Insights
What is a discovery phase in a software project, and do you need one?
By Rare Agentics engineering teamLast updated First published 7 minute read
A discovery phase is a short, paid piece of work before a build that turns an uncertain idea into a plan someone can price: who the users are, what they need to do, which systems are involved and what could go wrong. You need one when the problem, the users or the data are genuinely unclear; when you already know what you want built, free scoping is usually enough. Either way, it should end in a written plan, a price and the risks, not just a slide deck.
What is a discovery phase for?
Every software project starts with uncertainty. A discovery phase exists to remove the parts of it that would otherwise be paid for during the build, at a much higher price.
The questions it answers are simple to state and often hard to settle:
- Who will use it, and what each of them needs to get done.
- How the work moves today, including the spreadsheets, inboxes and workarounds nobody mentions in the first meeting.
- Which systems are involved, what data lives in each, and whether they can be connected.
- What must never go wrong: the rules, the regulations and the data that need the most care.
- What the first version should be, and what can wait.
Done well, discovery is the cheapest part of a project and the one that saves the most money. Done badly, it is an expensive way to produce a document nobody reads. For the terms that come up, see our glossary.
What should a discovery phase produce?
The output is the test of a discovery. If you cannot hand it to two suppliers and get two comparable prices, it was not finished.
| Output | What it contains | Why it matters |
|---|---|---|
| Problem statement | The outcome the business needs, in plain language | Stops the build solving the wrong problem |
| Users and journeys | Each kind of user and what they must be able to do | Sets the scope, and therefore the price |
| Current workflow map | How work moves today, and where it breaks | Shows where software helps and where it does not |
| Systems and data | What exists, where it lives, what can be connected | Reveals the integration work early |
| Rules and risks | Permissions, regulation, data sensitivity, failure modes | Decides how much care the build needs |
| First release scope | What is in the first version, and what is not | Keeps the first build small enough to ship |
| Acceptance criteria | How each journey will be checked as done | Gives you a way to hold the build to account |
| Plan, price and risks | The order of work, a fixed price, the open questions | Lets you decide whether to go ahead |
Notice what is not on that list: a large set of polished screen designs. Some design is useful to settle a hard journey, but a discovery that spends most of its budget on pictures has usually skipped the harder questions underneath them.
When do you need a discovery phase, and when is scoping enough?
Not every project needs paid discovery. Many need an honest conversation and a written plan.
Free scoping is usually enough when:
- You know the outcome you want and who it is for.
- The workflow exists today, even if it runs on spreadsheets.
- The systems involved are known and few.
- The first version is clearly smaller than the whole idea.
Scoping a build you have in mind is free with us, and it takes up to a week. Within two working days you get a first view, and within a week a written plan, a fixed price and the risks. You approve the plan and the price before the build starts. We do not sell open-ended discovery: where something genuinely needs proving, we propose a proof sprint or, for agents, an Agent Readiness Review, each at a price agreed in writing first.
Paid discovery is worth it when:
- The problem is clear but the answer is not: several very different products could solve it.
- Several teams do the same work differently, and nobody has written down how.
- The data is scattered across systems that disagree with each other.
- The rules are heavy: health, finance, safeguarding or regulated records (see healthcare for how we handle them).
- You want to use AI agents but do not know where they would be safe.
There is a third option worth knowing about. If one part of the idea is uncertain and the rest is clear, a proof sprint from £3,000 can settle it faster than any document: one bounded feature, workflow or working prototype, live in your accounts in 1 to 2 weeks. Working software answers some questions that no amount of workshops can.
Is an Agent Readiness Review a discovery phase?
Yes, for work where AI agents would act. The Agent Readiness Review is a two-week, fixed-fee review of where AI agents can safely take on work across your business. It covers the same ground as a good discovery, with three questions added that only matter when software is allowed to act:
- What may it do on its own, and what must always wait for a person? Payments, deletions and changes to customer records always need a person.
- What data would leave your systems, to which AI providers, and what do they keep?
- What controls would each task need: which sources it may read, which systems it may act in, and what is recorded.
It ends with a ranked shortlist of agent opportunities, the controls each would need and a costed roadmap you can act on with us, with another partner or on your own. We cover it in detail in what an agent readiness review covers.
What does a bad discovery phase look like?
Discovery is easy to sell and hard to judge, so it is worth knowing the warning signs.
- No fixed end. Discovery that rolls on week after week with no defined outputs is a build on a day rate by another name.
- No price at the end. If the supplier will not price the build after discovery, ask what the discovery was for.
- Workshops without the people who do the work. Leadership knows the goal; the people doing the job know where it breaks. You need both.
- A deck instead of a plan. Personas and journey maps are useful inputs. On their own they cannot be built or priced.
- The output only works with that supplier. The plan, the scope and the acceptance criteria should be yours, readable by anyone you choose to build with.
- No mention of risk. Every project has a riskiest part. A discovery that does not name it has not looked hard enough.
Who should take part in a discovery?
The right people in the room matter more than the method. A discovery that only hears from leadership produces a plan for the business as it is described, not as it runs.
- A decision-maker who can say yes or no to scope, and who will be at the read-out.
- The people who do the work today, ideally two or three of them, because they know the exceptions, the workarounds and the steps that never made it into a process document.
- Whoever owns the systems and data involved, so access, exports and integrations are understood early.
- Anyone who must sign off later: finance for payments, a data protection lead for personal data, a clinical or safeguarding lead where they apply.
From the supplier's side, you want the people who will actually build it, not a separate team who hand over a document and leave. Every question answered in discovery by someone who will not be on the build is a question likely to be asked again.
What should you prepare before a discovery starts?
You can make a discovery shorter and cheaper by bringing a few things with you.
- A sentence on the outcome you need, and the one question the first release must answer.
- Examples of the work: a real spreadsheet, a sample form, an anonymised record, a screenshot of the tool you use now.
- A list of the systems involved and who has access to each.
- Any rules you already know apply: retention periods, consent requirements, approval limits.
- The constraints that matter to you: a launch date, a budget ceiling, a board meeting.
How long should a discovery phase take, and what should it cost?
Long enough to answer the questions, and no longer. For most founders and growing businesses that means days to a couple of weeks, not months. A discovery that needs months is usually trying to design the whole product at once, and the first release would be better off smaller.
On cost, it should be a small fraction of the build it informs, fixed in advance and tied to named outputs. If a discovery costs as much as the first release would, cut the discovery and shrink the release. Ask for the price, the length and the list of deliverables before you agree to it. Our MVP cost guide sets out what drives the price of the build itself.
How do you judge a discovery proposal?
Before you agree to a paid discovery from anyone, check that the proposal says yes to each of these.
- Is there a fixed length and a fixed price?
- Are the outputs named, and do they include a plan, a price and the risks?
- Will it talk to the people who do the work, not only to leadership?
- Will it look at the systems and data, not just the screens?
- Will it name what the first release includes and leaves out?
- Will it set acceptance criteria you can check yourself?
- Can another supplier price the build from what it produces?
- Do you own everything it produces?
- If AI agents are involved, does it cover what they may do, what waits for a person and what data leaves your systems?
If the answer to most of these is yes, the discovery is likely to pay for itself.
How Rare Agentics can help
For a bespoke software build you already have in mind, scoping is free: within two working days you get a first view, and within a week a written plan, a fixed price and the risks. For custom AI agents, the Agent Readiness Review is a two-week, fixed-fee review with a costed roadmap. If one part of the idea needs proving first, a proof sprint starts from £3,000.
Your code and data live in your own accounts, agents cannot reach production, and a person signs off every release. Start a conversation and tell us what you are trying to do.