AI Policy

Why this page exists

We install AI systems for other companies, so we publish how we use them ourselves. Clients ask, and increasingly their clients ask them.

Models and vendors

We build on the best available commercial models and change them as they change. We do not tie a client's system to a single vendor where a reasonable alternative exists, and we document which model handles which task.

Training on your data

We configure business and enterprise vendor settings so that client inputs are not used to train third-party models, and we record that configuration in the handover documentation.

Human in the loop

Every AI output that reaches a customer has a named human owner on the client side. Complaints, emergencies, pricing negotiations, clinical questions, legal questions and anything emotionally loaded are escalated to a person by design.

Disclosure

We disclose when a customer is talking to an automated assistant. Several of our markets require it. In the rest we do it anyway, because being caught concealing it costs more than disclosing ever does.

What we will not build

Employee surveillance through cameras or keystrokes. Automated hiring decisions. Fabricated reviews, testimonials or case results. Generated imagery presented as real completed work. Systems designed to conceal that they are automated.

Logging and audit

Automated actions that affect customers or money are logged with their inputs and approvals, so a decision can be reconstructed.

Regulation

Obligations differ sharply across our fifty cities. We map which systems fall into which category in each market and we tell clients plainly where they need qualified local legal advice rather than a template.