A free proposal has one job: to win the work. That is not cynicism — it is arithmetic. A vendor writing five speculative proposals a month cannot spend a week analysing each one, so they estimate from pattern-matching and price to be competitive. The number is a sales artefact, and everyone in the room knows it.

A paid discovery has a different job: to be right. That is why it produces a different number, and usually a different project.

What two weeks actually produces

A discovery is not a longer sales call. It has deliverables, and you should refuse to pay for one that does not:

• A written scope: user roles, the actions each can take, and what is explicitly excluded.
• A data model — the entities, their relationships and the ones that will be painful to change later.
• An architecture decision record: the stack, the trade-offs considered, and what it costs to run at three usage levels.
• An integration inventory: every external system, its documentation quality, and the risk each one carries.
• A phased delivery plan with an estimate per phase and the assumptions each estimate rests on.
• A risk register — the five things most likely to blow the timeline, and the mitigation for each.

That package is useful to any competent engineering team. Which is the point.

What it should cost, and the clause that matters

Expect 5–10% of the projected build cost — roughly $8,000–15,000 for a $150,000 project. Two to three weeks.

The clause that matters more than the price: you own the output unconditionally, and you may take it to another vendor. If a firm will only do discovery on the condition that they build it, they are not selling analysis. They are selling a longer sales process, and the conclusion is already known.

Why it pays for itself

Three things happen consistently:

1. Scope shrinks. Roughly a third of what founders describe in a first conversation turns out to be unnecessary for the first release. That reduction alone usually exceeds the cost of the discovery.
2. The estimate becomes defensible. A number built on a data model and a risk register survives a board meeting. A number from a sales call does not survive month three.
3. You learn how the team thinks before you are committed. Two weeks of real collaboration tells you more about a partner than any reference call. And if it goes badly, you have lost two weeks instead of six months.

When to skip it

Honesty requires a limit. Skip discovery when the project is genuinely small — under about $40,000 — or when the scope is fully defined by an existing system you are replicating. In those cases a well-structured proposal is sufficient, and paying for analysis is overhead.

Do not skip it when you are replacing a system a business depends on, when regulated data is involved, or when the budget is the round you just raised.

The most expensive thing you can do with a six-figure budget is start spending it before anyone has written down what "done" means.

Our discovery is fixed-price, delivered in two to three weeks, and yours to keep whether or not we build the thing. That is the only arrangement under which the analysis can be honest.