The boring but important bit
Smart about businesses. Careful with people.
Cortex learns from what customers say and do, so it holds real conversations about real people. This page is how we look after that: what the AI sees, where it runs, what stops it making things up, and how someone is forgotten.
Personal details stay out of the AI
Before customer data reaches a language model, phone numbers and email addresses are stripped out. The model needs to know who the business is, what it does and where. It never needs to know how to reach a person, so it is never given that.
The one thing kept is the business’s public website address, because it is already public and says a lot about the business.
Our own models, on our own servers
- Call recordings are transcribed on our own servers. No audio is sent to an outside AI service.
- Cortex’s reasoning over customer records (briefs, answers, call analysis) runs on models we host ourselves, or inside Spotzer’s own cloud account.
- Where a feature needs a commercial AI service, for example asking ChatGPT or Gemini how they see a business, it sends public information about the business, not customer records. Asking them is the whole point of that feature.
Guardrails: no made-up facts
- Every number has a source. On our pages, hover a dotted number to see where it came from.
- Numbers an AI invents are removed before anyone sees the answer, and claims with no evidence behind them are dropped. Both are counted, so we can see when it happens.
- When something can’t be answered, Cortex says why instead of showing a blank or a guess.
- People stay in charge. Posts, sites and dossiers are drafts for a person to review before anything is sent.
Benchmarks never point at one business
Comparisons like “plumbers in Leeds get about 40 reviews” are built from groups of at least five businesses, and below that they are not shown. A figure about a trade or a town can’t be traced back to a single customer.
Anonymisation that keeps the learning
When a customer cancels everything and the partner asks for their data to be anonymised, Cortex removes anything that names or reaches a person: contact details, the people mentioned, email bodies and note text.
It keeps only what the conversations were about (the topics, the sentiment, the order history), so the models still learn why customers stay or leave without knowing who they were.
The right to be forgotten
- One operation erases a customer across Cortex.
- Their numbers and email addresses are blocked first, so a later call or email can’t quietly recreate them.
- Every table is re-counted afterwards. If anything is left, the erasure reports a failure rather than a success.
- Raw call recordings and transcripts are held by the call system rather than Cortex. They are listed for removal as part of the same request, never silently skipped.
Questions about any of this? Ask your Spotzer contact. We would rather explain it than have you wonder.