Purpose

Diagnose why growth efficiency is weak, not just that it is weak, and route each root cause to the right lever.

Rules

• Get every metric from compute_saas_metrics(company_id, period). The definitions below are for interpreting and reviewing tool output. Never compute them yourself.
• The tool reports which variant it used (for example, EBITDA vs FCF margin in Rule of 40). Quote the variant.
• Screening thresholds live in project://policies and are set by the fund. Use them. Don't substitute remembered "industry benchmarks". For peer context, call get_benchmarks(metric, peer_set) and cite the result.

Metric definitions (house standard)

Data-quality checks (before interpreting)

• Bookings vs ARR. Multi-year prepaid contracts, one-time fees, or services revenue counted as ARR inflate every ratio. Check the ARR definition in the data inventory.
• Capitalized commissions. Under ASC 340-40, commissions may be capitalized. CAC must use commissions paid or expensed as incurred, not amortization. If you can't tell, flag it.
• S&M allocation. Confirm that customer success and account management are classified consistently. Moving them between S&M and COGS shifts both CAC and gross margin.
• Entity duplication. Logos with several legal entities distort customer counts and ARPA. Check with the entity-resolution notes in the data inventory.
• Period mismatch. Every ratio must use aligned periods. The tool enforces this, but check it when combining outputs.

Diagnostic trees

Weak CAC payback or magic number. Decompose in this order:

1. Spend: S&M growth vs new ARR growth, and mix by channel and segment.
2. Conversion: win rate and pipeline coverage trend (CRM).
3. Deal size: ACV trend by segment. If ACV is falling, check discounting with pricing-value-creation.
4. Capacity: ramped vs unramped reps, attainment distribution.
5. Margin drag: if subscription gross margin is low, payback is weak even with healthy acquisition. Go to the Low gross margin tree.

Weak NRR. Split into GRR and expansion:

• GRR weak → hand off to customer-retention.
• GRR healthy but expansion weak → packaging and upsell path. Hand off to pricing-value-creation.

Low gross margin. Hosting cost per customer or unit of usage, support cost per customer, and services mix. Automation candidates go to ai-opportunity-assessment.

High burn multiple with healthy payback. Look outside go-to-market: G&A and R&D growth vs revenue growth, from get_financials.

Opportunity framing

Never size "grow faster" as an opportunity on its own. Growth levers belong to pricing, retention, or a specific GTM change with evidence.

Output contract

• Metric table: metric, value, period, variant, evidence id, and peer percentile if benchmarked.
• Findings, one per root cause, each with evidence ids, recorded with record_finding.
• Hand-offs to other lever skills, with the reason.
• Data-quality flags that change interpretation.
• NEEDS_EVIDENCE items.

Numeric claims and server validation

Before writing a numeric finding or proposal, call get_numeric_sources(company_id). Copy an exact catalog key and use {{quantity:KEY}} in prose; for example Legacy share is {{quantity:pricing.legacy.legacy_arr_share}} only when that key is returned. The server renders metric, value, unit, company, period and evidence together. Do not paste bare values into record_finding or propose_opportunity text, invent keys, or use identifiers as numeric sources. Qualitative wording is allowed. Structured scenario-rate/cost fields remain assumptions and must not be recast as measured facts.

Proposal, draft and submission calls enforce sufficiency, source freshness, policy eligibility, evidence and overlap checks. If rejected, record the named gap and resolve it; rewording the same unsupported or overlapping proposal does not make it eligible. Worked examples below were regenerated through a scripted MCP replay on the date shown in each file. They illustrate executable calls and returned outputs, not a human acceptance session.

References

• references/worked-example.md: tool outputs and findings for the fictional Cedar fixture, generated from tool calls; see its scripted replay date.