Recognition, synthesis and the language model run on our own servers, so conversations do not depend on third-party services. This page sets out where data is processed, how long we keep it, what we can evidence, and which obligations stay with you.
Most voice-agent vendors rent recognition, the model and synthesis from third-party APIs. Every rented layer is another company holding your customer's voice. Ours are in-house.
Speech recognition, the language model and speech synthesis all run on servers we operate. Your call data stays in-region and does not pass to third-party speech providers to be processed.
A shorter processing chain is easier to describe to your legal team, because each additional processor is another agreement, another location and another breach surface to account for.
We can connect other providers where a project needs it. When we do, we tell you which layer changed and where that provider processes data, before it goes live.
Benerra is registered in Dubai, United Arab Emirates. That makes the transfer question a real one, so here is the honest structure rather than a badge.
Website form submissions and server logs are processed on our own infrastructure in the EU. The site itself is served over HTTPS.
Where staff at our Dubai headquarters access data belonging to EU residents, that transfer is covered by appropriate safeguards, including the European Commission's standard contractual clauses.
Because we are established outside the EU, we are supported by an appointed representative in the European Union under Article 27 GDPR. EU residents can address requests to them or to us.
A short chain, named safeguards, and a contactable representative. No badge does that work for you.
Contact and demo requests, counted from our last exchange with you, so we can follow up on an enquiry.
Web server logs: requested pages and timestamps, kept to keep the site secure and working.
Two cookies only. One remembers your cookie choice, one stores your pricing region. Nothing else is set.
We do not sell personal data, and we do not use it for third-party advertising. There is no exception to this.
Regional pricing uses a one-time country lookup from your IP address, only after you accept optional cookies. We store the resulting region, not your IP address.
Read the full privacy policyQuality assurance and compliance evidence are the same problem: you cannot attest to what nobody reviewed. A manual QA team reaches 3–5% of calls, which is our own figure from running live call operations.
Locator, our speech analytics, transcribes and scores every call against the checklist you agree with us: whether the script was followed, which objections came up, what was promised. When someone asks what the agent said on a specific call, there is a record and a score, not a sample.
A complete set of scored conversations, on the same criteria, across the whole period under review. That is the difference between asserting your calls are compliant and being able to show it on any call someone picks.
The fastest way to stall an AI calling project is to build the agent first and discover the legal basis for the list afterwards. This is the honest split, and none of it is legal advice: consult your own counsel for your jurisdiction.
You own the lawful basis for contacting the people on it, and the record of how that basis was obtained. This is the item that most often needs work before launch, and it is not something a vendor can supply for you.
Consent regimes, do-not-call registers, recording notices and rules on telling the person they are speaking to an automated system all differ by country, and collections calling has its own constraints. Your counsel decides what applies to you.
We configure the scenario to match those decisions: how the agent introduces itself, what it must say and must not say, when it stops and hands to a person, and what gets logged for every call.
Decide the rules first, then build the agent around them. That order is the whole difference.
