AI assistants and workflow automation.
Assistants that answer from your own documents and systems, document processing that turns paper and PDFs into data, and automations that move work between your tools — built with a person in the loop where decisions matter.
Customer-facing assistants
- Answers from your product information, policies and FAQs
- Hand-over to a person with the conversation attached
- Available on website, messaging channels and e-mail
- Every conversation logged and reviewable
Internal knowledge assistants
- Manuals, contracts, past projects, HR policies
- Access-controlled by role: the assistant sees only what the user may see
- Answers with citations to the source document
Document processing
- Invoices, delivery notes, tenders, applications, contracts
- Extracted into structured data with validation
- Written into ERP, CRM or accounting after review
E-mail and ticket handling
- Classification and routing
- Suggested replies for staff to edit and send
- Summaries of long threads
Content and language
- Product descriptions and translations for EU markets
- Proposal and report drafts from templates and data
- Always reviewed by a person before it leaves the company
Workflow automation
- n8n, Power Automate, Zapier or custom services
- Triggers from forms, e-mail and systems
- AI steps where judgment is needed; rules everywhere else
Retrieval instead of training
The model reads the relevant documents at answer time (retrieval-augmented generation). Nothing is trained on your data; updating a document updates the answers.
Permissions respected
Sources are indexed with their access rights. A user gets answers only from documents they could open themselves.
Guardrails
Scope limits, defined behaviour for questions outside scope, escalation to a person, and no actions in your systems without an explicit rule or confirmation.
Evaluation
A test set of real questions with expected answers. Quality is measured before launch, after every change and every month.
Logging and review
Every answer traceable to its sources; a sample reviewed by your team; feedback fed back into the evaluation set.
Transparency
Users are told they are interacting with an AI system, as the EU AI Act requires; the assistant does not pretend to be a person.
Define week 1
One process, one metric, the data sources, the people involved and what “good” looks like.
Build weeks 2–5
Sources connected, assistant or automation built, evaluation set created, guardrails set.
Test with users weeks 5–7
Real users, real cases, weekly adjustments; quality and time measured against the baseline.
Decide week 8
Numbers on the table: go, adjust or stop. If go, a roll-out plan and the monthly operation.
- Design and scope with the people doing the work.
- Connection to sources documents, systems, channels.
- Evaluation set and report quality measured, not asserted.
- Hosting and monitoring in the EU for the pilot period.
- Training for the users and the reviewers.
- Decision document with numbers, cost of operation and roll-out plan.
Which model do you use?
The one that fits the task and the rules: European-hosted options from the major providers, or open-weight models on EU infrastructure where data may not leave your environment. The choice and the reasons are documented.
Can it take actions — create tickets, send e-mails?
Yes, with explicit rules and confirmation steps. In a pilot we usually start with reading and drafting, and add actions once quality is proven.
Will it make things up?
Assistants built on retrieval, with citations and scope limits, answer from your sources or say that they cannot. The evaluation measures how often each happens, and the numbers are in the report.
What does operation cost afterwards?
Usage-based model costs, hosting, and a monthly retainer for monitoring, quality review and changes. The decision document states the figures for your case.
AI solutions catalogue
Twelve solutions working in companies of your size: customer-service and knowledge assistants, invoice processing, inbox triage, sales drafts, reporting — with pilot times and what you get.
Read more →AI for businessAI integration and operations
Bringing models into CRM, ERP, helpdesk and your own software — with data protection, evaluation and cost control.
Read more →AI for businessAI readiness and use cases
A 2–4 week assessment of processes, data and rules that ends in a ranked list of use cases and a pilot plan.
Read more →Which task would you hand to an assistant first?
Describe it in a few sentences. We reply within one business day with whether it is a good first pilot and what it would cost.