A GDPR review of a Zanus AI calling workflow starts with what personal information moves through it, for which purpose and under whose responsibility. This guide is a preparation worksheet for your privacy reviewer. It does not determine your lawful basis or certify that a vendor or configuration is compliant.
Map the workflow rather than the product name
Start at the phone connection and follow audio, transcripts, summaries, caller details and saved CRM actions. Include support access, exports and any separate services used to process the conversation. Record each organisation involved and the proposed purpose of its processing.
The ICO’s AI accountability guidance is a starting point for a UK assessment. The reviewer must consider the applicable current law and deployment, including any distinction between UK and EU requirements.
Prepare an evidence request
| Topic | Evidence to give the reviewer |
|---|---|
| Purpose and roles | Why each data item is processed and who makes the decisions |
| Supplier chain | Contractual terms, processing providers and locations |
| Retention | Periods and handling of copies, logs and backups |
| Individual requests | Access, correction, restriction or deletion procedures as applicable |
Request written deployment-specific answers from the vendor. Keep unknowns explicit instead of completing a questionnaire with guesses.
Use synthetic records for operational tests
Create an internal test call, then identify the copies it produces. Ask an appropriately limited staff role to retrieve or export it. Correct a detail and inspect the connected CRM and summaries to see where the correction propagates.
Test the ordinary process for removing a record and ask how retained copies are handled. These observations establish operational behaviour, not whether a request must legally be granted in a particular case. Preserve exceptions for the privacy reviewer to assess.
Keep deployment approval with accountable reviewers
Have the responsible privacy or legal professional determine roles, lawful basis, notices, contracts, transfers and whether an impact assessment is required. A data-residency answer alone does not settle all these questions. Recheck the assessment when purpose, supplier or configuration changes.
Only introduce real personal information after the required review and controls are in place. Keep a dated record of the reviewed scope, open questions and people responsible. A useful checklist supports that decision without turning a marketing claim into legal approval.
Questions and answers
Does knowing the storage region establish GDPR compliance?
No. The review also needs the processing purpose, responsibilities, access, contracts, retention and other applicable requirements assessed for the actual use.
Trace Every Data Category Before You Write a Single Policy
When a calling workflow handles audio, transcripts, summaries and CRM updates, a privacy review cannot begin with the vendor's documentation alone. It must begin with a complete inventory of data categories that the workflow touches, because the legal questions that follow depend entirely on what is actually collected, not what a feature description implies. The following sections offer structured ways to build that inventory and use it to stress-test your internal processes before any real caller data enters the system.
A data category inventory for a calling workflow should distinguish at minimum between these groups: voice audio, automated transcripts, AI-generated summaries or intent labels, caller-supplied details such as a name or account reference, agent-supplied annotations, CRM fields written automatically by the workflow, system logs generated by the platform infrastructure and any copies created during export, backup or support access. Each category may have a different retention behaviour, a different chain of processing responsibility and a different answer to the question of what happens when a caller exercises a right. Treating all of these as a single undifferentiated block makes the review harder to complete and harder to revisit when something changes.
For each category, record the proposed purpose in plain language. "Improving service quality" is not a purpose statement that a reviewer can work with; "retained to allow a supervisor to audit call handling accuracy within the current billing period" is. The specificity matters because purpose drives the lawful basis question, and that question belongs to your qualified reviewer rather than to a workflow configuration screen.
Build a Responsibility Matrix Alongside the Data Map
A data map answers where information goes. A responsibility matrix answers who makes decisions about it. For a calling workflow involving an AI vendor, your organisation and any downstream CRM or ticketing system, the matrix must name the decision-maker for each processing activity, not just the technical operator.
A useful responsibility matrix for this context covers at least the following dimensions for each data category: who determines whether the category is collected at all; who sets retention limits; who can authorise access by support or audit roles; who receives and acts on an individual's access or deletion request; and who is notified if a processing error or unexpected disclosure is discovered. If the answer to any of these is unclear, that is a finding to pass to the privacy reviewer, not a gap to fill with an assumption.
One practical way to reveal gaps is to construct a fictional scenario rather than a theoretical question. Consider an illustrative case: a caller, whom we will call Caller A, contacts a business to update their preferred contact address. The workflow captures the call audio, generates a transcript and writes a summary into the CRM. Caller A later submits a subject access request. Walking through each stage of that request reveals something concrete: who within the business receives it, which systems must be queried, whether the AI-generated summary is treated as personal data in its own right, whether the vendor's platform holds a retained copy that falls under a separate contract, and whether the response timeline is achievable given the access procedures actually in place. None of those questions can be answered by reading a product specification. They require the organisation to trace the actual path of Caller A's information through every system involved.
A second illustrative scenario adds a complication worth planning for separately. Caller B is not the account holder; they are a family member calling on behalf of someone else. The transcript may contain information about the account holder, about Caller B and about a third party mentioned in conversation. When the account holder later requests deletion of records relating to their account, the organisation must decide which records are in scope, whether Caller B's voice data has an independent standing and how to handle references to the mentioned third party. These are not questions a workflow setting resolves; they require human professional judgement on the specific facts. Mapping the scenario in advance, however, makes it possible to design a triage process so that the request reaches someone with appropriate authority rather than stalling at a helpdesk queue.
Design a Structured Pre-Trial Acceptance Exercise
An acceptance exercise using synthetic records is more useful than a checklist when it produces observable evidence rather than affirmative answers. The distinction matters because a checkbox can be completed with an assumption, whereas an observation either matches the expected behaviour or it does not.
A structured pre-trial exercise for a calling workflow might proceed as follows. First, generate a test call using entirely invented caller details, clearly marked as synthetic throughout. The invented name, reference number and scenario should be realistic enough to test the workflow's handling of typical field types, but contain no information that could accidentally identify anyone. Second, immediately after the call, document every location where a record appears: the platform's own interface, the connected CRM, any notification or summary sent by email or webhook, system logs and any scheduled export destination. Third, attempt to retrieve those records using a staff role with restricted access, to confirm that access controls operate as expected rather than as documented.
Fourth, correct a detail in the invented record and observe where that correction propagates. If the platform summary retains the original value while the CRM is updated, that divergence is worth noting. A privacy reviewer will want to understand whether an individual who corrects their own information would receive a complete correction across all copies or a partial one. Fifth, carry out the removal process for the synthetic record and document which copies are deleted, which are archived and which remain in backup or log storage with a different retention behaviour. The result of this step is not a compliance determination; it is operational evidence that the reviewer can assess alongside your retention policy and any contractual terms with the vendor.
The acceptance exercise should be signed off by someone named in advance, with a written record of what was observed and what questions remained open at the conclusion. Open questions belong in the review file, not in the completed-tasks column.
Plan for Configuration Changes as a Review Trigger
A GDPR review completed for one workflow configuration does not automatically cover the same workflow after a change. For a calling workflow, changes that could alter the data-protection picture include: adding a new downstream integration that receives transcript data; enabling a feature that stores caller audio for longer; changing the purpose for which summaries are used; bringing a new team or geographic location into scope; and any change to the vendor's own sub-processing arrangements. Each of these may affect the lawful basis assessment, the notice given to callers, the contracts in place or the transfer arrangements for data leaving the relevant jurisdiction.
A practical way to manage this is to maintain a brief configuration log that records the scope of each review, the date it was completed, the reviewer responsible and the specific configuration covered. When a proposed change is raised, whoever manages the workflow can compare it against the logged scope before implementation, rather than after. If the change falls outside the reviewed scope, the decision about whether a new review is required goes to the named reviewer rather than being resolved informally.
This approach also helps when staff responsible for the workflow change. A dated log with named reviewers and a clear scope description allows a new team member to understand what was assessed and what was left open, without reconstructing the history from email threads or version-controlled documents spread across multiple systems.
Prepare Callers and Staff for Transparency Obligations
Whatever form of caller notice is eventually required depends on facts your privacy reviewer must assess: the processing purpose, the lawful basis, the categories of data involved and the applicable law for your jurisdiction and deployment. What an organisation can prepare operationally, before that determination, is the content and delivery mechanism for potential notices, so that implementation time does not delay the final review.
For a calling workflow, the delivery mechanism question has practical dimensions beyond the legal one. Audio notices at the start of a call have a different user experience effect than a written notice on a booking confirmation. If the workflow serves existing customers, a privacy notice update distributed through an existing channel may be appropriate. If it serves inbound enquiries from people with no prior relationship, the notice may need to appear before the call is connected. Preparing draft options for each scenario, clearly labelled as drafts for legal review, means that the communication work can proceed in parallel with the legal assessment rather than waiting on it.
Staff training is a related preparation task. The people most likely to receive an individual's rights request are often not the people who designed the workflow. A customer-facing team member who receives an access or deletion request needs to know the name of the person to escalate it to, the expected response timeline and what they should and should not say to the caller in the interim. A brief internal briefing document covering those three points is achievable before the workflow goes live and reduces the risk of an ad hoc response that creates problems of its own.
Questions That Should Change a Purchase or Deployment Decision
When evaluating whether to proceed with a calling workflow deployment, the answers to the following questions materially affect the risk picture and the scope of work required before live use. They are framed as questions your team should put in writing to relevant parties, including the vendor, and retain the answers to.
Which organisations process caller audio and transcripts, and in which countries do those organisations operate their infrastructure? What contractual mechanisms govern those processing relationships, and are written agreements available for review before deployment? When a record is deleted through the platform's interface, which copies of that record are removed immediately, which are removed on a different schedule and which fall outside the deletion process entirely? If a support request requires access to a customer's call data, which individuals at the vendor can access it, under what conditions and with what notification to the organisation? What happens to retained data if the organisation ends its subscription?
None of these questions has a universally correct answer. The purpose of asking them is to produce documented, specific answers that a privacy reviewer can assess against the applicable requirements for your deployment. An answer that amounts to "contact us for details" is itself a relevant data point: it means the evidence required for the review is not yet in your possession, and deployment should not proceed until it is.
Completing this preparation work before a real-data trial keeps the review process grounded in the actual configuration your organisation will use, rather than a generalised product description. The result is a review file that a qualified professional can assess efficiently, a set of operational observations from the synthetic test, a documented scope that can be checked against future changes and a named individual responsible for each open question. That is a more defensible starting position than one assembled retrospectively after callers' information has already moved through the system.
Handle Edge Cases That Synthetic Tests Rarely Surface
A structured acceptance exercise with invented records tests the expected path through a workflow. It does not test what happens when a call ends unexpectedly, when a caller withdraws consent partway through a conversation, when a transcript is generated from a poor-quality connection and produces inaccurate text, or when two separate callers are confused with one another by a CRM field that was populated incorrectly. These edge cases are worth mapping before live deployment precisely because they are the situations most likely to create a rights request or a data-quality problem, and the least likely to have a ready procedure in place when they occur.
Consider an illustrative scenario involving a dropped call. A caller, whom we will call Caller C, is connected to the workflow and begins providing account details. The call disconnects before the conversation concludes. The workflow may still have generated a partial transcript, written an incomplete CRM record and logged the call attempt. Depending on the configuration, a summary label may have been applied based on fragmentary audio. Caller C later requests deletion of all records held about them. The organisation must then determine whether the partial transcript is identifiable, whether the incomplete CRM record contains enough information to be linked to Caller C with reasonable effort, and whether the applied label carries any inference about Caller C that warrants attention. None of these questions is answerable by checking a deletion button; they require someone with appropriate authority to assess the actual content of the partial records and apply judgement.
A second edge case worth planning for is a call in which a third party volunteers personal information about an individual who did not consent to participate in the interaction at all. A caller updating their own account may mention a family member's health condition, a colleague's contact details or a neighbour's complaint in passing. The transcript captures this. When the original caller later requests deletion of their own records, the organisation must decide how to handle the embedded references to the uninvolved third party. Designing a triage step that flags transcript content containing apparent third-party information, and routes those cases to a named reviewer rather than processing them automatically, is a preparatory decision that can be made before any real data enters the system.
A third edge case concerns staff error. A team member annotates a CRM record with a note intended for one caller but accidentally saved against another caller's file. The mislabelled record persists until either caller raises a query. An acceptance exercise that tests only the intended workflow will not reveal whether the correction process works across mislabelled records, because the test assumes correct labelling from the start. A supplementary test that deliberately introduces a mislabelling and then attempts to correct it produces more useful operational evidence than one that does not.
Document the Handover Between Workflow and Human Decision-Making
A calling workflow that routes, transcribes and summarises calls does not eliminate the points at which a human must make a decision; it moves them. Understanding exactly where those handover points are, and who is responsible for the decision at each one, is preparation work that sits alongside the data map and the responsibility matrix.
Handover points in a calling workflow typically include at minimum: the moment a call is escalated from automated handling to a human agent; the moment a CRM record generated by the workflow is reviewed or relied upon by a staff member; the moment a rights request is received and must be routed to the appropriate person; and the moment a discrepancy between the transcript and the caller's stated account is identified. At each of these points, a person makes a consequential decision. If the workflow does not make the handover explicit, the risk is that the decision defaults to whoever happens to be available, rather than to whoever has the authority and context to make it well.
For each handover point, a practical preparation step is to write a one-paragraph description of what information the receiving person needs, what decision they are being asked to make and what options are available to them. This is not a script; it is a decision brief. A brief for a rights-request handover might describe the request type, the systems that need to be queried, the name of the privacy reviewer to consult if the scope is ambiguous and the expected response window. A brief for an escalation from automated to human call handling might describe what the summary provided by the workflow should and should not be relied upon for, and what the agent should verify directly with the caller rather than accepting from the generated text.
Writing these briefs before live deployment also surfaces gaps in the workflow design. If the person writing the brief cannot identify who receives a particular category of decision, or cannot describe what information they would need to make it, that is a signal that the process is not ready for real caller data rather than merely underdocumented.
Assess Vendor Answers as Evidence, Not Assurance
When an organisation sends written questions to a vendor about data processing, retention and access, the answers it receives are a category of evidence, not a compliance determination. The distinction matters for how those answers are used in the review process. A vendor's written confirmation that data is stored in a particular region, for example, is a factual claim about infrastructure that the privacy reviewer can consider alongside questions about access, transfer and purpose. It does not resolve those questions by itself, and it should not be treated as doing so.
A practical way to handle vendor responses is to maintain a written log that records each question asked, the date it was sent, the date a response was received and the substance of the response. Where a response is incomplete, redirects to general documentation or declines to provide the specific information requested, that is recorded in the same log with the same date discipline. The log then forms part of the evidence file that the privacy reviewer assesses, including the gaps.
Where a vendor's response refers to a document rather than answering the question directly, the organisation should retrieve the specific section of the document that is claimed to address the question and include it in the evidence file. A general reference to a privacy policy or terms of service is harder to assess than a specific clause, and a reviewer working under time constraints is better served by a file that contains the relevant extract than one that contains a link to a lengthy document.
When a vendor response changes between the time of initial evaluation and the time of a review update, both versions should be retained. Changes in how a vendor describes its processing arrangements may be relevant to the review, particularly if a configuration or sub-processing arrangement has changed in the interim. The dated log makes this comparison possible without relying on memory or informal communication records.
Sources and next steps
This independent Sysevo guide proposes evaluation steps and reports no hands-on vendor test or compliance certification. Check the linked primary sources and Zanus AI’s current Front Office overview, then confirm the scope of your proposed deployment in writing.
Read Zanus AI data-residency questions for related context. To compare providers using the same requirements, explore Sysevo security and data controls or book a Sysevo demo. Sysevo publishes this guide and offers its own business software.
Related workflow guides
- My AI Front Desk vs Zanus AI: Compare Call Data Controls
- Zanus AI CRM Caller History: Keep Useful Context Accurate
For current Sysevo subscriptions and usage options, compare plans and pricing.


