Maintenance buyer guide · Updated 2026-11-13
How to audit and maintain your AI stack.
AI tools change fast: pricing shifts, free tiers shrink, security terms update, and some tools get discontinued outright. A stack that was right six months ago can quietly become expensive, redundant, or risky. This guide gives a practical review cadence and a keep/replace/add/cancel framework, so maintenance is a scheduled check, not a surprise. Use avoid only for a prospective tool you decide not to buy; use cancel for a subscription already in the stack.
A model is retiring: decide before your workflow breaks
Treat a model sunset as an immediate workflow review, not a reason to buy another AI subscription automatically. For example, GitHub announced on November 13, 2026 that six Copilot models will be deprecated across Copilot experiences on December 14. The announcement names replacement options, with one replacement still unannounced. Check the official Copilot deprecation notice for the exact affected models and exceptions before changing policy.
- Find dependencies. Inventory pinned model names in coding agents, editor settings, prompts, and team policies. Name an owner for each affected workflow.
- Test the substitute. Run the same representative tasks with the current model and the vendor-supported alternative. Compare correctness, reviewer effort, latency, and actual usage cost; do not assume the suggested model is equivalent.
- Check access before migration. Confirm plan eligibility, admin model-policy enablement, and any exception or unannounced replacement. Keep a tested fallback for workflows that cannot switch safely before the deadline.
- Record the decision. Mark each workflow keep-with-replacement, replace-tool, or pause, with an owner, migration date, evidence, and rollback path. Recheck after the vendor completes the sunset.
When to review your AI stack
- Set a fixed cadence: quarterly for actively used paid tools, immediately after any pricing, security, or feature-deprecation notice.
- Review before every renewal date, not after — most plans auto-renew and refunds are rare.
- Review after any team change: new hires, role changes, or a workflow that shifted since the tool was first chosen.
- A lighter monthly check is enough for free tools: are we still using it, and has the free tier changed?
Signs of duplicated spend
- Two tools solve the same job — two meeting-notes tools, two general assistants — because different people picked different defaults.
- A team-wide seat license exists, but usage data shows only a few people log in regularly.
- A tool was bought for a pilot that ended, but the subscription was never cancelled.
- Multiple automation tools overlap because ownership of "the automation platform" was never decided.
Workflow drift: when a tool no longer fits
- The team has started working around the tool instead of through it — manual exports, copy-pasting between tools.
- The original job to be done has changed, but the tool has not.
- A newer tool now does the job in fewer steps or with a better privacy/security posture.
- Adoption has quietly dropped and nobody has asked why.
Pricing and free-tier change triggers
Review immediately when:
- A vendor announces a pricing change, tier restructuring, or usage-limit reduction.
- Your usage grows past what the current tier reasonably covers.
- A comparison on this site shows a materially different price for the same job.
- An annual renewal is approaching and the tool's value has not been checked in the last quarter.
Security and privacy change triggers
Review immediately when:
- The vendor updates its terms, privacy policy, or data-training defaults.
- The vendor discloses a security incident or data breach.
- Your compliance requirements change: a new customer contract, a new regulation, or a new data type handled.
- Admin controls you rely on — SSO, retention settings, audit logs — move to a different plan tier.
- A connected assistant gains a new workspace, data source, or permission to send, publish, merge, deploy, or delete. Reapprove the expanded scope before treating the existing pilot as safe to keep or scale.
Shutdown and provider risk signals
- The vendor is a very early-stage company with no clear monetization path.
- The product has not shipped meaningful updates in a long stretch.
- The vendor is acquired, or a major feature is folded into a larger platform — often a precursor to a standalone shutdown.
- Support responsiveness drops, or the vendor goes quiet on its public changelog.
Suspected unsafe actions? Contain first, audit second.
Keep, replace, add, and cancel are routine purchasing decisions. They are not an incident-response sequence. If a connected assistant has made an unauthorized change, or sensitive information may have been exposed, do not wait for the next renewal review to choose a bucket.
- Stop further impact. Pause the affected automation and restrict the specific connector or credential when it is safe to do so. Keep critical business workflows on a reviewed manual fallback rather than disabling unrelated systems.
- Preserve what happened. Record the tool, connected accounts, affected actions or data, timestamps, and the person responsible for investigation. Avoid deleting evidence as part of routine subscription cleanup.
- Escalate and verify.Notify the appropriate security or system owner, check actual downstream records and permissions, and follow the organization's incident process. Restore access only after the owner approves the corrected scope.
- Then make the stack decision. Once the immediate risk is contained, choose keep with tighter permissions, replace with a tested alternative, or cancel with a verified offboarding plan. Record the decision and its owner.
For a planned permission change, use the connected-assistant permissions checklist. For a planned exit, use the vendor offboarding checklist. Neither replaces your organization's incident-response procedure.
The keep / replace / add / cancel framework
Sort every tool in the stack into one of four buckets before a renewal date or a scaling decision. This is the same framework the stack quiz's current-stack audit applies automatically once you add your current tools.
Keep
Actively used, still the best option for the job, and pricing/security terms are current. No action needed beyond the next scheduled review.
Replace
Still needed for the job, but a pricing, security, or fit change means another tool may serve it better. Before switching, rerun the same recurring job with the incumbent and strongest replacement under the same success criteria, allowed data, and review boundary. If the replacement wins, plan the migration before the next renewal.
Add
A real gap exists: a job is being done manually or with an unofficial workaround, and a tool would remove a measured bottleneck.
Cancel
Already in the stack, but low-use, duplicated, or no longer acceptable on security, privacy, or fit. Cancel or decline to renew after planning any needed migration.
Record the decision before changing the stack
A keep, replace, add, or cancel label is a starting decision, not a completed handoff. For each tool, record the current job, the evidence behind the decision, the accountable owner, the next review date, and the action that must happen before renewal or rollout. Use the AI Stack Decision Memo so the reason and stop condition survive a team or budget change.
For keep, name the next review trigger. For replace, keep the incumbent until the same-job pilot and migration owner are clear. For add, require a bounded pilot and an approved data-access scope. For cancel, confirm dependent workflows, exports, access revocation, and billing closure before marking the decision complete. If the evidence or owner is missing, leave the action open rather than silently treating the label as approval.
Buyer templates for this audit
- Connected AI Assistant Permissions Checklist
Use this before keeping or expanding an assistant whose connected apps or delegated actions have changed. Record the approved scope, human approval gates, and a tested revocation path.
- Vendor Pricing and Security Review Checklist
A due-diligence checklist for reviewing an AI vendor's pricing terms and security posture before signing.
- AI Tool Renewal Review Checklist
A copyable review checklist for deciding whether to renew, reduce, replace, or cancel an AI tool based on current usage, value, risk, and switching cost.
- AI Vendor Offboarding Checklist
A copyable exit checklist for turning off an AI vendor without losing data, leaving seats active, or breaking downstream workflows.
- AI Tool Pilot Checklist
A step-by-step checklist for piloting a new AI tool with one team before buying seats broadly.
Related workflow guides
- AI Tools for Workplace Productivity
A suite and inbox productivity stack for teams choosing where AI belongs across email, docs, meetings, and internal knowledge.
- AI Tools for Workflow Automation
A workflow-automation stack for teams deciding when to connect apps, add AI agents, and put human approvals around cross-system work.
Turn this into a real audit
Add your current tools to the stack quiz for an automatic audit.
The quiz's optional current-stack step applies the same keep / replace / add / cancel framework to the tools you already use, alongside the deterministic recommendation for your role, workflow, budget, and privacy bar.
Pricing-relevant comparisons
- ChatGPT vs Claude
A practical comparison for teams choosing a general AI assistant for writing, analysis, research, and lightweight coding help.
- Zapier vs n8n
A practical comparison for teams choosing between no-code AI orchestration and a more technical workflow automation platform.