For enterprise AI buyers

AI Vendor Assessment

Procurement questionnaires were not designed for AI systems. We assess what a vendor's AI actually does, what evidence supports it, and what you are committing to operationally and contractually.

Scope

What we test before you sign

Capability and evidence

Vendor AI claims mapped to the specific tasks you intend to run, then tested against evidence rather than demos.
Evaluation on your data shape and your edge cases, including failure and escalation behaviour.
Model dependency: which providers sit behind the product and what changes when they change.

Governance and commitment

AI governance: data handling, retention, training-use terms, human review and audit trail.
Lock-in: data portability, integration depth, and the realistic cost of leaving in year three.
Total cost under your projected usage, including inference-driven price escalation.

Claim review

A typical vendor claim, tested

Claim vs. evidence

Claim: “Your data is never used to train models.”
Evidence reviewed: Master agreement, DPA, provider sub-processor terms, logging configuration walkthrough
Finding: The commitment holds for the vendor's own training but the underlying provider tier in use does not carry the same term for a subset of traffic.
Confidence: Moderate
Investment impact: The assurance your risk committee relied on is narrower than stated. Require the enterprise provider tier in writing and a named sub-processor list before execution.

Claim vs. evidence

Claim: “Pricing is flat per seat regardless of usage.”
Evidence reviewed: Order form, rate card, uplift clauses, twelve months of comparable customer usage
Finding: Flat pricing applies below a per-seat inference ceiling; above it, overage bills at list.
Confidence: High
Investment impact: Budget certainty ends at the ceiling, which your projected usage crosses in year two. Negotiate the ceiling and a capped overage rate now, not at renewal.

Evaluating an AI vendor?

Bring the proposal, the contract draft and the use case. We will tell you which claims are supported, which are not, and what to change before signature.

Discuss an AI Investment See a sample report