Conversion Path Diagnosis

Variables

Replace each placeholder before you run the prompt.

• {{funnel_data}}Stage counts: leads, MQLs, SQLs, opportunities, closed-won. Plus stage-to-stage conversion rates.

  90 days. Leads: 480. MQLs: 220 (46%). SQLs: 60 (27% of MQL). Opportunities: 22 (37% of SQL). Closed-won: 5 (23% of Opp).

• {{channel_context}}Where the leads come from, what we know about lead quality by source.

  Mostly Google Ads paid search. Some PMax. No social. Sales feedback: "PMax leads are flakier."

Diagnose the conversion path below. Identify which stage transitions are broken vs healthy, and propose hypotheses for each broken stage.

FUNNEL:
{{funnel_data}}

CONTEXT:
{{channel_context}}

OUTPUT:

1. Stage-by-stage health assessment: each stage transition (lead→MQL, MQL→SQL, SQL→Opp, Opp→Won) gets healthy / weak / broken with the rate and your reasoning.
2. For each weak/broken stage, propose 3 distinct hypotheses about why. Don't pile up hypotheses about the same thing — make them genuinely different (e.g. "intent mismatch", "qualification too strict", "sales handoff bug").
3. For each hypothesis, propose a diagnostic test that would confirm or rule it out.
4. A prioritized investigation plan: which hypothesis to test first and why.
5. Channel-level callouts: do any of the hypotheses apply differently to PMax vs paid search?

Do NOT recommend "improve sales follow-up" as a primary hypothesis unless the funnel data clearly shows MQL→SQL is the broken stage.
Do NOT recommend "test more landing pages" without naming the specific failure mode you'd be testing for.

Expected output shape

Stage assessment + 3 hypotheses per broken stage + diagnostic tests + prioritized investigation plan.

Why we wrote it

CPL alone is a vanity metric in B2B. Multi-stage funnel diagnosis is the real work — and most agencies skip it because it requires coordinating with the sales side. This prompt structures it.

How to use

1. Open Claude or ChatGPT. The recommended model for this prompt is claude-opus-4-7 — opus when the prompt requires deep reasoning, sonnet for the rest.
2. Replace every {{variable}} with content specific to your account. The examples above are starting points, not templates to ship as-is.
3. Paste the prompt and run.
4. Read the output against the expected shape above. If the model produced a structurally different response, re-prompt rather than accept the drift.