Before You Buy AI Software, Run the Exit Test

AI software is often bought for what it can do on day one. The expensive question is what happens on the day you need to leave.

A polished demo can show faster writing, smarter analysis or a cleaner customer-support queue. It rarely shows whether you can retrieve your inputs, outputs, settings, audit history and attachments in a usable form. Nor does it reveal how long a migration will take, what continues to be billed during the move, or whether the supplier will confirm deletion after the contract ends.

That is why an exit test belongs in AI-software due diligence. It is not a prediction that the product will fail. It is a practical check that your organisation can change direction without losing control of its information, workflow or budget.

What the exit test actually asks

A credible exit is more than finding an “Export” button. Before buying, establish whether your team could:

• retrieve the records it needs in a structured, reusable format;
• understand what the export includes and excludes;
• rebuild or transfer essential workflows without undocumented dependencies;
• stop renewals and charges on a known timetable;
• remove users, integrations and credentials safely; and
• obtain a clear account of retention, backup and deletion after termination.

This is a commercial resilience test, not merely a privacy exercise. The UK Information Commissioner’s Office explains that the legal right to data portability applies only in specified circumstances and to particular personal data. It does not guarantee a complete export of a business workspace, generated material, configuration or supplier-created data. Buyers therefore need contractual and technical evidence beyond a general reference to “GDPR compliance”.

1. Test the export, not the promise

Ask for a sample export or use a trial account to create a small but representative workspace. Include text, files, tags, user roles, comments, dates, model outputs and one automation. Then export it.

Open the result without relying on the original service. A folder of PDFs or screenshots may be readable, but it is not necessarily reusable. CSV, JSON or another documented structured format is usually more useful for records that need to be searched, transformed or imported elsewhere. UK government guidance for cloud tools similarly recommends exporting data in a reusable format, giving CSV as an example.

Check completeness field by field. Does the export preserve identifiers, timestamps, relationships between records, attachments, version history and status? If an AI tool stores prompt templates, knowledge-base chunks or evaluation results, ask whether those are included. A supplier may truthfully offer “data export” while leaving out the parts that make the data operationally useful.

2. Measure reconstruction, not just extraction

An export is only the first half of portability. The second half is the work required to use it somewhere else.

List every dependency the tool has accumulated: integrations, webhooks, API keys, custom fields, prompt libraries, approval rules, user permissions, reports and staff habits. Mark which items can be recreated from documentation and which depend on supplier-specific behaviour. Government guidance on technical lock-in recommends considering exit costs early, periodically retesting switching assumptions, and favouring open standards and formats where that trade-off makes sense.

Do not demand zero lock-in. A specialist feature can create genuine value. The decision question is whether the benefit is worth a known and accepted switching cost. A team may rationally choose a proprietary workflow if it saves substantial work, provided the dependency is visible, budgeted and reversible enough for the risk involved.

3. Read cancellation, retention and deletion together

These are three different controls. Cancellation stops renewal under the contract. Retention determines how long the supplier keeps information. Deletion concerns what is removed from live systems and, where relevant, backups.

Before purchase, find the answers to five questions: Who is authorised to cancel? How much notice is required? When does access end? How long are exports available after cancellation? What confirmation is supplied when deletion is complete? Guidance for organisations choosing SaaS notes that a customer may need to ask the supplier to delete data securely and confirm that it has done so.

Watch for a dangerous sequence: access ends immediately, the export takes days to prepare, and the contract gives no recovery period. The safer sequence is to create and verify the export, transfer the required records, revoke integrations, cancel, then follow the documented deletion process.

4. Check account ownership and handover

A technically portable system can still be trapped behind the wrong account. Confirm that the organisation—not an employee, contractor or agency—controls the primary administrator identity, billing relationship and recovery route.

• Use a managed work address for the owner account.
• Require more than one authorised administrator where appropriate.
• Record who controls single sign-on, API credentials and connected storage.
• Document how ownership transfers when a staff member or supplier leaves.
• Test removal of a user without breaking shared automations.

This is particularly important for small teams. A founder may remember how everything connects today, but a buyer, successor or new operations lead needs evidence that does not live in one person’s memory.

A practical evidence table

Question	Useful evidence	Hidden cost if unclear
Can we retrieve complete records?	Sample export, field list and format documentation	Manual copying, lost context or unusable archives
Can another system use the data?	Successful test import or documented transformation	Cleaning, mapping and developer time
Can we rebuild the workflow?	Integration map, configuration record and runbook	Downtime, duplicated work and retraining
Can we stop charges predictably?	Renewal date, notice period and cancellation confirmation	Overlap subscriptions or another billing term
Will data be removed appropriately?	Retention schedule, deletion process and confirmation route	Security, compliance and record-management exposure
Can control be handed over?	Named organisational admins and credential inventory	Account recovery delays or loss of access

A 60-minute pre-purchase exit test

1. Create representative data. Add several records, an attachment, a custom field, two users and one integration or automation.
2. Export it. Time the process and note whether support or a paid tier is required.
3. Inspect it elsewhere. Open the files independently and compare important fields with the live workspace.
4. Map the rebuild. Estimate the people, permissions, transformations and retraining needed to move.
5. Read the termination terms. Record renewal, notice, access, retention, deletion and refund rules.
6. Ask one precise question. Send the supplier the most important unresolved point and keep the answer with the buying record.

If the test cannot be completed in an hour, that is not automatically a rejection. It tells you where deeper due diligence is needed. Enterprise systems will require a longer exercise; a low-risk individual tool may justify a lighter one.

Illustrative switching-cost calculation

This example is illustrative, not a customer result. Suppose four people use an AI tool at £25 each per month, so the base subscription is £100 per month. A move requires 18 staff hours at an internal cost of £30 per hour (£540), two months of overlapping subscriptions (£200), and six hours of retraining (£180). The estimated exit cost is £920—equivalent to 9.2 months of the original subscription.

The arithmetic is simple, but it changes the buying decision. A cheaper monthly price may not be cheaper over the full relationship. Conversely, a higher-priced product with a complete export, documented APIs and clear termination process may carry a lower total risk.

Who should proceed—and who should pause

Proceed with greater confidence when the supplier provides a representative export, explains exclusions, uses documented formats, states cancellation and retention rules plainly, supports organisational admin control, and answers evidence questions consistently.

Pause or limit the pilot when exports are available only after purchase, critical records arrive only as screenshots or flattened documents, the workspace belongs to an individual account, renewal terms are hard to find, or deletion is described vaguely. The same caution applies when the product’s value depends on an integration that cannot be transferred or recreated.

The alternatives are valid

Buying now is not the only sensible option. You can continue with an existing tool, run a month-to-month pilot with non-critical data, separate source records from the AI service, choose a product based on open formats, or postpone the purchase until a clear workflow owner exists. For a broader financial comparison, use the 12-month SaaS build-versus-buy test rather than comparing sticker prices alone.

Compare with an exit plan in hand

If you are exploring acquisition or software opportunities, write down your required export, handover and deletion evidence before browsing. MOC Marketplace is part of the MaryChuks.com business ecosystem. Its catalogue can be used for general discovery, but a marketplace card is not a substitute for verifying the seller’s evidence, terms and a working transaction path.

Explore the MOC Marketplace catalogue with the exit-test checklist beside you.

---

Sources and further guidance

• Information Commissioner’s Office: Right to data portability
• GOV.UK: Managing technical lock-in in the cloud
• GOV.UK: Using cloud tools securely
• GOV.UK: Securing SaaS tools for your organisation