Faayez MohammadFounder and CEO, Nexvix
Published September 20268 min read
- Decide before build
- What leaves, to whom, where, for how long
- Design rule
- Minimise before it leaves
- Tell customers
- In one plain paragraph
One: what actually leaves your systems
For each step, list the fields sent outward. Most requests need far less than the whole record: an answer about delivery needs an order status, not a customer’s full history.
Minimising at this level is the single most effective privacy control available, and it usually makes the system cheaper and faster too.
Two: which providers see it
Get the providers named in writing (model, hosting, telephony, verification), with their region and retention policy. A supplier who cannot name them cannot tell you where your data goes.
“We use industry-standard AI” is not the name of a company.
Three: where the processing happens
If contracts or client policy require processing in a region, that constraint changes which providers are eligible and belongs in the design, not in a launch-week surprise.
Four: how long anything is kept
You should be able to state retention in one sentence per system: transcripts for X days, records for Y years, nothing used for training. Then check the provider settings actually match the sentence.
Five: what you tell customers
One plain paragraph in your privacy notice covering what the AI does, what it can see, and how to reach a human. Most notices omit this entirely, which is exactly why customers assume the worst.
- What the system is used for
- What data it can access
- Whether a person reviews output
- How to opt out or reach a human