Half of every company's AI tool spend in 2026 is on tools that are barely used. SaaS auto-renewal makes this invisible. Here are the five signals that say cancel, with what to do instead.
Cancel when: (1) weekly active usage is below 25%, (2) the workflow it serves can be done in Claude/ChatGPT, (3) you can't name a single person who would notice if it disappeared, (4) you can't articulate the ROI in one sentence, (5) renewal is coming up and you're not sure why.
Pull the usage dashboard for any AI tool you pay for. If less than 25% of seats are active weekly, you are paying for storage of unused accounts.
Action: reduce seats to active users. Often saves 50-70% without anyone noticing.
Many specialized AI tools are thin wrappers on Claude or GPT. If your team's actual workflow could happen in your LLM platform (which you also pay for), the specialized tool is redundant.
Action: build a Project that does what the tool does. Test with one workflow owner for 30 days. If it works, cancel the tool.
Mental test: if the tool stopped working tomorrow, would anyone email you? If no — cancel.
Action: send a survey: "would you be affected if [Tool] went away?" Crickets = cancel.
If the original buyer of the tool left or rotated roles, the rationale often left with them. Ask the current team to articulate the ROI; if they can't, the tool is invisible.
Action: investigation period. If after 2 weeks no one can defend it, cancel.
Vendors time auto-renewals to minimize cancellations. If renewal is approaching and you can't justify it, that's the signal.
Action: always set a calendar alert 60 days before renewal. Use that time to decide consciously, not by default.
Block 2 hours every January and every July. List every AI tool you pay for. Apply the 5 signs above. Cancel ruthlessly. This single discipline saves most $5M-$50M companies $20K-$100K per year.
Twice a year minimum. Quarterly if your spend is over \$10K/month.
Downgrade if there's a real cheaper tier. Cancel if usage is genuinely zero.
Read the contract. Usually you can give notice but pay through the term. Worth it if the tool is unused.
Yes — useful feedback for them, and they may surprise you with retention pricing if the tool is actually useful.
Then it wasn't really unused. The audit was wrong. Rare but possible — usage data sometimes misses real value.