An unused AI licence is easy to describe as an adoption problem. It may be evidence that the organisation bought the wrong tool, chose the wrong task, designed a poor workflow or failed to establish safe conditions for use.
The source claims two-thirds of mid-market firms pay for unused AI after a year and supports the argument with recurring “single-digit” adoption anecdotes. No research basis appears in the project. The decision does not need a market percentage. A firm can inspect its own use, cost and outcomes.
Past licence fees are sunk. Future cost includes renewal, internal effort, integration, assurance, support and the opportunity cost of continuing. The relevant question is whether a credible next intervention has more expected value than those future costs and risks.
Define what “unused” means
Seat login is a weak measure. Some tools support infrequent, high-value tasks. High use can also be harmful if staff apply a tool outside approved boundaries.
Review:
- Intended users and tasks
- Frequency and consequence
- Active use by task, not login alone
- Output accepted, changed or rejected
- Complete-workflow time and cost
- Errors, incidents and near misses
- User reasons for use or avoidance
- Client, privacy, security and professional constraints
Segment the answer. Five researchers using a specialist tool well may justify a smaller contract even if the original enterprise rollout failed.
Diagnose the constraint
The source uses four categories that overlap. A more precise diagnosis can combine them.
Problem fit. The purchase addressed a symptom or a need the team did not consider important. The source's document-summary example is memorable: researchers struggled to find the right documents, rather than summarise them. Verify the consultancy case before publication. It shows why workflow observation should precede procurement.
Task and capability fit. The tool performs well on high-volume, repeated work and the firm handles low-volume, highly varied cases, or vice versa. Test representative work instead of extrapolating from a demo.
Workflow fit. Users must leave approved systems, duplicate data or perform more review than the saved step warrants. The tool may be competent and the complete task worse.
Information and integration readiness. Permissions, data quality, source access or records are inadequate. Treating this solely as adoption can push staff towards unsafe copying.
Trust and change conditions. People may lack training, explanation, time, senior example or confidence that use is approved. They may also have a valid objection based on quality or client duty.
Governance boundary. Privacy, security, confidentiality, intellectual property, employment or sector concerns may properly constrain use. Low adoption can be the control working.
Write the evidence and confidence. “People resist change” is not a diagnosis.
Four decisions, including stop
Repair the use
Choose this when the task creates value and the constraint is addressable. Redesign the full workflow, configure approved sources, involve users and run a controlled comparison.
Training should use representative work and include limitations, data boundaries, review and escalation. Senior sponsorship supplies authority and capacity; it should not force use when evidence is poor.
Set a short decision period based on task frequency. Agree success and stop criteria before spending more.
Narrow the deployment
A tool may work for one team, role or task. Reduce seats, features, environments and integration to match that value. Confirm whether a narrower deployment changes risk and support.
The source claims firms routinely cut licence cost by 40% to 60%. Remove that unaudited range. Negotiate from actual usage, renewal leverage and alternatives.
Renegotiate or replace
Contract terms may allow a lower tier, different service, credit or revised period. Another product may fit better, though switching introduces evaluation, migration and supplier cost.
Do not use replacement to avoid fixing an unclear problem. A new interface cannot resolve weak ownership or source data.
Retire
Stop when value is absent, the task is wrong, risk is unacceptable or the credible repair costs more than the opportunity.
Retirement includes data export or deletion, account removal, integration shutdown, records, user communication, supplier confirmation and financial control. Check whether any downstream process now depends on the tool.
Read the contract before the decision window closes
Record renewal date, notice period, minimum term, price movement, data return and deletion, audit evidence, subcontractors, model changes, termination assistance and liability.
The source describes a 200-person financial services firm missing a 90-day notice period by two weeks and paying for another year. Verify the engagement and permission. The operational lesson is to assign contract ownership and alerts well before the deadline.
If the supplier supports other important systems, coordinate the commercial relationship without allowing that dependence to silence a poor product decision.
Explain the decision without rewriting history
The source proposes telling the board that the original investment was reasonable and the mismatch unknowable. That may be true. It should not become a script that excuses weak procurement.
A useful account says:
- What the firm expected
- Evidence available at approval
- What happened in real work
- Which assumption failed
- Cost and risk of each option
- Recommendation and learning
- Change to future evaluation
If the original pilot was too broad or integration went untested, say so. A board can accept a learning cost more readily when the organisation changes its method.
Budget redeployment is not the same as recovering sunk cost. Show the future choice accurately.
Preserve the learning
Before closing or renewing, answer:
- Which constraint mattered most?
- Which evidence would have revealed it earlier?
- What data, workflow or governance prerequisite was missing?
- Which users and tasks did create value?
- Which harms or near misses appeared?
- What should procurement test next time?
Update the AI inventory, approved-use guidance, supplier review and evaluation library. Keep representative failed cases as well as successes.
The source's best line survives: the licence fee is gone; the learning does not have to be.
Make portfolio review routine
Review AI tools by use, owner, value, risk, contract and next decision. Avoid waiting for an annual budget exercise while notice windows pass.
Distinction offers an AI portfolio review. The keep-renegotiate-cut promise should expand to include repair, narrow and governance stop, and the current service scope needs confirmation.
Doing nothing is still a decision with cost. The answer is not automatically cancellation. It is an explicit, evidence-led choice to repair, narrow, renegotiate, replace or retire before patience becomes the only business case left.
