I learned this one on the job. A product manager changes domains every couple of years: payments, then support tooling, then banking, each with its own vocabulary and its own unspoken rules. The people who ramped fastest were rarely the ones who read the most. They asked better questions of better material. An AI makes that move cheap enough to run in an afternoon, provided you stop asking it to summarize.
Raw material first
"Research the CRM market" gets you an encyclopedia entry. Eight competitor landing pages, three earnings call transcripts and a forum thread full of complaints get you a model that can think about the market. The ceiling on what you learn is set by what you load, far more than by how you phrase the prompt.
| Source | What it shows you |
|---|---|
| Competitor landing pages | Who each company thinks its customer is |
| Earnings call transcripts | What leadership worries about, between the lines |
| Two- and three-star reviews | The distance between the promise and the product |
| Forum complaints | Pain, in words customers use when nobody is selling to them |
| Job postings | Where a company is spending, and what it cannot do yet |
| Pricing pages | Which customer the business model is built for |
Thirty minutes of gathering beats three hours of analysis on thin inputs. If you can only get three sources, take the complaints, the reviews and one transcript.
The question ladder
Questions come in depths. Most people stop at the second rung. The useful material starts at the third.
| Rung | Kind | Example |
|---|---|---|
| 0 | Factual | Who are the competitors? |
| 1 | Structural | What does every successful player understand that customers never say? |
| 2 | Assumption | Which three beliefs is this whole market built on? |
| 3 | Fragility | What would have to be true for each of those beliefs to be wrong? |
| 4 | Adversarial | Which question would sink this thesis, and where does it still break at its strongest? |
Five questions to ask first
With the material loaded, ask these one at a time, and read each answer before asking the next. The third answer usually changes how you read the first.
- The unspoken insight. What does every successful player here understand that their customers never say out loud?
- The agreement. What does this field need everyone to agree on in order to work, and who gains most from that agreement?
- The money. Follow value from where it is created to where it is captured. Where does it pile up, and where does it leak?
- The real customers. Group the customers by the job they need done, ignoring what they say they want. Which jobs are overserved and which are ignored?
- The clock. What was true here two years ago that is no longer true? What will be true in two years that is not yet?
Rank the assumptions by how easily they break
Every field rests on a handful of beliefs that have to hold for things to stay the way they are. Write three to five of them down, and for each one note what would make it false, what happens if it fails, and whether it feels like granite or sandstone.
The most fragile one is the most important line on the page. If you are already in the market, it is your biggest risk. If you are entering, it is your opening.
Make the AI argue with you
An AI has no thesis to defend, no career riding on the answer and nothing sunk. Use that. Run four rounds:
- The destroyer. "Write five questions a sharp investor, competitor or critic would ask to sink this understanding. Answer each one using only the material I gave you."
- The steel man. For every weak answer: "What is the strongest version of this argument, and where does it still break?"
- The red team. "You are the smartest competitor here, with the same material. What do you do that I am missing?"
- The blind spot. "Which question should I be asking about this field that I have not asked?"
Each weak answer feeds the next round. Stop when answers hold under pressure, or when what is left can only be settled by real-world data. Write those down too; they are your research plan.
The map you keep
field-map.md — copy this
# Field map: [domain] Last updated: [date] ## Brief Need to understand: [what, and why now] Horizon: [a decision this week | a 30-day ramp | a plan] Loaded: [each source and what it covers] Missing: [what would make this sharper] ## Structure [what the five questions turned up] ## Assumptions, most fragile first | belief | what breaks it | if it breaks | granite or sandstone | ## Who matters [players, and how they relate] ## Open questions [ranked; each with how you would find out] ## Watch list [the specific numbers or events that would change this map] ## Session log [date — what you asked, what changed]
Keep the file and come back to it. A map that gets a session a month for half a year is worth more than any single deep dive.
Next door
Cartography builds maps; the Filip Method breaks walls. A map often shows you which wall is worth breaking, and the plan file from that method is where the breaking gets written down. For the moment your search turns something up, read Calibrate before you search.