Custom AI assistants
AI assistants tailoredto each role.
A custom AI assistant is built for a single role in the company, such as quoting, in-house legal or technical support. Unlike a general chat, it knows that role's rules, terms and limits, and prepares the routine work for the specialist, who makes the decision.
A generic chat does not know how your role works
General AI tools answer anything, but they do not know what discount you are allowed to give, which clause is unacceptable to the company or which check comes next in a diagnosis. The output has to be rewritten every time, and there is a risk that someone uses it as it is.
- Each specialist writes their own instructions, and quality depends on who is using the tool
- The role's rules — thresholds, approvals, forbidden wording — are not applied automatically
- The tone and structure of documents differ from the company standard
- There is no clear line between what the assistant can prepare and what a person must decide
Who it is for
- Quoting teams that regularly prepare quotes using the same pricing rules and terms
- In-house legal departments that review contracts against their own set of accepted clauses
- Technical support or service teams that diagnose problems using manuals and case history
- Managers who want a specialist's experience to be available to newer colleagues too, under that specialist's supervision
What a custom AI assistant does, by role
- 01
Draft quote
For quoting: suggests the structure and text of the quote from the catalogue and pricing rules, flagging what needs approval.
- 02
Clause review
For in-house legal: compares a contract with the list of accepted clauses and flags deviations, without giving a legal opinion.
- 03
Diagnostic steps
For technical support: suggests the next questions and checks based on manuals and resolved cases.
- 04
The company's tone and format
Writes in the style and structure set by the team, with approved wording and wording to avoid.
- 05
Escalation rules
Recognises situations outside its remit and passes them to the designated person.
How it works
How a role-based assistant works
The steps marked in yellow are approved by a person.
- Input
- Action
- AI
- Human in the loop
- Result
- 01
Input
The specialist's request
Your colleague describes the task: a quote, a contract to review, a technical problem.
- 02
Action
Rules and sources
The assistant applies the role's instructions and consults only the sources approved for it.
- 03
AI
Draft in the agreed format
The model prepares the draft or analysis in the agreed structure and tone.
- 04
Action
Flagging exceptions
Points that exceed the role's thresholds or rules are highlighted separately.
- 05
Human in the loop
The specialist's decision
The specialist corrects, approves or rejects; nothing goes out without them.
- 06
Result
Final document
The approved version is saved, and the corrections help fine-tune the assistant.
What information a role-based assistant receives
- A description of the role: what it does, what it does not do and who it escalates to
- The role's written rules: price thresholds, accepted clauses, diagnostic procedures
- Approved template documents: quotes, contracts, service reports
- The role's sources: catalogue, manuals, history of resolved cases
- Examples of good work and common mistakes, used for testing
Possible integrations
- Microsoft 365
- Google Workspace
- Microsoft Teams
- Slack
- Microsoft SharePoint
- HubSpot
- Odoo
- Jira Service Management
- Microsoft 365
- Google Workspace
- Microsoft Teams
- Slack
- Microsoft SharePoint
- HubSpot
- Odoo
- Jira Service Management
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What stays a human decision
- Any quote, price or discount sent to the customer
- Any legal conclusion — the assistant flags deviations, it does not give legal advice
- Technical interventions that may affect the safety of equipment or people
- Changing the role's rules and instructions, done only by the designated owner
Limitations and how we handle them
- The assistant applies the written rules it is given, which is why the first step is turning the specialist's experience into clear instructions
- It stays up to date through its sources: when prices, legislation or products change, updating the sources keeps it accurate
- It may miss a deviation or flag one that does not exist, which is why the specialist's review remains mandatory
- An assistant built for one role is not suited to questions outside that role
How we measure a role-based assistant's quality
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Volume of corrections
- How much the specialist edits the drafts — tracked on a sample of real work.
- Compliance with the role's rules
- Tests with edge cases (exceeded thresholds, forbidden clauses), run before every change.
- Correct escalations
- Whether situations outside its remit reach the right person, checked on real cases.
- Specialist feedback
- Simple ratings on each suggestion and observations discussed regularly with the team.
How we build a custom AI assistant
- 01
Observing the role
We watch how the specialist works, which rules they apply and where the time goes.
- 02
Writing the rules
We turn experience into clear instructions, examples and limits, validated by the specialist.
- 03
Build and test
We configure the assistant and test it on real work, including difficult cases.
- 04
Pilot with the specialist
The specialist uses the assistant for a while, and the rules are adjusted based on their corrections.
Use cases
These scenarios are conceptual examples showing what can be built. They are not delivered implementations or results achieved for clients.
- PVC window and door manufacturer
Quoting assistant
Based on the requested dimensions and options, the assistant suggests the quote using prices from the current list and flags discounts that need approval.
Conceptual example - Group of companies with in-house legal
Supply contract assistant
Incoming contracts are compared with the clauses the company accepts; the lawyer receives a list of deviations and decides what to negotiate.
Conceptual example - HVAC equipment service company
Assistant for technicians
The technician describes the symptoms, and the assistant suggests the next checks from the manuals and previous service visits.
Conceptual example
Frequently asked questions
How is it different from a knowledge base assistant?
A knowledge base assistant answers any employee's questions about procedures. A custom assistant works for a single role and prepares that role's work — quotes, contract analyses, diagnostics — according to its rules.
Does the assistant replace the specialist?
No. It takes on the repetitive part — the first draft, the checklist review, searching the manuals — while the specialist decides and is accountable for the result.
Can the legal assistant give a legal opinion?
No. It flags differences from the clauses set by the company and explains why it flagged them; interpretation and the legal opinion remain with the lawyer.
What does the assistant do when it does not have enough information?
It says what is missing, asks for the details it needs or passes the case to the designated person, instead of filling in the gaps itself.
How are the role's documents protected?
The assistant only has access to the sources approved for that role and can only be used by that role's users. The model provider and settings are chosen according to the level of confidentiality, and the processing of personal data is assessed in line with the GDPR.
Can we have several assistants for different roles?
Yes. Each has its own rules, sources and access rights, even if they share the same technical infrastructure.
How is it different from a general AI chat set up by each colleague?
A general chat depends on the instructions each user writes. A custom assistant has rules, sources and access rights set at company level, tested on real work and adjusted together with the specialist, so that results are consistent across colleagues.
Can’t find the answer? Write to us and we’ll reply directly.
Want an assistant that knows your team's rules?
Pick a role where routine work takes up a lot of time. We will show you what the assistant would look like, which rules it gets and where the specialist decides.
Related solutions
- RAG systemsAssistants that search the configured sources before answering and show where each piece of information comes from.
- AI assistant for your internal knowledge baseEmployees ask how something is done in the company and get the answer from internal procedures and policies, with a reference to the source document.
- AI agent for salesPrepares quotes, follow-ups and replies from your CRM data, while the salesperson reviews them and decides what goes to the client.