Skip to content
All articles
Product Strategy9 min read18 November 2026

Is a paid discovery phase worth it before you build?

What a paid discovery phase produces, what it should cost, when to skip it, and how to tell real discovery from a sales upsell. The honest version, from an AI and product studio.

An agency asks you to pay a few thousand dollars before they write a single line of code. Your instinct says upsell. Hold that thought, because in most builds this is the cheapest money you will spend, and skipping it is how five figure projects quietly become six figure ones.

This is the honest version: what a paid discovery phase produces, what it should cost, when you can skip it, and how to tell a real discovery engagement from a sales call with a deliverable stapled to the end. We run ours as a fixed scope engagement called the Discovery Sprint, but the principles below apply whether you buy ours or anyone else's.

What a paid discovery phase actually produces

Discovery is the work that happens before anyone builds: the phase where a fuzzy idea or a real operational problem becomes something an engineering team can price and ship. We call it Phase 0, the phase before production code, and the thing you walk away with is a Build Plan.

A good discovery phase produces, at minimum:

  • A problem and workflow map. Where time, money, and errors actually go today, defined precisely enough to build against.
  • An MVP roadmap. What ships first, what waits, and why. The version that earns the right to the rest, instead of the everything at once build that never launches.
  • Architecture and a stack, in plain language. System design, data flows, integrations, and model choices where AI is involved, written so you can decide from it and any competent team can build from it.
  • A cost to run estimate. What the system costs to operate once it is live, at low, medium, and high volume. For AI products this is the question founders forget to ask until the first bill.
  • Acceptance criteria. How success gets measured, agreed in writing before anything is built.
  • An honest verdict: build, buy, or do not build.
  • A fixed price for the build itself, not an estimate that doubles later.

Our Discovery Sprint packages exactly this, and you keep every deliverable whether or not you build with us.

Why paying for discovery saves money

Software has a brutal cost curve. A decision that costs an afternoon to change on paper costs a week to change in a half built codebase, and a month to change after launch. Discovery front loads the decisions while they are still cheap.

Three places the money comes back:

  • Fewer scope changes. Most mid build chaos is not scope creep, it is scope discovery: the team learning, halfway through, what they should have decided at the start.
  • Accurate estimates. A fixed price written against a real architecture holds. A number quoted off a two paragraph brief is a guess, and guesses are padded or wrong, sometimes both.
  • Less rework. The expensive rewrites happen when an early assumption turns out false after it is already load bearing. Discovery is where you find the false assumption for the price of a conversation instead of a sprint.

What a discovery phase should cost

Paid discovery is usually a fixed fee, not an hourly meter, and the range across the market is wide because scope varies. As a rough map:

  • Lightweight (a well understood problem, few systems): roughly $3,000 to $7,000, one to two weeks.
  • Standard (most new products, mid complexity): roughly $5,000 to $15,000, two to four weeks.
  • Comprehensive or regulated (complex systems, compliance, or an architectural decision that is expensive to reverse): $10,000 and up.

For AI specifically, a working proof on real data tends to run higher, commonly $12,000 to $20,000, because it is engineering, not a document.

Our Discovery Sprint sits at $3,000 to $9,000, fixed at signing. We publish the range because you should not need a sales call to learn the shape of the number. The exact fee is set on a short fit call, before you pay anything, and it does not move mid sprint. What moves it inside the range:

  • Scope surface: how many workflows, user types, and systems the problem touches.
  • Integration depth: the APIs, legacy systems, and data sources we have to design against.
  • A working proof: at the top of the range, we demonstrate the riskiest assumption running on your own data, not on a slide.
  • Technical unknowns: model feasibility, data quality, regulated domains.

One number to anchor against: if discovery costs a few thousand and the build it scopes runs forty or fifty thousand, discovery is a single digit percentage of the project, protecting the other ninety plus percent from being spent on the wrong thing.

When you can skip discovery

Discovery is not a tax on every project. Skip or compress it when:

  • The scope is genuinely defined. You have a real, validated spec and you need hands to build it, not a plan. Talk to a team and get the build priced directly.
  • It is a rebuild of something that already exists. The working product is its own spec. A lighter audit usually beats a full discovery.
  • You are augmenting a team that already owns the architecture. The decisions are made; you need capacity, not direction.
  • The scope is one small, well understood automation. A single workflow with an obvious shape can often be quoted off a call.

Everywhere else the pattern holds: the fuzzier the idea or the higher the build cost, the more discovery earns its fee.

How to tell real discovery from a sales call

Not all discovery is worth paying for. Some of it is a sales call with a template at the end. The tests that separate the two:

  • It can end in do not build. If the only possible outcome is a proposal to build with them, it is a pitch, not an analysis. Real discovery sometimes tells you the thing does not need building, or that an off the shelf tool already does it. That answer is worth more than the fee.
  • The output stands alone. You should be able to hand the plan to any competent team and have them build from it. If it only makes sense as a reason to hire that one firm, it is marketing.
  • It surfaces unknowns instead of hiding them. Good discovery names what it is still unsure about and how it would resolve it. Bad discovery projects false confidence.
  • Senior people do it. Discovery is the highest leverage week of a project. Handed to the most junior person on the team, the leverage is gone.
  • It is paid, and the fee is real work. Free discovery is sales theater: the cost is recovered in the build price, and its incentive is to say yes. Paid discovery by senior engineers is accountable to the deliverable, not to closing you.

A note on the credit trick. Some firms offer discovery free if you build with them, or credited against the build. It sounds generous. It quietly removes discovery's most valuable feature, the ability to honestly recommend against building, because the provider only gets paid if you build. We charge a standalone fee for exactly that reason, and never credit it against a build.

The part most discovery skips: proof

Most discovery ends at paper: diagrams, a scope document, an estimate. Paper is necessary and it is not enough, because the riskiest questions in a build are rarely answered by a diagram. Can the model actually return numbers a business can trust? Will the latency hold on real data? Does the integration behave the way the API docs claim?

The strongest discovery answers at least one of those in working code. At the top of our range, the Discovery Sprint includes a working proof: the single riskiest assumption, built thin and run against your own data, so you see it answer before you commit to the full build. It is the difference between we think this will work and we watched it work.

Discovery as risk reduction

Step back and discovery is really buying down three risks at once:

  • Product risk: is this the right thing to build. The verdict and the MVP roadmap answer it.
  • Technical risk: can it be built, and what will it cost to run. The architecture, the cost to run estimate, and the working proof answer it.
  • Relationship risk: is this the right team to build it. A paid discovery is a low stakes way to see how a team thinks, writes, and handles hard questions before you hand them a large build.

Any one of those going wrong late is more expensive than the whole discovery. Buying all three down for a single digit percentage of the project is not an upsell. It is the cheapest insurance in software.

Both of our longest engagements started this way, a paid discovery that earned the build. See the case studies, or read how the Discovery Sprint works.

FAQ

What does a discovery phase cost?

Most paid discovery runs between $3,000 and $15,000 as a fixed fee, scaled to scope. Our Discovery Sprint is $3,000 to $9,000, fixed at signing, with the exact number set on a free fit call before you pay. An AI working proof on real data sits at the top of that range because it is engineering, not a document.

How long does discovery take?

Usually two to four weeks. A well understood problem with few systems lands near two; one that includes a working proof on your data runs closer to four. Your own time investment is small, a few hours across the engagement, because the team does the work between the calls.

Is a paid discovery phase just an agency upsell?

It can be, which is why the tests above matter. Real discovery can end in do not build, produces a plan any team can execute, and is run by senior people. If the only possible outcome is a proposal to build with that one firm, treat it as a pitch.

Is this the same as a proof of concept?

Not the throwaway kind. A classic proof of concept proves technology in a vacuum and dies in a demo folder. Discovery proves the decision: whether to build, what to build first, and what it will cost to build and run. At the top of our range it includes a working proof on your real data, the useful half of a proof of concept without the throwaway half.

Do I have to build with the firm that runs discovery?

You should not have to. A good discovery deliverable is portable by design: take it to any team, or build in house. Ours is written to be executed by anyone, the fixed build price stays valid for 30 days, and the fee is never credited against the build, so there is no lock in baked into the offer.

Ready to build something?

Book a free 30-minute call, or send us what you're building. No paid gate to start the conversation, and no obligation.