← all hypotheses

Vendor Claim Cross-Examination Pack for Support Directors

exhausted [TRIANGULATED] signals: 1 independent
What is this?
A pre-renewal interrogation tool for directors of support at 50-500 person B2B SaaS companies managing 3+ external support vendors (BPO desks, AI deflection, automated triage, KB suppliers). The director pastes the vendor's verbatim QBR claim — '38% L1 deflection over Q1', 'AHT reduced 22%', 'CSAT +0.4' — including any footnotes, methodology slides, or caveats. AE's adversarial multi-model debate engine interrogates the claim's internal structure and produces a ranked cross-examination pack: which ticket types are silently excluded, which time window is cherry-picked, which baseline is undefined, which escalations are not counted as failures, which caveats gut the headline number. The 6-pattern autopsy taxonomy maps directly: Temporal & Transmission Blindness flags undefined windows; Concession Laundering flags hedges that nullify the claim; Cosmetic Confidence flags unfalsifiable phrasing. The director walks into QBR with 8-14 specific questions the vendor must answer before any renewal signature, and a graded artefact for CFO defense. AE never sees helpdesk data; the director's own Zendesk pull confirms the structural gap AE predicted.
Why did we consider it?
Support directors face vendors with home-field forensics advantage at renewal time; AE's adversarial debate plus 6-pattern autopsy is uniquely shaped to produce CFO-defensible cross-examination packs at solo-operator scale.
What breaks?
  • Episodic usage guarantees high churn: QBRs and renewals are infrequent, making a recurring subscription model fundamentally unviable.
  • Execution burden remains on the buyer: The tool only suggests questions; the Director still has to manually pull and analyze Zendesk data to prove the vendor wrong.
  • Commoditized output: Standard LLMs can already identify statistical cherry-picking and missing baselines in marketing claims without requiring a specialized engine.
What did we learn?
Killed: evidence_search_exhausted.

Evidence

Signal D — Demand proxy

{"found":true,"summary":"A LinkedIn article on vendor governance identifies that vendor management is fundamentally an evidence-generation problem and that spreadsheet-based approaches fail — directly validating the hypothesis that current vendor claim review processes are inadequate and need structured tooling.","sources":["https://www.linkedin.com/pulse/vendor-governance-third-party-evidence-chains-from-christiansen-vm7xc"],"reason":"Single LinkedIn article discusses the vendor governance evidence gap. However, no forum discussions (Reddit, HN, Stack Overflow), GitHub issues, or broader comm…

Evaluation history

WhenStagePhase
2026-05-09 16:43evidence_searchevidence_hunt
2026-05-09 16:37evidence_searchevidence_hunt
2026-05-09 16:25evidence_searchevidence_hunt
2026-05-09 16:19evidence_searchevidence_hunt
2026-05-09 16:13evidence_searchevidence_hunt
2026-05-09 16:06evidence_searchevidence_hunt
2026-05-09 15:54evidence_searchevidence_hunt
2026-05-09 15:49evidence_searchargument
2026-05-09 15:42audience_simulationargument
2026-05-09 15:36red_team_killargument
2026-05-09 15:24steelmanargument
2026-05-09 15:21genesisargument