Area 14 · Data, AI and consulting

Artificial intelligence

In most companies AI already came in through the side door: one person drafts customer replies with an assistant, another pasted a supplier contract into it. The useful question is no longer whether the tool impresses. It is which of your repetitive tasks it genuinely lightens, what becomes of the data you hand over, and what the monthly bill adds up to. We start from those figures; picking a model comes later.

4
services gathered in this area
1 process
per pilot, the rest wait their turn
EU
hosting region for sensitive material
EUR 95
per hour excl. VAT, or a fixed price

Where the gain shows, and where there is none

A good candidate has three traits: the task returns dozens of times a week, a single mistake costs little, and the output passes a human eye anyway before it leaves the company. When one of the three is missing, we say so before invoicing anything.

Shared mailbox and complaints

A person approves the routing. The message is read, classed as a complaint, an order or a billing query, then offered to the right department with a draft reply to correct.

Product pages for a catalogue

Nothing publishes unread. From the technical specification a first description is drafted for the shop, then reshaped for a marketplace that imposes its own template.

Assistant on your procedures

The answer cites its source. A new joiner asks about the leave policy or how to handle a customer return; the assistant returns the exact passage from the SharePoint document and its link.

Contracts sent by your clients

The decision stays with the lawyer. The incoming draft is compared with your usual template and the gaps are highlighted: penalties, payment terms, liability clauses, length of commitment.

Minutes of meetings

A participant reads before sending. The recording of a Teams meeting yields a list of decisions and dated actions, ready to paste into Planner or your own tracker.

Your customers' data has no business inside a consumer service opened on somebody's personal account. Before any trial we establish with you what may travel to an outside vendor and what stays inside your Microsoft 365 tenant or with a host in a European region. As soon as personal data appears you need a processing agreement, a line in the record of processing activities and, depending on the case, an impact assessment. Those decisions are written down and dated for your DPO; that is what makes the practice defensible to the CNIL, not the quality of the model you picked.

How a first project runs

A short pilot on real data before anything is generalised. Workshops, configuration and acceptance happen remotely, on Teams or Google Meet, Mon-Fri 8:00-18:00 CET.

01

One process, one measure

We put a figure on the time the task eats today and agree a success threshold together, for example eight messages in ten reaching the right department without correction.

02

Pilot inside real work

A handful of people use the tool on the day job for an agreed period. Every output passes human approval while we record accuracy and the minutes saved.

03

Decision taken on figures

The result is compared with the starting point. If the threshold is missed we tell you and the pilot ends there: a failed trial costs far less than a rollout nobody uses.

04

Rollout and usage rules

The tool opens to everyone concerned, together with a short usage policy, quality monitoring and the documentation your DPO will need.

Questions and answers

The European regulation on artificial intelligence does not target vendors alone: duties depend on how you use a system and on the role you play with it, and they arrive in stages. We help with the factual part, recording which tools are genuinely in use, for what purpose, on which data, and writing the internal rules. Deciding how your situation qualifies in law belongs to your lawyer or DPO, and we do not stand in for them.

That depends entirely on the contract and the configuration, not on the brand. We choose business offers whose terms exclude training on your content, we read those terms with you before the start because they change, and we select a European processing region. Where professional secrecy applies, or where health data hosting rules come into play, the model runs inside your own environment instead.

Not by tipping it at random into a free service, no. You need a legal basis, a written purpose, a processing agreement with the vendor, an entry in the record of processing activities and an impact assessment where the case calls for one. Often the simplest answer is to send less: an order number and a complaint reason are enough for the model, while the full identity stays inside your own system.

It happens, and it is a result like any other. Some tasks carry too many special cases, some rest on documents nobody ever structured, others occur twice a month. We would rather hand you a reasoned negative conclusion than extend a project out of habit. The money spent then stops at the pilot, and the report sets out what would have to change for the question to be worth asking again.

A pilot covers a single process and is quoted as a fixed price or at EUR 95/hour excl. VAT with a ceiling agreed in advance. Model consumption billed by the vendor comes on top: we estimate it before starting and track it during the trial. If you prefer a continuous arrangement, AI can also sit inside a Start, Business or Premium plan, while a dedicated team is priced on a custom quote.

Let us find the process worth the effort

Describe the task your teams repeat every day and the system it lives in. We will look at the data available and propose a pilot with a figure attached.

When we are around
Weekdays, 8:00 to 18:00 CET; answers land inside one working day
Talking it through
A call on Teams or Google Meet, whenever writing is not enough

We set strictly necessary cookies only: they keep the site running and remember the city you chose. Nothing here is used for advertising or tracking. More in our privacy policy.