AI can build the comparison matrix, but the buy decision still needs a human in the room.

The matrix structure is repeatable, but each evaluation involves unique vendor documents, a specific company context, and shifting market conditions. It's not a rote task — the judgment layer changes every time.

The matrix dimensions are well-specified, but 'best fit for a mid-market SaaS company' is underspecified — the agent doesn't know the company's existing stack, data team maturity, or budget ceiling, all of which drive the recommendation.

The RFP responses are presumably provided, but real pricing (especially negotiated enterprise tiers) is rarely in RFP docs and requires vendor conversations. Vendor stability signals require access to funding databases, G2/Gartner reviews, and recent news — accessible but noisy.

A CDP selection at 500M events/month is a multi-year, six-figure commitment. A flawed recommendation based on misread pricing or missed integration gaps could cost the company significantly in migration pain and sunk implementation costs.

The final recommendation requires weighing vendor relationship risk, internal engineering capacity, and strategic fit — factors that depend on organizational context the agent cannot access. Procurement decisions of this scale routinely hinge on intangibles AI cannot reliably model.