Product growth diagnosis

Product Growth Diagnosis

What this skill does

It helps someone figure out where to focus their product growth efforts — not by guessing, but by mapping the full opportunity space and then filtering it down to a small, defensible shortlist. The output is a "growth map" (a driver tree) and a ranked list of focus areas, ending with a clear top 3.

The core belief: most teams waste effort on low-leverage work, or on drivers they can't actually move, or on one-off wins that don't compound. A good diagnosis steers them toward areas that are high-leverage, genuinely influenceable, and compounding.

This skill stops at where to focus. Designing the specific experiments/interventions for a chosen focus area is a separate job — hand off cleanly at the end (see "Closing the diagnosis").

The method, end to end

1. Understand the situation (conversational intake).
2. Define the objective — the single success metric to drive. This becomes the root of the tree.
3. Build the driver tree — decompose the objective into its drivers; this is the growth map.
4. Find the leverage — where would improvement have outsized impact?
5. Assess influence — how much can this team actually move each driver?
6. Assess compounding — would success loop and build on itself over time / across metrics?
7. Rank into a shortlist with reasoning, and surface a top 3.
8. Deliver the markdown growth map and close with a handoff.

Work through these conversationally. Don't interrogate the user with a giant form — pull what you need as the conversation flows, and start building the tree as soon as you have enough.

Step 1 — Understand the situation

Gather these naturally over the conversation (not as a checklist read aloud):

• The objective they care most about (and why now).
• Business model — this shapes the whole tree (SaaS, marketplace, e-commerce, subscription, fintech, ad-supported, social/UGC, usage-based, token-based AI). Ask if unclear.
• A quick outline of the product, the company, and the ideal user.
• Stage — pre-PMF, early traction, scaling? (Changes what "leverage" looks like.)
• Current metrics if they have them. If they don't, that's fine — you'll score High/Medium/Low.
• What they've already tried — both wins and dead ends.
• Key pains / opportunities they're already chasing or worried about.

If the user is vague, make reasonable assumptions, state them, and move forward — a diagnosis that keeps moving beats one stuck waiting for perfect inputs.

Step 2 — Define the objective

Pin down one top-level success metric (e.g. MAU, ARR, GMV, paid conversion). If they name several, help them choose the one that matters most right now; the others usually show up as drivers inside the tree anyway. The objective sits at the root: the tree is objective × business model.

Step 3 — Build the driver tree (the growth map)

Read references/driver-trees.md and pull the template matching their business model. Then adapt it to their actual product. A strong tree:

• Is 3–5 layers deep — shallow trees hide the real drivers.
• Has branches that are MECE (mutually exclusive, collectively exhaustive) — or close enough that nothing important is missing.
• Is multiplicative where possible (e.g. Revenue = Traffic × Conversion × AOV × Frequency), so the math of leverage is visible. It's fine to start thematic and harden into formulas over time.
• Nests two shapes: a multiplicative tree at the top, with additive growth-accounting (New + Retained + Resurrected − Churned) at any "active users" or "recurring revenue" leaf. This nesting is the most important structural move — explain it to the user as you do it.

If their objective isn't the template's natural root, re-root the tree onto their objective and pull the relevant branches up.

Where the user lacks hard numbers, annotate each node H/M/L for volume and for conversion/drop-off based on what they do know. Estimates are enough to find leverage.

Render the tree as a Mermaid diagram so it genuinely reads as a map — see "Rendering the growth map" below for the exact pattern to follow.

Rendering the growth map (Mermaid)

Draw the tree as a Mermaid flowchart LR (left-to-right reads best for wide trees; use TD top-down for narrow ones). This keeps the map inside the markdown as editable text while rendering as a real diagram on GitHub, Obsidian, Notion, and most viewers.

Encode the scoring visually so the map alone tells the story:

• Structure — the objective is the root; each layer branches into its drivers. Where a node is a product of its children, say so in the parent label (e.g. Revenue = Accounts × Usage × Price), so the multiplicative logic is visible.
• Leverage & Influence scores on the candidate drivers — on each driver you actually score, append the two lens reads, e.g. Activation: first value<br/>Lev: H · Inf: M. (Optionally add · Cmp: H for compounding if it doesn't crowd the node.) Leave purely structural/parent nodes unscored to avoid clutter. This lets the reader spot the High/High winners — and the high-leverage-but-low-influence traps — directly on the map.
• Where the action is — color nodes: red #ffd6d6 for a clear leak or a low-influence trap, green #d6f5d6 for a top-3 focus area. Leave neutral nodes unstyled. Add a one-line legend.
• Where it helps explain a leverage score, you can still note the volume/conversion read that drove it (e.g. Vol: H · Conv: L) on that node.

Keep it legible: aim for ~12–20 nodes. If the full tree is bigger, show the top 2–3 layers in the diagram and push deeper detail into the shortlist table. Pattern:

flowchart LR
  R["Revenue = Accounts × Usage × Price"] --> ACC["Active accounts"]
  R --> USE["Usage / account"]
  R --> PR["Price per unit<br/>Lev: L · Inf: M"]
  ACC --> NEW["New signups<br/>Lev: M · Inf: H"]
  ACC --> RET["Retained<br/>Lev: H · Inf: H"]:::focus
  ACC --> RES["Resurrected"]
  USE --> ACT["Activation: first value<br/>Lev: H · Inf: M"]:::focus
  USE --> MON["Free→paid<br/>Lev: H · Inf: L (trap)"]:::leak
  classDef leak fill:#ffd6d6,stroke:#cc0000,color:#111;
  classDef focus fill:#d6f5d6,stroke:#2e7d32,color:#111;

Legend: green = top-3 focus area · red = leak or low-influence trap · Lev = leverage, Inf = influence (H/M/L).

Always also include a brief nested-bullet version of the tree right after the diagram, so the map is fully readable even in viewers that don't render Mermaid.

Steps 4–6 — Score every candidate driver on three lenses

For each driver worth considering, score High / Medium / Low on each lens. Explain the reasoning, not just the letter — the reasoning is what makes the diagnosis sharp.

Leverage — if we improve this driver ~10%, does the top metric move a lot? Multiplicative trees make this visible. Hunt for leverage where:

• No one has experimented yet — untested ground often hides the biggest gains.
• High volume meets low conversion / high drop-off — a lot of traffic leaking at a bad step is classic high leverage.

Influence — can this team actually move this driver, or will it barely budge regardless of effort? Judge from intuition, their past experiments, and market/benchmark signals. A high-leverage driver you can't influence is a trap.

Compounding — would success here loop and build on itself over time and across metrics? The best focus areas have a flywheel quality: a win that feeds more growth (e.g. activation that improves retention that improves referral). Prefer compounding drivers over one-off wins. This lens is what separates a great diagnosis from a merely correct one.

Consider all levers, not just product experiments. A driver might be best moved through pricing, marketing, support, partnerships, or sales — not only product/feature changes. Name the most promising lever for each shortlisted driver.

Step 7 — Rank into a shortlist

Rank candidates by the combination of the three lenses, surfacing High/High/High drivers to the top and using the reasoning to break ties. Leverage and influence gate inclusion; compounding breaks ties and elevates. Produce a ranked shortlist, then call out a clear top 3 with a sentence on why each earns its place.

Be willing to deliver an uncomfortable finding: the most common one is that the team is pouring effort into acquisition when the real leak is retention or activation. Say so plainly when the tree shows it.

Step 8 — Deliver the growth map (markdown file)

Always produce a markdown file with this structure:

# Growth Diagnosis: [Product] — [Objective]

## Situation
[2–4 sentences: product, model, stage, current state]

## Growth map (driver tree)
[A Mermaid flowchart of the tree, rooted at the objective, with Leverage/Influence scores on the
candidate drivers and color-coded leaks/focus areas + a one-line legend — followed by a nested-bullet
version for non-Mermaid viewers]

## Top 3 bets
1. [Driver] — why it wins (leverage + influence + compounding, in one or two sentences)
2. ...
3. ...

## Focus-area shortlist (ranked)
[The full ranked list of candidate drivers, each with Leverage / Influence / Compounding scores
and a short reasoning, including the best lever to pull. The top 3 above are drawn from this list;
this section shows the complete picture behind them.]

## What we assumed / what to validate
[Key assumptions made, and the data that would confirm or change the ranking]

Lead the reader with the answer: the top 3 bets come before the full shortlist so the headline recommendation is the first thing they see, with the complete ranking available below for the reasoning. (You still derive the shortlist first internally — this is just the reading order.)

Closing the diagnosis

End by naming the handoff: the diagnosis says where to focus; the next step is designing the specific experiments or interventions for the top bet(s). Offer that as the natural follow-on rather than drifting into experiment design here.

Common pitfalls to steer the user away from

• Optimizing the wrong stage — improving acquisition when retention/activation is the real leak.
• Chasing low-leverage drivers — polishing a step that can't meaningfully move the top metric.
• Targeting drivers they can't influence — high leverage, but immovable in practice.
• Favoring one-off wins over compounding ones — ignoring the flywheel.
• Tunnel vision on product experiments — forgetting pricing, marketing, support, partnerships, sales.
• A shallow tree — 1–2 layers that hide where the real opportunity sits.