can move funds from an existing expense or approve a new line item. Ask what they would stop paying for, what budget they would use, and what result would justify the transfer. Don't convert every hour saved into cash: some released time becomes more capacity, while the payroll stays the same.

Compare the startup with the customer's real alternatives. A contract-review assistant competes with the legal operations team's current workflow, existing contract software, and perhaps a lower-cost template change. If a simple rule or process fix removes most of the pain, the startup needs another reason to earn recurring spend.

Identify the user, buyer, and approval path

Map the people in a live purchase: the worker who uses the product, the manager who owns the result, the person with budget, and the teams that can block deployment. Ask each what they must see before approval. For an AI tool that reads contracts, legal, security, procurement, and IT may all have a say. The enthusiastic champion may have no authority to sign.

Ask the buyer which cost centre would pay for year two, what amount they can approve, and when their budget resets. Request a recent example of a similar purchase moving through the process. A letter of intent from an innovation team isn't the same evidence as a signed order from the operating budget.

An a16z survey of 100 CIOs in 2025 reported AI spending moving from pilot funds toward recurring IT and business-unit budgets among its respondents. That survey doesn't settle a particular startup's sales case. It makes the source of the customer's recurring budget a concrete diligence question.

Test what the pilot actually proves

Separate a free demo, a paid pilot, and a production contract in the pipeline. For each pilot, ask who chose the use case, what data it used, who evaluated the result, what success threshold the buyer agreed to, and who decides whether to buy afterward. A pilot with no decision date or named budget owner can run for months without testing willingness to pay.

Look at the completed task, not the first model output. In a contract-review example, count documents the team reviewed to its normal standard, changes people made to the AI draft, and missed clauses. Compare time and quality with the old method. Then ask whether the customer agreed to a price before seeing a successful pilot result. A free test can prove product capability while leaving the commercial question open.

Use clear labels in the startup's revenue report. Pilot fees that won't recur belong in a separate column from contracted recurring revenue. Y Combinator's Startup School discussion of SaaS metrics warns against describing one-time payments as recurring revenue.

Count the cost of becoming part of the workflow

A buyer may like the output yet reject the implementation. Ask what systems the product must read and update, whether the customer can grant access, and how staff see the result inside their existing work. Check data rights, security review, procurement, training, and support. Price the work on both sides: the startup's integration and onboarding effort, and the customer's staff time.

Inspect the last three deployments, including a slow or failed one. How long passed between signature and regular use? Which steps needed founder intervention or custom code? If each customer needs a new connector, prompt set, or manual quality team, the apparent software revenue may carry a large service obligation. A narrow but repeatable integration can be more valuable than a broad demo that depends on bespoke setup.

Ask why the customer will renew

Interview customers after the initial excitement has passed. Which staff still use the product each week? Which work would become slower or more costly if they turned it off? What measure appears in the budget owner's review, and did the customer meet it? Ask about teams that stopped using the product as carefully as teams that expanded.

For early companies without renewal dates yet, look for a sequence of stronger commitments: a paid pilot, production deployment, repeat use, wider team access, and a budget owner who has agreed to the next contract. Don't call a pilot renewal. At a later stage, inspect contracted renewals, net changes in spend, and reasons for cancellation by customer cohort. Y Combinator's discussion of retention treats continued use as evidence that customers keep getting value.

Make an investment decision from evidence

Build a short commercial evidence file for each customer: the recent job and present workaround; annual spend or owned budget; user and signatory; pilot terms and result; deployment effort; regular use; and the buyer's stated renewal reason. Mark founder claims separately from invoices, system records, and direct customer interviews. Choose a few reference calls that include a buyer and an operator.

At seed stage, a startup may have little revenue and still have a strong case if customers show urgent work, commit time and money, and can explain a path to repeat purchase. If the only evidence is demo applause and free experiments, write down the next commercial test: a named buyer, agreed price, production task, decision date, and a purchase from the budget that will fund continued use. That test tells you more than another polished prompt.