Shared mailbox and complaints
A colleague confirms the routing. Each incoming message is sorted: the model tells a complaint from an order or an invoice query, and suggests an owner plus a draft reply.
Start with whichever one fits. The usual order, though, runs like this: decide what is worth automating, get a working tool, give it a proper home, and make sure people know how to use it.
A shortlist of processes with estimated savings, a view on data quality and on risk. For each idea we flag AI Act questions to check with a lawyer, and the result becomes a roadmap for the coming year.
A real tool living where your staff already work: Outlook, Teams, a WooCommerce shop or Comarch ERP Optima. Think routing incoming mail, drafting replies or writing Allegro listings.
Models running in your own cloud account in an EU region, with a vector database and GPU spending kept on a leash. Open models too, including the Polish PLLuM, for data that must not reach an outside provider.
Live online workshops built on tasks the participants bring, not textbook examples. One module never changes: what stays out of the chat window, and a two-minute routine for spotting an invented answer.
A strong candidate ticks three boxes: the task repeats dozens of times a day, one slip is not a disaster, and a person looks at the output before it goes anywhere.
A colleague confirms the routing. Each incoming message is sorted: the model tells a complaint from an order or an invoice query, and suggests an owner plus a draft reply.
Someone reads it before it goes live. Product attributes turn into a Polish draft description, and into German or Czech ones when you also sell on Amazon.de or Kaufland.
Answers point to their source. A new starter asks about remote working rules or the returns procedure, and the assistant quotes the relevant SharePoint passage with a link.
A lawyer makes the call. The model sets the draft against your template and flags altered penalty clauses, payment terms and liability caps.
An attendee edits and sends them. A recorded Teams call becomes a list of decisions and dated actions, ready to drop into Planner or your CRM.
Company information has no business in random free chatbots. Before anything goes live, we agree with you what an outside model is allowed to see and what stays inside your Microsoft 365 tenant or in a European data centre. Where personal data is involved, we add a processing agreement, an entry in the record of processing activities and, if needed, a DPIA. All of this is written down, so you can show it to your data protection officer, an auditor or the Polish supervisory authority.
A small pilot on real data comes first, a company-wide rollout second. Workshops, configuration and sign-offs all happen remotely over Teams or Google Meet.
We measure how long the team spends on a repetitive job today and set a success threshold, for instance “80% of emails reach the right person without correction”.
A handful of people use the tool for an agreed period. A human checks every output while we record hit rate and time saved.
We compare the outcome with the baseline. If the threshold is missed, the pilot ends there and you hear it plainly. Stopping early is far cheaper than a rollout nobody benefits from.
The tool reaches every user, and the company gets a short AI usage policy together with training on handling data.
The AI Act places duties on more than just the makers of AI systems, but what applies depends on how AI is used, and the rules come in stages. We help you list which tools the business uses and for what, and draw up rules for working with them. Which obligations apply to your company is worth assessing with a lawyer.
That depends on the service terms and configuration. We choose business editions whose terms exclude training on your content, review those terms with you before launch since they can change, and pick EU regions. For especially sensitive material, medical records for example, the model runs in your own cloud account, or we use an open model that sends queries nowhere.
Sometimes. PLLuM and other models trained on Polish handle official language and inflection well, and they can run on infrastructure you control. Large commercial models can be stronger at other tasks, and the picture changes quickly. Rather than guess, we test two or three options on your own documents and compare the results side by side.
Liability is a question for your lawyer; in practice the model output is only a suggestion, which is why our projects keep a human sign-off at the end. We do not automate jobs where a single error costs a lot and could slip through unnoticed. While the pilot runs, we track the error rate and agree up front on a floor below which the tool will not go into use.
A pilot covers a single process and is priced as a fixed fee or at PLN 190/hour excl. VAT. Model usage comes on top at the provider’s rates, which we estimate for your volume before launch. The point is to find out cheaply whether the idea holds up, long before any spending on licences and integrations.
Describe the jobs your team repeats every single day. We will pick the strongest candidates and propose a pilot with a measurable goal.
Your enquiry has reached us
You will hear back within one working day, and if you have reported an outage that is holding up work, it goes to the front of the queue.
No match for that name. Try a different spelling or pick a bigger town nearby - all our support is delivered online, so your choice has no effect on the service.