Clients ask, reasonably, why the first two weeks of an engagement don't produce software. Discovery has a branding problem: it sounds like the phase where consultants interview people and return with a deck. Done badly, that is exactly what it is. Done properly, it's the only stretch of the project where changing your mind is still free, and the entire point is to spend that freedom on the decisions that would otherwise get made accidentally, in code, by whoever picks up the first ticket.
The output is a set of closed decisions
Every artefact discovery produces should record a decision that has been made and shut. Not “here are the personas we identified,” but “these are the two user types the first release serves, and this is the one we are deliberately not serving yet.” Not “the system will integrate with the payments provider,” but “here is the specific API we tested against, here is what it returns for a partial refund, and here is what our data model has to look like as a result.”
If a discovery deliverable can be read two ways by two people in the same room, it hasn't decided anything. The argument is still ahead of you — it will simply happen in sprint four instead, when settling it costs a rebuild rather than a conversation.
What should be in your hands at the end
Concretely: a scope that names what's out as clearly as what's in; a walkthrough of the primary flow at enough fidelity that a stakeholder can disagree with it; an architecture sketch showing where data lives, what talks to what, and which third parties you now depend on; a ranked list of the assumptions the plan rests on, ordered by how badly you'd be hurt if each turned out false; and an estimate with its uncertainty attached, so a range still reads as a range when someone quotes it back at you in month three.
That last one carries more weight than it looks. The most useful thing discovery produces is often not the plan but the honest confidence interval around it — knowing which two features are genuinely well understood and which one is a guess wearing the costume of a ticket.
One thing that should not be in the pile is a specification detailed enough to build from without any further conversation. That document takes weeks to write, is wrong in exactly the places that matter most, and quietly transfers responsibility for every gap in it onto whoever reads it next. Discovery's job is to remove ambiguity from the decisions that are expensive to reverse, and to deliberately leave the cheap ones open so the people building can settle them later with better information than anyone has in week two.
The riskiest assumption gets built, not discussed
Where discovery earns its fee is the prototype, and not a polished prototype of the product — a deliberately ugly one of the single riskiest thing. If the business case depends on importing data out of a legacy system nobody has opened in six years, the right use of two weeks is to actually import it, badly, and discover that a fifth of the records have no usable timestamp. That one finding is worth more than every other page of output combined, and it costs three days now instead of three months of denial later.
Which assumption is riskiest is usually obvious the moment someone asks the question out loud, and it is almost never the one people want to prototype. Teams gravitate to the exciting screen. The value sits in the unglamorous thing that would quietly invalidate the schedule.
How to tell it went badly
There are reliable tells. If nothing in the plan changed between the kickoff and the readout, discovery was a transcription exercise — you paid to have your existing assumptions typed up neatly. If every open question is still open, nobody was willing to close anything, and the phase produced a summary rather than a direction. And if the output is entirely documents with no working artefact anywhere, then nobody touched the parts of the problem that resist being described.
The good version is mildly uncomfortable to sit through, because at least one thing everybody was sure about turns out not to be true. Better uncomfortable in week two than committed in month six. That discomfort is the actual deliverable; the document is just where it gets written down.









