The Exit Demo: Testing Whether an AI Tool Will Give Your Data Back

Most software demonstrations are designed to show what a product can do once your data is inside it. The unspoken assumption is that the difficult part is getting started. For UK SMEs adopting AI tools, the more consequential question may be the reverse: if the relationship ends, what exactly comes back out?

Data portability is often treated as a contractual footnote. In practice, it is an operational capability that should be tested before signature. An AI tool that can import a spreadsheet in minutes may still be unable to return the workflow logic, prompt history and derived data that give that spreadsheet its value. The result is not simply inconvenience. It is a switching cost that is difficult to see at the buying stage and difficult to escape once embedded.

Why import is the easy half

Suppliers have a commercial incentive to make onboarding straightforward. Import wizards, template libraries and migration support reduce friction at the point of sale. Export is different. It is used less often, requested less often and, in some cases, designed with less care.

For a UK business, the risk is cumulative. A tool may begin as a simple assistant for drafting or summarising. Over time it accumulates approved prompt libraries, workflow triggers, user permissions, evaluation notes and AI-generated classifications. Each layer makes the system more useful and the exit more complicated. By the time portability becomes urgent, the business may be dependent on records it never thought to request in a reusable format.

The practical response is to run an exit demo alongside the conventional product demo. Ask the supplier to show, not describe, how a customer would leave. The exercise should be treated as part of due diligence, in the same way a buyer would test access controls or review a service level agreement.

What usable export actually means

A CSV file containing names and email addresses is not a meaningful export for an AI workflow. Usable portability requires that the receiving system can reconstruct the working context, not merely the underlying records.

Four categories deserve particular attention.

1. Structured records and export formats

Ask for a sample export from a realistic account, not a blank template. Confirm whether the supplier provides machine-readable formats such as CSV, JSON or XML, or only PDF reports. Check whether field names are preserved, whether timestamps include time zones, and whether attachments are included or referenced by links that may expire.

A supplier that offers only screen-level reporting is not offering portability. It is offering visibility. The distinction matters when the data must be moved into a replacement system or retained for audit.

2. Workflow and prompt history

AI tools increasingly hold operational logic in prompts, agents, rules and workflow definitions. A buyer should ask whether these can be exported in a form that another platform could interpret, or at least reviewed and manually reconstructed.

Request an example of a prompt library export, including version history where available. Ask whether workflow definitions are exported as configuration files or merely described in a support article. If the supplier cannot provide a structured output, the business should assume that rebuilding the workflow will be a manual project.

3. Embeddings and derived data

Many AI systems create embeddings, indexes, classifications or summaries derived from source material. These are not always essential to portability, but their status should be clear. Can they be exported? Can they be regenerated from the source data? If they cannot be exported, what is the cost of rebuilding them elsewhere?

This is not an argument for demanding every derived artefact. It is an argument for knowing which ones matter to the business and whether the supplier treats them as customer data or as proprietary system state.

4. Deletion evidence

The exit process is incomplete if data remains in backups, logs or model-training pipelines without a clear retention rule. Ask what deletion means in practice, how long it takes, and what evidence the customer receives. A written confirmation is useful; a documented process with dates and scope is better.

For UK organisations, data protection obligations make this more than a commercial preference. The buyer should be able to explain to a regulator, auditor or customer how data will be removed and evidenced at the end of the contract.

The cost of leaving

The most important number in a portability assessment is not the export fee. It is the internal effort required to reconstruct the working environment elsewhere. That effort may include rewriting prompts, rebuilding integrations, retraining staff, validating outputs and maintaining parallel systems during transition.

A supplier may charge a reasonable exit fee, particularly where bespoke extraction work is required. The concern arises when the fee is unpredictable, the format is deliberately limited, or the timescale is controlled by the supplier. A business should ask for the exit fee schedule before signing and confirm whether it is capped or subject to change.

A simple scoring sheet

The following sheet can be completed during a demo or procurement review. Score each item from zero to two: zero for no evidence, one for a partial or verbal commitment, and two for a demonstrated and documented capability.

Test	Evidence to request	Red flag	Score
Structured export	Sample JSON or CSV from a realistic account	PDF-only reporting	0 to 2
Field fidelity	Export preserves field names, dates and attachments	Truncated or renamed fields	0 to 2
Prompt history	Export of prompt library and version history	No structured output	0 to 2
Workflow definitions	Configuration export or documented rebuild guide	Screenshots only	0 to 2
Embeddings and indexes	Clear statement on export or regeneration	Undefined ownership	0 to 2
Deletion evidence	Documented process with confirmation	Verbal assurance only	0 to 2
Exit fees	Published schedule or capped fee	Discretionary pricing	0 to 2
Timescale	Defined export window and support commitment	Open-ended process	0 to 2

A total below eight should prompt a serious conversation before contract. A score above twelve is not a guarantee, but it suggests the supplier has thought about the return journey as well as the arrival.

Putting the test into the buying process

The exit demo should be scheduled before the final decision, not after implementation. It can be run in thirty minutes if the supplier is prepared and considerably longer if they are not. Either outcome is useful.

Record the session and retain the sample export with the procurement file. Ask for the relevant clauses to be reflected in the contract, including export format, timescale, fee and deletion evidence. Where the supplier cannot commit, the business should price the rebuild risk into its decision.

In 2026, AI tools are becoming part of routine UK operations rather than experimental additions. That makes portability a mainstream operational concern. The question is not whether a supplier can accept your data. It is whether, on the day you need to leave, they can give it back in a form you can actually use.