Free decision worksheet
Make the tool prove its value before the subscription renews.
A one-page, vendor-neutral checklist for testing real work, total cost, permissions, human review, and a safe exit.
Eight gates
AI tool buyer checklist
- 01
Outcome
Name one repeated job and the measurable result that must improve.
- 02
Baseline
Record current time, cost, error rate, and accepted-output rate.
- 03
Evidence
Test the tool on 10–20 representative tasks, including hard cases.
- 04
Review
Measure correction time and define who approves consequential output.
- 05
Access
List what the tool can read, write, retain, share, and trigger.
- 06
Economics
Count seats, credits, overages, setup, review, and switching costs.
- 07
Ownership
Verify output rights, training-data terms, deletion, and export paths.
- 08
Fallback
Document the manual path and test revocation before rollout.
Compare AI tools with a weighted selection matrix
Start with two or three candidates for one job, including your current manual process as a baseline. Remove any candidate that fails a required data, accessibility, integration, or budget condition before scoring. A high average cannot override a failed requirement.
| Criterion | Weight | Evidence to collect |
|---|---|---|
| Accepted-output quality | 30% | Pass rate on the same representative inputs, including difficult cases. |
| Full cost | 20% | Seats, usage, overages, API charges, setup, training, and review time. |
| Privacy and controls | 20% | Applicable plan terms, retention, deletion, permissions, and model-provider access. |
| Usability and review effort | 15% | Intended users complete the task; count corrections and approval minutes. |
| Integration and exit | 15% | Test required connections, export fidelity, access revocation, and manual fallback. |
Score each criterion from 1 (fails the agreed target), through 3 (meets it with manageable work), to 5 (exceeds it with repeatable evidence). Leave unverified items unscored and request evidence. Multiply each score by its weight and add the results for a total out of 5. For example, scores of 4, 3, 4, 3, and 2 produce 3.35/5 with these weights. This is an illustrative calculation, not a product rating.
Nonprofits handling donor information may give privacy more weight; a lean operations team may prioritize integration and review effort. Keep weights at 100% and apply them equally to every candidate. Choose a time-limited pilot only after required checks pass, with an owner, spending cap, success target, and cancellation date.
Download the editable scoring matrixCheck hidden costs before upgrading or switching
Compare the free and paid plans on the same workload. An upgrade earns its place when it removes a measured limit or supplies a required control. Include annual commitments, minimum seats, credit expiry, retries, storage, paid connectors, and separately billed model APIs. Divide the full pilot cost by accepted outputs rather than generated outputs.
Before replacing a tool, export a small sample and check formatting, attachments, permissions, and automation dependencies in the replacement. Keep a backup and manual fallback; revoke the old connections after migration is verified. Browse the pricing index, estimate subscription costs, or read our free versus paid decision guide.
Keep it current
Get the checklist and the weekly decision briefing.
We cover one meaningful change, one practical workflow, and one clearer buying decision each week. No daily launch dump.
Want to see the email first? Read a complete sample issue →