What we build
- Website assistant: answers questions about services, hours, required documents and next steps, using your own pages and FAQ as its source.
- Inbox triage: reads incoming email to a shared address, labels it by type (new inquiry, existing client, billing, vendor) and routes it to the right person or folder.
- WhatsApp assistant: handles first replies and collects details before a team member takes over.
- Workflow automation: onboarding steps, document requests, approvals, reminders and recurring reports.
- Document processing: OCR to read scanned forms and PDFs, and template-based document generation such as engagement letters or checklists.
Rules first, model second
The useful part of an assistant is not the language model; it is the set of rules around it. Before we configure anything, we agree in writing which topics the assistant may answer, which fields it must collect for each request type, and which situations trigger a handoff. A home services company, for example, might let the assistant confirm service areas and collect an address and photos, but route any pricing dispute straight to the office manager.
| Request type | Assistant does | Hands off when |
|---|---|---|
| General question | Answers from approved content | No approved answer exists |
| New inquiry | Collects name, contact, service needed, timing | Details are complete, or the visitor asks for a person |
| Existing client | Confirms identity fields, logs the request | Anything account-specific or sensitive |
| Complaint or urgent issue | Acknowledges and records details | Immediately |
How a project runs
- You describe the requests you receive and who handles each type today.
- We map the flow and propose rules, fields and handoff points.
- BIT LOGIC DEV S.R.L. sends a written scope and quote; nothing is billed before you confirm.
- We configure the assistant or automation on a test channel and you try it with real example questions.
- We adjust the rules, then switch it on for live traffic and review early conversations with you.
What to expect, honestly
AI assistants make mistakes, especially with unusual questions or ambiguous wording. That is why we restrict them to approved content, keep a visible way to reach a person, and log conversations so your team can review them. They are not suitable for giving legal, medical, tax or financial advice, and we do not configure them to do so.
AI automation projects are scoped and quoted in writing; there is no public price. Read our guide on when an assistant should hand off to a human before you decide what to automate, then list the requests you want handled for a written scope.
Common questions
Which AI model do you use?
We choose the model per project based on the task, the data involved and cost, and state it in the written scope. What matters more for reliability is the content the assistant is allowed to use and the handoff rules around it.
Will the assistant see client data?
Only what is needed for the task. We agree which fields it collects, where conversations are stored and who can read them. Sensitive categories are routed to a person rather than processed by the assistant.
Can it book appointments or update our CRM?
Often yes, if your tools offer an API or integration. We confirm what is possible with your specific systems during scoping and include it in the written quote.