"We never train on your data" is a sentence that appears on a lot of websites. It is only useful if it is precise. This article explains what we mean by it, in plain language, so you can check our answer against your own requirements.
Your calls are not used to train the shared models that other customers use, so your AI voice agent training data stays with your account. Your transcripts and recordings are not used either. Fixing an answer in your own account changes your agent and nobody else's, and you choose what is kept and where it is hosted.
Two different things people mean
When a buyer asks whether their data is used for training, they usually mean one of two things.
- Is my data used to improve a shared model that other customers benefit from? This is training in the strict sense.
- Is my data used to improve how my own agent behaves? This is configuration and learning inside your own account.
We treat these very differently. Your calls are not used to train shared models. Your own corrections, such as a fixed answer in your knowledge base, change how your agent behaves and nobody else's.
Fixing an answer in your account improves your agent. It does not teach anyone else's.
Where a call goes
A call has several stages, and the data lives in different places at each one.
- The audio stream. The caller's speech is processed in real time to understand it and to produce a reply.
- The transcript. A text record of what was said, used for the call record and for your team to review.
- The recording. Where your rules allow it, the audio is stored so you can replay it.
- The structured result. Outcome, summary, actions taken and next steps, written into your systems.
You choose what is kept. A business that must not retain recordings can switch recording off and keep only the structured result. A business that must keep everything for compliance can do that instead.
In-region hosting
Data residency matters, particularly for banks, healthcare and government. VoiFlow can be hosted in the UAE, Malaysia, the UK and the US, so that your call data stays in the region your regulator expects. The region is a deployment decision made with you, not a default we pick behind the scenes. For the UAE rules, see UAE PDPL and AI voice agents.
What is kept, and for how long
Retention is set per customer. Common patterns include:
- Keeping transcripts and summaries for as long as the customer record is live.
- Keeping recordings for a fixed period, then deleting them.
- Redacting sensitive details, such as card numbers or identity numbers, before anything is stored.
Whatever you choose, it is written down as a rule and applied consistently. Ask for the retention you need in writing before you start.
Third parties in the chain
A voice agent relies on other services, such as speech and language processing. The question to ask any vendor, including us, is whether those providers retain or train on your data. Your contract and our data processing terms should state the answer clearly. If they do not, treat that as a gap.
What "never reused" covers
For the avoidance of doubt, we do not use your customers' calls, transcripts or recordings to train shared models, and we do not share one customer's content with another. Content in your account is used to run your service, to show you your own records, and to improve your own agent when you tell it to.
Controls that sit alongside it
Not training on data is only one part of handling it responsibly. The rest is control, covered on the control page:
- Permissions per action, so the agent can only do what you allow.
- Consent and opt-out handling per channel.
- Redaction of sensitive details.
- An audit trail of what the agent did and why.
- An instant pause if something looks wrong.
Questions to put to any vendor
Whoever you buy from, ask these and expect a written answer:
- Is any customer content used to train or fine-tune models shared with other customers?
- Which sub-processors touch the audio and the text, and where are they hosted?
- Do those sub-processors retain or train on the content?
- How do I delete a customer's records, and how long does it take?
- What can I see in the audit trail, and for how long?
A good answer is short and specific. A vague answer, or one that says "industry standard", is a reason to keep asking. For a wider list, see the security checklist.
What to do next
Write down your requirements before you talk to any vendor: region, retention, recording, redaction. Then ask each one to answer in writing.
Read more on our trust page, or ask us for the data processing terms alongside a pricing conversation.
Frequently asked questions
Does VoiFlow train shared models on my calls?
No. Your calls, transcripts and recordings are not used to train models shared with other customers, and one customer's content is never shared with another.
Does correcting an answer change other customers' agents?
No. A fix in your own account improves your agent, and it does not teach anyone else's.
What should I ask a vendor about training?
Ask whether customer content trains models shared with other customers, and whether sub-processors retain or train on it. Ask for the answer in writing, and treat a vague reply such as industry standard as a reason to keep asking.
Can I stop recordings from being kept?
Yes. You can switch recording off and keep only the structured result, or set a fixed retention period. To talk through your retention needs, contact us.
