Trust and security

How we handle security and sensitive data

The answers we give in pre-sales, published here word for word.

Sensitive data in AI

With concrete controls: PII and confidential information filters so nothing sensitive leaves the perimeter, the option to deploy models in the client's own environments — on-premise or private cloud — classification by sensitivity, and granular monitoring of what the agents consume and produce. In digital asset solutions, key custody can sit with the client, with us, or in a hybrid arrangement.

NDAs and corporate compliance

Yes. We work regularly with banks and energy companies, where confidentiality, security and compliance are a requirement, and we adapt to each client's procurement, NDAs and audits. Building regulatory compliance matrices is part of what we deliver on projects of that kind.

AI governance

Before scaling we install strategy, measurement and governance: indicators, visibility and cost control, an approved tool set and granular usage monitoring. Adoption is organised into three environments: exploratory with synthetic data, self-managed by the business, and enterprise when criticality, sensitive data or deep integrations are involved. Adopting without measurement ends in cost without return, as happened with the RPA boom.

From pilot to production

Three practices: design starts from business value and integration with existing systems, not from the demo; the same team that designs builds and runs it, so the architecture is thought for production from day one; and operations exist as an explicit service, with support and evolution after go-live.

Results and expectations

We are deliberately honest about expectations: improvement targets are set as hypotheses and the first phase confirms or adjusts them. With generative AI, accuracy improves iteratively and a share of cases always goes to a person, which is why the architecture includes human review from the design stage. Results are measured against objective references agreed with the client.