AI assistant setup: answers from your sources, hands the rest to your team
An AI assistant is software that answers customer questions on a website or in WhatsApp using the business's own sources. ALTAI Digital builds AI assistants that speak only from the information you provide. When a question has no answer in those sources, the assistant does not invent one; it hands the conversation to your team with a summary. Customers are told clearly that they are talking to an AI assistant. Written conversations are kept and linked to the customer record.
Many business websites have a chat window with nobody behind it. Messages land in an inbox and are read the next day. Other sites run an off-the-shelf chat tool that answers every question with the same stock phrases. The customer asks about price, opening hours or availability, and the tool shows a menu or misses the point. The customer closes the window and picks up the phone, which nobody answers at the busiest hour.
An AI assistant built on a language model interprets the question and answers only from your sources. Those sources are your service list, pricing policy, opening hours, frequently asked questions and documents. If a question has no answer there, the assistant does not guess. It passes the conversation to someone on your team. That person sees the summary so far and the customer's record, so the customer does not have to start again.
The assistant's limits are written down at setup. A conversation goes to a person in three cases: a question the sources cannot answer, a customer who asks for a person, and topics you list. Complaints, price negotiation and contracts are on the default list. The greeting tells the customer plainly that they are talking to an AI assistant, and the option to reach a person stays open. Hiding it costs trust.
Below you will find the scope of an AI assistant build, the steps and the questions we hear most. Sector pages are listed below. For phone calls, see our voice AI assistant page; for a WhatsApp line, see WhatsApp automation. We work from Rotterdam and Istanbul, and conversation data is handled under GDPR. Pricing is given after the discovery call, once the channels and the volume of source material are clear.
What goes wrong today
- 01
Nobody behind the chat window
A message typed into the website chat lands in an inbox. Questions sent after hours wait until the next day. The customer writes to a competitor in the meantime, and the window becomes decoration.
- 02
The off-the-shelf chat tool misses the question
A rule-based chat flow only responds to preset buttons. When a customer asks in their own words, it falls back to a menu. After a few tries the customer gives up and tries to call instead.
- 03
Fear of wrong facts
A general-purpose model tries to answer everything. A chat tool with no sources that quotes the wrong price, hours or terms puts the business in a difficult spot. That worry keeps many businesses from switching an AI assistant on at all.
- 04
Customers repeat themselves after handover
An off-the-shelf chat tool passes an unsolved question to the team, but not the conversation. The team member asks the customer what they need. Explaining it twice wears the customer's patience thin.
- 05
The same questions fill the team's day
Opening hours, address, price range and availability come in all day. Each takes a few minutes and breaks up real work. The questions repeat, yet nobody collects them in one place.
- 06
Conversations are not recorded
Website chats are not linked to the customer record. Nobody knows which questions come up most or which chats turned into sales. When the same customer writes again a week later, the earlier chat cannot be found.
Scope of the build
Answers only from your sources
The assistant is given your service list, pricing policy, hours, FAQs and documents. Answers come from those sources. When there is no match, the assistant does not invent an answer; it hands over to a person. When a source is updated, the answers change the same day.
Handover to a person in three cases
A conversation goes to the team for an unanswerable question, at the customer's request, or when a listed topic comes up. Complaints, price negotiation and contracts are on the default list. We write the list with you at setup.
Handover with a summary
The person taking over sees the summary so far and the customer's record. The customer does not repeat themselves. The record also shows who took the conversation over.
Same AI assistant on the website and WhatsApp
The assistant runs on the website chat and the WhatsApp line from the same sources. There are not two knowledge bases to maintain. On WhatsApp, Meta's twenty-four-hour rule and approved templates apply.
A plain greeting and a way to a person
The greeting tells customers they are talking to an AI assistant. The option to reach a person is open in every message. You write the greeting or approve ours.
Conversation records linked to customers
Written conversations are stored and linked to the customer record. When a customer writes again, the earlier chat is there. Conversations are not used to train models, and storage follows GDPR.
Monthly question and handover report
Each month we report the most common questions and how many chats the assistant answered or handed over. Gaps in the sources are listed and filled, and the handover count is tracked month by month.
How the build runs
- 01
Discovery (1 week)
We map the questions arriving on each channel, gather source documents and write the handover list and greeting with you.
- 02
Sources and setup
We organise the sources, connect the assistant to the website and WhatsApp and route handovers to your team's screen.
- 03
Trial week
The assistant runs with your team first. Wrong or missing answers go back into the sources and handover rules are adjusted.
- 04
Launch and measure
The assistant opens to customers. After the first month we report questions, answers and handovers. The first live flow opens in four to eight weeks.
An AI assistant page for your sector
Products used in this build
Chat and voice assistant
Questions on your website, on WhatsApp and on the phone are answered at once with your own information; anything without an answer comes to you.
Request handling
Every request from WhatsApp, your web form and the phone gets an answer straight away and lands in one list.
WhatsApp Business integration
Questions on WhatsApp are answered with Meta's official AI feature and you step in whenever you like. Every chat builds up in the business account.
Terms used on this page
- AI assistant
- Software that reads a written customer question and answers from the business's own sources, passing anything else to a person.
- Source material
- The service list, pricing policy, opening hours, FAQs and documents an AI assistant relies on when it answers.
- Handover
- The assistant passing a conversation, with its summary, to a person on the team, so the customer does not repeat themselves.
- Handover list
- The written list of topics that always go to a person; complaints, price negotiation and contracts are on it by default.
- Greeting
- The opening message that tells customers they are talking to an AI assistant and shows how to reach a person.
AI assistant setup Frequently asked questions
The assistant speaks only from the sources you provide. When a question has no answer there, it hands the conversation to a person instead of inventing one. Gaps show up in the monthly report so the sources can be completed.
In three cases: a question the sources cannot answer, a customer who asks for a person, and topics you list. Complaints, price negotiation and contracts are on the default list.
No. The person taking over sees the summary so far and the customer's record, and picks up where the assistant stopped.
Yes. The greeting says so, and the option to reach a person is always open. Being open about it also helps customers ask the assistant the right questions.
Yes. It uses the same sources on both channels, so there is one knowledge base to maintain. On WhatsApp, Meta's twenty-four-hour rule and approved templates apply.
Written conversations are stored and linked to your customer records. They are not used to train models. Where personal data is kept is set out at setup, in line with GDPR.
The sources are your documents and information. When prices, hours or services change, the sources are updated and the assistant answers with the new information the same day. Updates can also be part of a monthly maintenance plan.
The first live flow opens in four to eight weeks. The timing depends on the number of channels, the volume of source material and the systems the handover connects to. Your own timeline is set out in the discovery report.
Terms on this page
- AI assistantAI assistant is the name for software that answers questions in writing on a website or a messaging channel from the sources it was given. A rule-based chat flow offers a written route and a list of options. An AI assistant reads a free-text question and answers from those sources. When a question has no answer in them, a well-built assistant hands the conversation to a person.
- AI voice assistantAI voice assistants answer the phone with a synthetic voice and turn the conversation into text. The greeting is written by the business. The assistant speaks only from the sources it was given. Where a question has no answer in those sources, it makes nothing up. The call goes to a person, along with a written summary.
- IVRIVR (interactive voice response) is a phone system that answers callers with recorded or synthetic speech and routes them by keypad or voice. The caller picks an option from a menu, and the system sends the call to a team, a queue or a recorded message. The menu is written in advance, and it recognises only the options it was given.
We write the proposal after we have found the problem together
In a thirty-minute call we listen to how you work and talk through where time and customers are slipping away. If you do not need a digital fix, we will tell you that too.
30 minutes · no preparation needed
The call and the week of discovery are both free.Read the terms
