2. Desk research

§The question this chapter answers: what does the world already know about this problem, and who has already paid to learn it?

§Desk research is the least glamorous step in the sprint and the one with the highest return per hour. It cannot tell you whether your idea is good. It can tell you, with great reliability, which parts of your idea someone has already tested, and what it cost them.

§ 2.1Three things to find#

§The market as it is. Not the market size slide. How people currently handle the problem, what they pay for it, and how often it comes up. Pricing pages tell you what the category believes it is worth. App store and review site comments tell you where the current tools fail, in the customer's words, for free. Job postings tell you which companies are hiring to solve the problem by hand, which is the strongest signal that the problem is real and unsolved.

§Competitors and adjacent tools. Direct competitors are easy to list and rarely the interesting part. The interesting part is the adjacent tool people are bending to the job: the spreadsheet, the group chat, the notes app, the general-purpose model being prompted into a specialized one. Christensen's observation holds: the substitute you should fear is not the one that looks like you, it is the one the customer is already using badly and does not want to leave.

§The ones who failed. Every space has a graveyard. CB Insights has catalogued post-mortems for over a decade; Failory collects them by hand; founders write their own on personal blogs and in farewell emails. Read them for the space you are entering. You are looking for one thing: whether they failed because the problem was not real, or because they were early, or because they executed badly. Only the first should stop you. The other two are information about timing and about what not to repeat.

§ 2.2What desk research is not#

§It is not validation. Finding a large market and no dominant competitor is not a sign of an opportunity; it is more often a sign that the problem does not pay. The 2026 wave of AI companies has made this mistake at scale: the desk research showed an enormous, fragmented market of knowledge workers, and the product research, done later or not at all, showed that most of them were happy with a general chat tool and a copy button.

§It is also not a phase to finish. You will come back to it after every set of interviews, because the interviews will name competitors and workarounds you did not find, and you will need to read about them before the next conversation.

§ 2.3How to hold what you find#

§One page, not a deck. Three short lists: how the problem is handled today and what that costs; who else is in the space and what each is actually for; who failed and why, in one line each. Under the lists, the assumptions from your research plan, each marked with what the desk told you: supported, contradicted, or silent. Silent is the important one. Those are the questions you carry to the experts.

§ 2.4What you leave with#

§A one-page landscape: current handling, adjacent tools, competitors, failures. Your assumptions marked supported, contradicted or silent. And a shorter list than you started with of things you still need a person to tell you.