AI assistant in business: what it can do, what it costs and how to build it on your company’s knowledge

What an AI assistant is — and how it differs from a chatbot and an agent

These three words are often used interchangeably, although they mean different things. A chatbot is a program for conversation. It used to run on rigid scripts (“press 1 if…”); today it increasingly runs on a language model. An AI assistant is a chatbot with a defined role that knows the company’s documents, knows what tone to use and has a specific task, for example answering employees’ questions about leave procedures instead of sending them to HR or to a forty-page policy document.

An AI agent goes one step further. It doesn’t just answer, it acts on its own: it sends an email, creates a task in the CRM, updates a spreadsheet. We have devoted a separate article to agents, AI agent in marketing and PR. In most businesses it makes more sense to start with an assistant. The risk is lower, because the assistant suggests and a person decides. Where assistants and agents fit into the bigger plan is shown in our guide to AI in business.

Online, the terms “AI assistant” and “AI personal assistant” are also popular. They usually mean the same thing, sometimes an assistant with a female voice or a name. Round-ups along the lines of “best AI assistant 2025” date quickly, because tool features change every few months. For a business, the name doesn’t matter. What matters is what knowledge the assistant works on and who is responsible for it.

Types of AI assistants in business

An AI assistant for business can look very different, from a feature built into the office suite that every employee has to hand in their email and documents, to a chat window on the website where it is customers, not company staff, who talk to the assistant. Below are the four most common types. From the simplest to the most demanding.

Who it is for and where it works

An artificial intelligence assistant makes sense where people answer similar questions every day based on the same documents. Is HR explaining the remote working rules for the tenth time this month, while sales reps rummage through folders looking for the current product sheet? An assistant will take over this routine with ease.

• HR. Employee questions about policies and leave, or about benefits. We write more about what is and isn’t allowed in HR in the article on AI in HR and recruitment.
• Sales. Quickly looking up product specifications, drafting a reply to an enquiry, preparing for a meeting.
• Customer service. Suggested replies for agents or an assistant on the website. We compare both options in the article on AI in customer service and chatbots.
• Marketing and PR. An assistant that knows the brand’s tone of voice, key message and company glossary helps with first drafts of copy.
• The board and executive assistants. Meeting summaries, notes, sorting email.

What an AI assistant does well — and what it does not

An assistant is good at summarising and searching, organising and suggesting first drafts. It does worse where data is missing or the answer requires judgement. The biggest limitation? If the assistant can’t find the answer in the documents, it may make one up. So add the sentence “if it isn’t in the documents, say you don’t know” to its instructions and regularly check its answers on a random sample of conversations, ideally every month and always after changing the documents in the assistant’s knowledge base.

The technique whereby the assistant first retrieves passages from documents and only then answers is called RAG (retrieval-augmented generation). We explain it in plain language in the article on RAG and AI that works on company knowledge.

How to build an AI assistant step by step

You can launch your first assistant in a few weeks without developers if the company has a business plan for ChatGPT, Gemini, Claude or Copilot. The order of the steps matters more than the choice of tool, because even the best model won’t help if it gets out-of-date documents and a vague task, while an average model with good instructions and well-organised knowledge is surprisingly often enough.

1. STEP 01Choose one task

   Specific and measurable: “answers questions about the remote working policy”, not “helps employees”. Choose one where mistakes are easy to spot.

2. STEP 02Collect and organise the documents

   Only current versions, no duplicates. An old price list in the folder means an old price list in the answers.

3. STEP 03Write the role instructions

   Who the assistant is, who it works for, what tone it uses, what it doesn’t do and what it should do when it doesn’t know the answer. We have collected ready-made templates for such instructions in the guide prompts for marketing.

4. STEP 04Test it on 30 real questions

   Collect the questions people actually ask and rate each answer as correct, partly correct or wrong. Then improve the instructions and documents.

5. STEP 05Share it with a small group

   A few people for two or three weeks, with a simple way of reporting errors. Appoint an owner of the assistant who is responsible for keeping its knowledge up to date.

6. STEP 06Measure and decide

   Measure response time, the number of errors and the number of questions that still reach people. Only then decide whether to roll the assistant out to the whole department.

How much an AI assistant costs and what drives the budget

An internal assistant in a suite the company already has costs mainly time. That means preparing documents and testing, work that someone in the department has to do alongside their normal duties and that usually nobody writes into the project budget. A customer-facing website assistant is a separate project. The licence prices below come from official price lists (USD, excluding VAT, as of 24 September 2026).

An AI assistant on your company website and what to bear in mind

A website assistant talks to customers, so every mistake it makes is visible to the outside world. It promises a discount that doesn’t exist, or gives a delivery date plucked out of thin air, and the customer will cite it in an email, in a complaint, and sometimes in a public review that other customers will read. So give it a narrow scope and clear boundaries (“I don’t quote prices, I’ll put you through to a sales rep”). And an easy handover to a human.

An AI assistant in the WordPress editor, on the other hand, is a tool for the website editor, not for the customer. It helps write and edit content. The same rules apply here as with any AI-assisted text: a person checks it and is accountable. We show how to do this without losing quality in the article on copywriting and content marketing with AI.

When an assistant pays off and when it doesn’t yet

An AI assistant pays off when there are many questions, they are similar to one another, and the answers are in documents that someone keeps in order. It doesn’t pay off when the company’s knowledge lives in the heads of a few people and is written down nowhere, or when every question requires an individual judgement of the situation that no document can describe, because it depends on the customer’s history, the relationship and intuition.

How can you tell it’s time for an assistant? People in several departments answer the same questions, new employees spend weeks looking for information, customers write after hours about simple matters. There are three warning signs. Nobody wants to own the assistant. The documents are out of date. The company has no rules on what may be pasted into AI tools.

How to measure whether the assistant helps

Measure three things: how many questions the assistant handled without human involvement, how many answers were wrong (in a random sample checked every month) and how much time the team saved. The number of conversations alone says nothing. An assistant can hold hundreds of them and get every one wrong.

Emails to HR falling, or enquiries from the website rising? Before you declare the assistant a success, check whether a department that doesn’t have it yet is seeing a similar change, and look at data from the same weeks last year. We help companies that want to combine an assistant with the automation of sales and service processes as part of marketing and sales automation. It is also worth keeping in mind the wider context described in the article on internal communication. An assistant is part of it. Not a substitute.

What to avoid when setting up a company AI assistant

Read next