Claim
The growth PM's first move is to understand from first principles what is actually happening — not to pull data that justifies a pre-existing plan. Understanding precedes identification of the lever; identification precedes execution. Inverting this sequence ("identify, justify, execute") produces fast motion in the wrong direction.
Mechanism
Most growth roadmaps start from the team's existing model of the user and the funnel. Data is then mined to support that model. The problem is that the model is usually a quarter or two stale, and the supporting data confirms the team's blind spots. Understand-work — qualitative interviews, trace review, talking to non-users, sitting with the actual workflow — surfaces the live picture. Only with the live picture does data analysis identify the right lever. Execution becomes faster because the lever is correctly chosen.
Conditions
Holds when:
- The team has access to real users and time to do qualitative work.
- Leadership accepts a slower start in exchange for faster mid-course execution.
- The product or category is changing fast enough that stale models are common.
Fails when:
- The category is stable and the team's model is still accurate; understand-work returns the same picture.
- Time pressure forces immediate action; understand-work feels like avoidance.
- The team uses "understanding" as cover for indecision.
Evidence
"First you have to really understand from first principles what is actually going on... understand, identify, execute."
Bangaly's complement: impact = environment + skills. Where are you structurally hindered? Where are your skills lacking? What do you control? The answer guides where to invest.
— Bangaly Kaba on Lenny's Podcast, 2026-04-28
Signals
- Roadmap proposals open with a "what is actually happening" section, not a "what we should do" section.
- Pivots happen earlier in the cycle because understanding catches model drift.
- Cross-functional alignment is easier because the team is debating reality, not strategy.
Counter-evidence
At well-known categories with stable user behaviour, the understand step is overhead. Casey Winters and other growth practitioners argue that the better move in mature categories is to copy known-working playbooks from category leaders.
Cross-references
- Force intuitions into explicit predictions so you can find out where you are wrong — what to do with the understanding once you have it
- Evals are systematic data analysis on your LLM application — start with error analysis, not tests — same shape applied to AI applications