B2B evidence protocol¶
Readiness verdict¶
Decision and four options are ready; the demo baseline is reproducible. Organizational fitness for purpose, owner, rights and prior expectation are not established. The counterfactual is designed but unassigned; no outcome is observed; causal attribution is blocked pending a comparator and 2–3 cycles.
Reproduced baseline¶
Calculations derive from the simulated commercial dashboard observed July 21, 2026. Repository CSVs are audit aggregates, not real CRM data.
Compliance and Proposal¶
- June 2026:
381,061.4 / 475,657 = 80.11%, the lowest weighted compliance from January 2025 to June 2026; all 8 demo sellers were below 90%. - Jan–Mar 2026 Proposal baseline:
(46.5 + 50.4 + 47.2) / 3 = 48.03%. - Apr–Jun deterioration:
(37.4 + 34.5 + 37.7) / 3 = 36.53%; change−11.50 pp.
The pre-decline window requires sensitivity against full-year 2025 and seasonality before real use.
Channels¶
| Channel | Won / closed | Win rate | Wilson 95% |
|---|---|---|---|
| Event / trade show | 5 / 56 | 8.93% | 3.87%–19.26% |
| Referral | 10 / 126 | 7.94% | 4.37%–13.99% |
| Digital inbound | 4 / 111 | 3.60% | 1.41%–8.90% |
The ranking favors event and referral, but intervals overlap widely. “Significantly superior” is unsupported. Option B remains blocked pending more observations, mix adjustment or a specific test.
Preferred design: hold-out + difference-in-differences¶
Define eligible sellers, regions or portfolios; match on prior conversion, volume, ticket, channel mix and seniority; assign intervention and operational comparator; preregister action, expectation and horizon; calculate:
[ \widehat{\tau}{DiD}=(Y) ]}-Y_{I,pre})-(Y_{C,post}-Y_{C,pre
Review parallel trends, composition changes and atypical deals; report estimate, interval and sensitivity. Without hold-out, historical series or forecast is weaker and normally remains observed-noncausal.
Metrics and learning¶
Primary: Proposal→Negotiation conversion. Secondary: sales compliance. Guardrails: quality prospects, cycle time and margin/discount. Process: 0–10 rubric. One cycle inspects implementation, not effectiveness; two is the case minimum and three preferred. Do not institutionalize good outcomes from weak process; review hypotheses after poor outcomes from sound process. Log every post-action definition, group or expectation change.