Why data location matters before a firm uses an AI tool
Ask where prompts, files, outputs and logs are processed, stored and accessed for the exact product tier. Check contracts and subprocessors, not only the vendor's headquarters or a residency badge.
Include confidential, privileged, personal, special-category and commercially sensitive information in the review, along with records used as regulatory evidence. Examine public chat tools, enterprise workspaces, legal platforms and private workflows separately; the category name does not establish where their data goes.
Before client data enters a service, identify the receiving systems, countries or regions, retention, training use, human access and available logs. Include the path beyond the main workspace, rather than relying on a location label for the vendor.
The direct answer for UK professional services firms
Assess processing and storage locations against the proposed task, contract, data protection requirements and client confidentiality. Ask for written terms covering locations, subprocessors, retention, training use, deletion and access for the exact product tier. A general security statement does not answer those questions.
For each proposed data class, check the exact service terms, tenancy, access, recipients and location evidence. Record the reviewer and decision conditions. Neither internal use, enterprise branding nor a UK GDPR reference establishes suitability for client material.
Map AI data flows before client data is exposed
What to check in the vendor contract and security pack
Request answers for the exact plan and workspace under consideration. Check which terms and controls apply to that tier, including whether settings differ from another plan. Cover each part of the data flow:
- Processing locations: where prompts, files, outputs, metadata and support logs are handled.
- Storage locations: where live data, backups, vector indexes, analytics logs and audit logs are stored.
- Subprocessors: every third party that can touch the data, including infrastructure, model, support and monitoring providers.
- Training and improvement use: whether prompts, uploaded documents or outputs can be used to train, fine-tune, evaluate or improve models.
- Retention and deletion: how long each data type is kept and what happens when a file, workspace or contract is deleted.
- Human access: whether vendor staff or contractors can review prompts, files or flagged outputs, and under what approval process.
- Audit evidence: whether the firm can see who sent what, when, from which account, and what the system returned.
Compare the DPA, training terms, retention, residency and subprocessor chain for the specific tier. A paid plan or certification should not replace that review. Check the scope of any certification against the service and data flow being assessed.
How the answer changes by tool type
For a public AI tool, check the terms for conversation retention, monitoring, service improvement, processing region, human review and deletion. Keep client-confidential data out of a use the firm has not approved. Do not infer the available controls solely from whether the account is free.
For an enterprise workspace, verify the DPA, training terms or opt-out, administrator controls, SSO, logs and retention settings. Check region coverage and subprocessors for the actual data types. A UK or EEA residency option is not a guarantee that every prompt, log, backup or support access remains there.
For a legal or professional-services platform, examine matter permissions, document handling and any claims about privileged workflows. Identify foundation-model providers and other subprocessors, the data each receives and the contract for each layer. Specialist branding does not establish confidentiality or privilege.
For a custom or privately hosted workflow, identify who operates logging, permissions, deletion, model-use rules, monitoring and quality review. Check external services as well as the private host. The hosting arrangement alone does not establish control over the entire data path.
Governance controls to put around data location
Record approved uses, prohibited information, user groups, processing and storage locations, retention, training position, DPA status, subprocessors, log availability and the review owner in a tool register. Link each entry to the terms and evidence examined.
Apply the registered limits in the workflow. Approval for internal marketing drafts should not extend to client matter files. For proposed client work, assess identity controls, permissions, logs and retention, and name the professional responsible for review.
Turn AI vendor checks into an operating model
A practical review sequence
- List the workflow and data categories the team wants to use with the AI tool.
- Separate public, internal, client-confidential, privileged and special-category data.
- Confirm whether the vendor is a processor, controller, or both for each data flow.
- Read the DPA, subprocessor list, data residency terms, retention schedule and security pack.
- Check whether prompts, files, outputs or logs are used for training, evaluation or product improvement.
- Decide what evidence must be kept: approval record, DPIA, vendor assessment, contract extract, audit logs and human review notes.
- Approve or reject the use case, then document the decision in a register staff can actually follow.
What a good policy decision looks like
State which tools and contracts cover each permitted task and information category, who reviews the work and what evidence to keep. Staff need a decision they can apply, including when to pause and ask for approval.
A firm's policy might permit internal knowledge drafting in a checked workspace, prohibit unapproved public-tool use for client documents, require legal approval for privileged material and specify review of regulated outputs. Set those decisions for the firm's actual tasks and information, rather than treating the example as a universal permission.
Implement AI controls in live workflows
