The same questions, fifty times a day
Opening hours, pricing, availability, delivery times, what to bring. Answered by people who should be selling or treating patients.
An hour a day of repeated answers is more than 250 hours a year, per person.
Premium solutions · applied AI
AI assistants that do the work, not just the talking
Trained on your data, connected to your systems, and allowed to take real actions — book, quote, route, escalate — within limits you set.
A chatbot that answers questions is a support widget. An assistant that reads a request, checks your calendar, books the slot and tells the right person is a member of staff. We build the second kind.
We reply the same working day. No obligation.
12 years
building business systems
120+
projects delivered
6–12 weeks
typical build time
3+ years
average client relationship
Situations
Three signs an assistant would pay for itself.
Enquiries arrive out of hours and wait
Someone asks at nine in the evening and hears back at ten the next morning, by which time they have asked someone else.
Requests are routed by whoever opens the inbox
A technical question reaches sales, an urgent one waits behind a routine one, and nothing is tracked.
Benefits
Trained on your data
Your services, pricing, policies and documents. Not a generic model guessing.
Connected to your systems
It reads real availability and real stock, then acts on them.
Clear limits and escalation
What it may do alone and what goes to a person is defined and visible.
Measured in resolved requests
Not in conversations. What matters is how many needed no human at all.
What you get
What an AI assistant system includes.
Not a chat window. An assistant with access, permissions and accountability, like any other member of the team.
Assistant types
What we build most often.
Booking assistant
Interprets the request, checks real availability, books the slot and sends the confirmation.
Sales qualification assistant
Handles first contact, asks the qualifying questions, and hands a prepared summary to the salesperson.
Customer support assistant
Answers from your documentation, checks order status, and escalates with context when it cannot resolve.
Internal knowledge assistant
Your team asks it about procedures, contracts or specifications instead of asking a colleague.
Document processing assistant
Reads invoices, orders or contracts, extracts the data and files it in the right system.
Triage assistant
Reads incoming requests across channels and routes each to the right person, with urgency flagged.
Process
Five steps, starting from real conversations.
We read your actual inbox
Three months of real questions tell us what the assistant must handle. Assumptions do not.
Scope and limits
What it answers, what it may do, what it must escalate. Written and agreed before anything is built.
Build and connect
Knowledge base, live data connections and the defined actions, with permissions on each.
Supervised trial
It runs with every answer reviewed by a person, and the corrections improve it before it goes public.
Launch and continuous tuning
We watch escalations and unanswered topics, and close the gaps month by month.
How we build
The limits matter more than the model.
Choosing a model is the easy part. The engineering is in what the assistant is allowed to do, what it must never claim, and what happens when it is unsure.
Premium solution
An assistant with permissions
We treat an assistant like a member of staff: it gets access to specific systems, permission for specific actions, and a clear instruction on when to ask someone. It answers from your material rather than from general knowledge, and when it does not know, it says so and hands over — with the conversation summarised.
- It says when it does not knowThe most important behaviour we build. An assistant that invents an answer costs more than one that escalates.
- Answers from live dataAvailability, stock and order status read at the moment of asking, not from a stale copy.
- Actions, not just repliesBooking, quoting, order lookup and routing, each behind its own permission.
- Every conversation reviewableYour team can read, correct and improve. Corrections feed back into the knowledge base.
- Your data stays under your controlWe agree in writing what leaves your systems. Sensitive material can be processed without being used for training.
- Cost per conversation is visibleYou see what it costs to run, so extending it is a business decision rather than a guess.
For Cabinero we built an assistant that interprets the patient request, routes it to the correct specialty and books the slot in the calendar — no phone call, no operator, and escalation to a person whenever the request is ambiguous.
Worth knowing: an assistant is only as good as the material behind it. If your pricing, policies and procedures are not written down anywhere, that documentation work comes first — and we will tell you before you commit.
Document processing
Invoices, orders and contracts arriving in different formats, with the data extracted, validated and filed automatically. Exceptions flagged rather than guessed.
Suited when tens or hundreds of documents a month are keyed in by hand.
Internal knowledge assistant
Procedures, specifications, contracts and past decisions, searchable in plain language, so people stop interrupting the person who remembers.
Suited to companies where institutional knowledge sits with two or three people.
Qualification and routing
First contact handled, qualifying questions asked, urgency detected, and a prepared summary passed to the right person.
Suited when enquiry volume exceeds what the team can triage carefully.
| Basic chatbot | Platform assistant | Custom assistant | |
|---|---|---|---|
| Answers from your data | scripted only | partly | fully |
| Reads live availability or stock | no | rarely | yes |
| Takes real actions | no | limited | yes, with permissions |
| Escalates with context | no | basic | with full summary |
| Improves from corrections | no | limited | yes |
| You control the data policy | no | no | yes |
Our rule: If your questions are few and always the same, a scripted chatbot is cheaper and we will say so. A custom assistant earns its cost when requests vary, when answers depend on live data, or when the assistant needs to actually do something rather than explain it.
Measurement
The number that matters is requests resolved without a human.
Conversation counts flatter everyone and mean nothing. We agree the target before building and verify it after.
A concrete example: the Cabinero assistant completes the whole path from request to booked appointment, which previously required a phone call answered by a person during opening hours.
What is included
What an assistant system includes
Projects
View full portfolioWe do not promise. We demonstrate.
AI systems delivered and running in production.
Frequently asked questions
How is the price set?
We work on an hourly rate: €50 per hour, excluding VAT. The price comes from the number of estimated hours rather than from a fixed package. You receive a proposal in which each module carries its estimated hours and cost.
Is this an off-the-shelf product?
No. This is a system built around how your organisation already works. We start from your workflow, your rules and your existing software, then build what fits. That is why it costs more than a subscription tool — and why people actually use it.
How long does it take to build?
A focused assistant takes 6–10 weeks. One connected to several systems and taking multiple actions takes 8–12 weeks. We always run a supervised trial before it goes public, and that period is part of the timeline.
Will it make things up?
That is the risk we engineer against. It answers from your material rather than from general knowledge, and when the material does not cover something it says so and escalates. We test that behaviour deliberately during the supervised trial, including with awkward questions.
What happens to our company data?
We agree in writing what leaves your systems and what does not. For sensitive material we use processing arrangements that exclude your data from model training, or local processing where the case requires it. The policy goes into the project documentation.
Can it actually do things, or only answer?
It can act: book an appointment, look up an order, generate a quote, create a ticket, route a request. Each action sits behind its own permission, and anything above the thresholds you set goes to a person for approval.
What if it gets something wrong?
Every conversation is reviewable by your team. Corrections feed back into the knowledge base, so the same mistake does not recur. Thresholds decide what runs alone and what needs approval, and we tighten them wherever the cost of an error is high.
Which channels does it work on?
Website, WhatsApp, email and internal chat, sharing the same knowledge base and the same limits. We usually start with one channel, prove it, then extend.
Do we need our documentation in order first?
Largely, yes. The assistant is only as good as the material behind it. If pricing, policies and procedures are not written down, we do that work first — and we tell you before you commit, rather than discovering it halfway through.
Estimate
Tell us what you need. We reply with an estimate.
Five steps, under two minutes. The more context you give us, the closer the estimate will be to reality.
ProjectStep 1 din 5
What would you like us to build?
You can pick more than one.
Tell us what your team keeps answering.
You do not need a specification. Send us the questions that arrive most often and what currently happens to them. We reply with what an assistant could handle, what it should not, and the estimated hours.
Get an estimateOr email us directly: contact@divasweb.ro+40 755 336 514