An hourly rate that justifies the investment.

The digital transition of the law and accounting professions is not a topic we bring in from outside: in France the national body of chartered accountants publishes its data and AI work, relayed by the regional bodies; the higher council of the notariat has issued its legal guide to artificial intelligence; and the national bar council makes a comparison of legal AI tools available. In other words, your own institutions already frame how the profession works with these tools — our work sits inside that frame, never around it.

The economics, meanwhile, read straight off the fee structure. Every hour spent on traditional tasks — data entry, filing, archiving of documents — is an hour of advisory lost; and advisory is what your clients pay for, responsiveness what they judge you on. A system that absorbs this volume does not change your profession: it gives the firm back the billable hours that admin quietly erodes, with dashboards that show where the team's time actually goes.

We rarely start from zero: our inbound-request classifier has been calibrated and tested on exactly this kind of audience — nuanced requests where “urgent” does not mean the same thing depending on whether it comes from a long-standing client or a stranger. A typical first engagement starts there: the assessment maps how requests arrive today (shared inbox, switchboard, forms), then the sprint puts triage in place on a limited scope, with your staff in the validation loop through the first weeks.

Starting with intake is not an accident: it is the most visible process for the whole team, the most measurable — you count what is correctly classified — and the least intrusive on the content of files. Trust is built there, before going further. From that first sprint, the relationship continues as a retainer, at the pace you set.

Many vendors treat your ethical obligations as friction to route around. We take them the other way: they dictate the architecture. Minimisation first — the system passes the AI model only the fragment strictly required, pseudonymised where possible; never the whole file. Mapping next — every data flow is written down in the assessment: which information, to where, why, for how long, and which third-party models are used. Validation last — nothing goes to production until you have seen and approved that schema.

A dispute, a succession, a tax litigation are not “data” like any other, and we will never pretend otherwise. That is exactly why the first engagement stays clear of the core of files — and the later ones venture there only at the pace your professional ethics allow. Where the data lives and which models touch it is documented in the assessment and decided with you, not for you.