AI Decision Point 5 — The 90-Day Vendor Review: Kill, Scale, or Renegotiate

I watched a founder sign a six-figure AI vendor contract and then disappear into daily operations for three months. When he came back up for air, the vendor had delivered nothing that mapped to his actual processes, and his team was quietly working around the tool.

Most AI vendor reviews fail because they happen too late, or they happen as a vibe check instead of a decision. At day 90, you need a hard stop: keep the vendor, kill it, or renegotiate the scope. No third option.

Definition: AI vendor review

A structured checkpoint—usually at 90 days—where the owner decides whether an AI vendor, integration, or pilot is delivering measurable value or consuming budget without return.

Definition: Kill criteria

Pre-agreed conditions under which you stop working with a vendor. They exist so you do not have to argue emotions at month three.

What does a 90-day vendor review actually decide?

It decides three things only:

1. Is the vendor solving the problem we hired them to solve?
2. Is the cost—money plus internal time—justified by the output?
3. Do we continue, renegotiate, or terminate?

Everything else is theater. If you walk into this review without written answers to those three questions, the vendor has already won because they control the narrative.

How do you prepare for the review?

Start before the vendor signs. At kickoff, agree on what success looks like at day 30, day 60, and day 90. Write it down. The vendor should sign off on it too.

At day 90, collect:

• The original scope and success criteria.
• Vendor delivery logs and actual outputs.
• Internal hours spent managing the vendor.
• Team feedback: who is using the tool, who is avoiding it, and why.
• Financials: license fees, implementation costs, overages.

If any of these items are missing, postpone the review. You cannot review what you did not measure.

What goes into the vendor scorecard?

Use a simple one-page scorecard. Rate each item as met, partially met, or missed.

If more than two criteria are missed, the default decision is kill or renegotiate—not "give them another quarter."

Tool tip (AiAdvisoryBoard.me): Before you sign any AI vendor, run a Plan → Fact → Gap check on your own processes first. If you cannot see what your team actually does today, you will not know whether the vendor changed anything at day 90. See how the 7-day diagnostic works.

What are the kill criteria?

Kill the vendor if:

• They miss the primary success metric by more than 30%.
• Your team is spending more time fixing vendor output than doing the work manually.
• The vendor cannot explain why a result happened (no audit trail, no logic).
• Data privacy or security concerns surface that were not disclosed upfront.
• The vendor asks for scope creep payments to reach the original promise.

Do not wait for all five. One is enough.

What does renegotiation look like?

Renegotiate when the technology works but the commercial model does not. Examples:

• Reduce scope to the one module that actually works.
• Shift from fixed fee to outcome-based pricing.
• Cut the license count to match actual users.
• Extend the pilot by 30 days with new, written success criteria.

Get the new terms in writing. Verbal promises from AI vendors age poorly.

Internal link candidates

If the review goes badly, you will need to communicate the decision upward or to the board without spin. See how to deliver unwelcome news upward without spin.

If you are evaluating whether to build internally instead of buying, see how to get your own AI project to a working state.

If you want to understand the true cost of tokens and API usage before you sign, see AI tokens: how to avoid overpaying for your own experiments.

Manager scan (2-minute digest example)

• Vendor: [name], contract value: [amount], live since: [date]
• Primary success metric: [target] vs actual: [result]
• Internal hours spent on vendor management this month: [number]
• Team adoption: [X] of [Y] licensed users active weekly
• Gap between promised output and actual output: [description]
• Decision: keep / renegotiate / kill

Micro-case (what changes after 7–14 days)

A 120-person services company hired an AI vendor to automate invoice matching. At day 90, the owner reviewed the scorecard and saw the vendor had reduced errors by 12% but consumed 18 internal hours per week in exception handling. The team was bypassing the tool and emailing spreadsheets instead. The owner killed the contract, brought the one working sub-task in-house, and redirected the budget to internal training. Within two weeks, the team had a working automation they owned, and the vendor dependency ended.

Note on this case: This example is illustrative — based on typical patterns we observe with companies of 30–500 employees, not a single named client. Specific numbers are rounded approximations of common ranges, not guarantees.

FAQ

Who should run the 90-day review? The owner or COO, with input from the team using the tool. Do not delegate it entirely to the vendor or to IT.

What if the vendor says they need more time? More time is acceptable only if you agree on new, written success criteria and a hard end date. Open-ended extensions are how pilots become permanent costs.

Can we renegotiate mid-contract? Yes, but only if the contract allows it. If not, treat the 90-day review as your data for the next negotiation or termination conversation.

What if we have no baseline to compare against? Then the review is premature. Build the baseline first—see what your team actually does for seven days—then reassess the vendor.

Conclusion

A 90-day vendor review is not about punishing vendors. It is about refusing to pay for hope. If the AI vendor cannot show you Plan vs Fact vs Gap in plain terms, they are not a partner—they are a line item.

Do this today: pull the last vendor contract, find the success criteria, and schedule the review before day 90 passes.

If you want a system that surfaces the Plan → Fact → Gap automatically—every day, across the company—see how the 7-day diagnostic works. https://aiadvisoryboard.me/?lang=en