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.
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.
You can come in through any of the four. The cheapest sequence, though, stays the same: decide what deserves automating, run one tool on real work, give it a compliant place to execute, then teach your teams to use it without feeding it whatever comes to hand.
An inventory of repetitive tasks, an estimate of the time they consume and a ranking by likely gain and by difficulty. GDPR and AI Act questions are noted beside each idea, so your lawyer or DPO can settle them.
A tool that appears where your teams already work: in Outlook, in a Teams channel, on the product form of your PrestaShop store or as a draft entry in Sage. With a before and after measurement, not a demonstration.
A closed environment on your own account with OVHcloud, Scaleway, Clever Cloud or an equivalent host in a European region. GPU capacity rented by the hour, a vector store for your documents, a watched budget. We sell no hardware.
Online workshops built on the casework participants bring, not on textbook examples. Two points return in every session: what is never pasted into an assistant, and how to check an answer in two minutes.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
The tool opens to everyone concerned, together with a short usage policy, quality monitoring and the documentation your DPO will need.
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.
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.
Message received
An answer follows inside one working day. Report an outage that is stopping people working and it moves ahead of everything else.
Nothing here under that name. Check the spelling, or simply choose the nearest large city instead. Since every engagement runs remotely, whichever you pick changes nothing about what we do for you.