Correction, 28 October 2026: an earlier version of this article described how Zanus processes and stores data, and what it charges, without a published source. Zanus's public pages reviewed for this guide do not state where data is hosted. Those statements have been removed, along with an oversimplified account of international transfer rules.

If you run a UK or EU business and an AI voice agent answers your calls, you need to know where the recordings, transcripts and customer notes are processed and stored. For AI data residency with Zanus, the starting point is that its public pages reviewed for this guide, including its AI Phone page and pricing page, do not state a hosting region; they link to separate trust and security pages. So the location of your data is a question to put to Zanus in writing. This guide explains what to ask, why it matters, and how to check the answers.

Why Data Residency Matters for AI Voice Agents

A call handled by an AI agent creates personal data from the first second: the caller's number, their voice, the transcript and anything they say about themselves. Under UK and EU data protection law you need to know which organisations process that data, in which countries, and on what legal basis.

Processing outside the UK or EU is not automatically unlawful. Transfers can rely on an adequacy decision (the EU and UK each recognise a list of countries, and US companies certified under the EU-US Data Privacy Framework or the UK Extension), or on safeguards such as standard contractual clauses with a transfer risk assessment. What matters is that the route is documented and that you can show it. Some businesses, contracts or sectors require data to stay in a particular region regardless; if that is you, residency becomes a hard requirement rather than one option among several.

How AI Data Residency Zanusai Works in Practice

An AI voice call usually passes through several systems: the telephone carrier, speech recognition, the language model that decides what to say, text-to-speech, and the database where the transcript and record are kept. Each can be run by a different company in a different country. A vendor can store the finished transcript in Europe while speech recognition runs elsewhere, so "where is my data stored" is not the same question as "where is my data processed".

For Zanus, ask for the list of subprocessors and where each runs, for every step from the incoming call to the stored recording. Zanus's AI Phone page describes transcripts and summaries written to the customer record and a voicemail register that keeps recordings with transcripts; those are the records whose location and retention you need confirmed. Read Zanus's trust and security pages alongside its data processing agreement.

Questions to Ask Before Deployment

  • In which countries is each step processed: telephony, speech recognition, the language model, text-to-speech and storage?
  • Who are the subprocessors, and how will you be told when they change?
  • Where are backups and disaster-recovery copies kept?
  • What transfer mechanism applies to any processing outside the UK or EU?
  • How long are recordings and transcripts kept, and can you set your own retention period?
  • Are your calls used to train or improve the vendor's models, and can you opt out?
  • How do you export everything, and how quickly is it deleted from all copies when you leave?
  • Which independent security reports or certifications can you see, and how recent are they?

A clear vendor answers each of these in writing. Vague answers, or answers that cover storage but not processing, are worth following up before you sign.

EU Data Residency Software and Compliance Standards

Some voice AI vendors offer a choice of hosting region; others run in one region only. Region choice usually depends on every component of the stack being available there, which is why a vendor may offer EU storage before it can offer EU processing for every model. Ask what region choice applies to your plan, not what the vendor offers in general.

Independent assurance helps. A recent SOC 2 Type II or ISO 27001 report shows how a vendor runs security, but it does not by itself say where data is located. Check the scope and the date of any report you are given.

GDPR Data Location and AI Agent Architecture

Trace a single call end to end on paper: where it arrives, where speech becomes text, where the AI decides on a reply, where the transcript and recording are stored, where backups go, and where any reporting or analytics tools copy the data. Every box on that diagram should have a named organisation, a country and a legal basis. Gaps usually appear in analytics exports and third-party tools added after launch.

The fewer separate vendors handle a call, the easier that diagram is to complete. Platforms that combine voice agents with a built-in CRM, including Sysevo, reduce the number of handoffs to document, though you still need the same answers from whichever vendor you choose.

Frequently Asked Questions

Where does Zanus store call data?

Its public product and pricing pages reviewed for this guide do not state a hosting region. Ask Zanus for its data processing agreement and subprocessor list, and read its trust and security pages.

Is data residency in the EU the same as GDPR compliance?

No. Location is one part. Compliance also covers lawful basis, security, retention, people's rights, processor agreements and breach handling.

Can I use a US-hosted AI voice platform under GDPR?

Often yes, if the transfer has a valid basis, such as the vendor's certification under the EU-US Data Privacy Framework (and the UK Extension for UK data), or standard contractual clauses with a transfer risk assessment. Some contracts and sectors require data to stay in region regardless, so check your own obligations.

What if I am outside the EU but serve EU customers?

GDPR can still apply to you when you offer goods or services to people in the EU. That does not automatically mean EU hosting, but it does mean the transfer and processing rules apply. Take advice for your situation.

Data location shapes which voice AI vendors you can use. Get the answers in writing before you deploy, not after. Book a call if you would like to talk through your data residency requirements.

Mapping the Ownership of Each Data Point Before You Sign

Before a single call is answered by an AI voice agent, a business needs to establish something distinct from data residency: data ownership. Residency asks where data sits; ownership asks who can read it, copy it, share it, or act on it without your instruction. These are related but separate questions, and conflating them is a source of later friction during audits, contract renewals and terminations.

For AI voice platforms specifically, consider three categories of data with different ownership implications. The first is operational data: call recordings, transcripts, summaries and the notes written to a customer record. These should be contractually yours from the moment they are created. The second is derived data: inferences the platform makes from patterns across calls, such as which topics cause callers to abandon or which responses reduce call duration. Whether derived data belongs to you, to the vendor, or is pooled across the vendor's customer base is a question whose answer varies and is often buried in terms of service rather than in a data processing agreement. The third is model-improvement data: whether your call content is used to train or refine the AI's underlying models. A vendor may keep this right unless you explicitly opt out, or it may be unavailable to opt out of on lower-tier plans.

When reviewing a data processing agreement, look for language that distinguishes these three categories. If a contract treats all three identically, that is worth clarifying in writing. A useful way to frame the question to any vendor is: after our contract ends and we have deleted our account, in what form does any data derived from our calls still exist within your systems, and who can access it?

Running a Practical Handover and Exit Test Before You Commit

One of the most revealing evaluations a business can conduct before deploying an AI voice platform is a structured exit test during the trial or onboarding period, not after problems arise. The purpose is to confirm that the data you put in can come out completely, in a usable format, and within a timeframe your operations can absorb.

To design a useful exit test, consider the following. First, put a small volume of representative data into the platform: fictional caller names, synthetic phone numbers and scripted enquiry transcripts. Keep a record of exactly what was entered and when. Then, at a point of your choosing, attempt a full export. Observe whether the export includes every record type the platform creates: not just transcripts, but voicemail recordings, call summaries, flagged items and any CRM-style notes generated by the agent. Check whether the exported files are in a format your other systems can read without specialist conversion tools.

After the export, submit a deletion request and ask for written confirmation that deletion has been completed across all copies, including backups and disaster-recovery systems. Note how long the confirmation takes and whether it specifies backup cycles. If a vendor's standard deletion confirmation does not address backup copies, ask a follow-up question in writing before treating the process as complete. The answers you receive during this test are more informative than documentation alone, because they reflect what the vendor's support and operations teams actually do rather than what the policy says they do.

This kind of acceptance exercise is worth conducting even if you have no immediate intention of leaving, because it surfaces gaps in portability and deletion that affect your ongoing data governance obligations, not just your exit options.

Distinguishing Customer Identity from Individual Requests and Responsibility

AI voice agents introduce a particular complexity that is worth working through operationally before deployment: the difference between who called, what they asked for, and who is responsible for what happened next.

Consider an illustrative scenario. A caller contacts a business using an AI voice agent and, during the conversation, asks to cancel a service and also raises a complaint about a previous charge. The transcript records both requests under a single call identifier. If the cancellation is processed and the complaint is not routed to a human reviewer, a question arises: was the failure to act on the complaint a system gap or a process gap? And whose responsibility was it to ensure the complaint was flagged?

This matters for data governance because the transcript of that call is a data record that documents both the customer's request and the business's response, or absence of one. If a customer later exercises a right of access and receives a transcript that records their complaint but shows no corresponding action record, that creates an accountability question. The data itself becomes evidence of process failure.

To manage this, businesses deploying AI voice agents should design call handling with explicit routing logic for different request types, and should ensure that the data record created for each call includes not just what the caller said but what action category the system assigned to each part of the conversation. Where a call contains multiple distinct requests, each should ideally generate a traceable action record rather than a single composite note. Reviewing a sample of complex multi-request transcripts during a trial period will reveal whether the platform's record structure supports this level of granularity or flattens everything into a single summary.

Evaluating Subprocessor Notification Arrangements

The subprocessor list a vendor provides at the time of contracting is a snapshot. AI platforms in particular tend to change subprocessors with some frequency, because model providers, infrastructure vendors and specialist services evolve. A subprocessor notification arrangement tells you how you will learn about changes and what your options are when a change affects your data residency or transfer position.

When reviewing notification arrangements, the relevant questions are not only how many days' notice you will receive but also what the notice will contain. A useful notification tells you which subprocessor is being added or replaced, what function it performs, where it operates, and what transfer mechanism applies. A less useful notification tells you only that the subprocessor list has been updated and invites you to review it.

Consider also what your contractual options are if a new subprocessor creates a residency conflict with your own obligations. Some vendor agreements allow you to object to a change and, if the objection is not resolved, to exit the contract without penalty. Others allow you to object but reserve the right to proceed regardless. The latter position creates a practical problem: your obligations do not pause while a vendor resolves its infrastructure preferences. Understanding this before you sign gives you time to negotiate or to select an alternative rather than discovering the limitation when a change notification arrives.

A comparison table for evaluating notification arrangements across vendors might include the following criteria and the reasons they matter for operational decision-making.

Criterion Why it matters
Notice period before change takes effect Determines whether you have time to assess the impact and make adjustments
Content of notification Affects whether you can assess the data residency and transfer implications without follow-up requests
Your right to object Shapes your leverage if a change conflicts with your own contractual or regulatory position
Consequences of unresolved objection Determines whether you have a usable exit right or are effectively bound regardless of your objection
Method of notification Email to a named contact is more reliable than a dashboard update you may not see promptly

Designing an Internal Accountability Record for AI Voice Deployments

Many businesses focus their pre-deployment effort on asking questions of the vendor and less on building the internal record that demonstrates they asked the right questions and acted on the answers. An internal accountability record is a working document, not a filed policy, and its purpose is to be useful to the person who needs to explain a decision rather than to satisfy a checklist.

A practical accountability record for an AI voice deployment might include the following sections. A description of the processing activity: what the agent does, which customer interactions it handles, and what data it creates. A record of the questions put to the vendor in writing and the answers received, with dates. A summary of the transfer mechanisms in place for any processing outside your home jurisdiction, cross-referenced with the subprocessor list as it stood at contract signature. A note of any residency or ownership questions that received incomplete answers, and what was agreed as a follow-up. A record of the exit test described earlier, including what was exported, what deletion confirmation was received, and any gaps identified.

This document is also the right place to record staff responsibilities. An AI voice agent may handle calls unattended, but a named person within your business should be accountable for reviewing flagged calls, handling escalations the agent cannot resolve, receiving subprocessor notification emails, and reviewing the accountability record when circumstances change. Distributing that responsibility without naming individuals tends to mean it belongs to no one in practice.

Reviewing the accountability record at the point of contract renewal is more useful than leaving it as a deployment artefact, because the vendor's subprocessors, your own obligations and the regulatory landscape may all have shifted in the intervening period. The questions that mattered at the start may need different answers by the time you renew.

A business that can produce a clear, dated, internally maintained record of its vendor questions, the answers received, the transfer mechanisms documented and the staff responsibilities assigned is in a materially stronger position when questions arise than one that relies solely on vendor documentation. Build that record as you go rather than reconstructing it from memory, and treat the first renewal conversation with your vendor as an opportunity to verify that the original answers still hold.

Independent buyer's guide published by Sysevo. Sysevo is not affiliated with, endorsed by, or partnered with Zanus, and Zanus is the trademark of its owner. Product details change often, so confirm anything that matters to your decision with the vendor directly before you buy. This guide is general information, not legal advice.