How do you find your company's knowledge with an AI assistant?
Your procedures, your past quotes and the right answers to customers already exist. The problem is finding them. An AI assistant connected to your documents answers your questions with its sources, on 2 conditions: up-to-date documents and respected access rights.
:quality(80))
"Ask the person who was here in 2019, they'll know."
Every SMB has its own version of that sentence. The procedure exists somewhere. So does the quote for a similar project. Someone already answered that customer question 6 months ago, in an email. But nobody knows where.
The knowledge is there. It's just hard to find. That's exactly the kind of problem an AI assistant handles well, as long as it's set up properly.
Why is information so hard to find in an SMB?
Because it's scattered: in emails, shared folders, software and people's heads. Each place has its own logic, and none of them talk to each other.
The procedure for a product return is in a Word document from 2021. The price of the last similar project is in a quote, somewhere in a customer file. The right answer about the warranty is in an email sent by someone who's on vacation.
And the most important part isn't written down anywhere: why things are done this way, what went wrong last time, which customer has special requirements.
The result: you ask someone. Again and again. The person who knows everything becomes the company's search engine, between 2 tasks that were really theirs.
How do we handle a return for Ville du Lac?
What does information you can't find cost?
More than you think: time, mistakes and dependence on a few people.
Time, first. For example, if each person on a team of 10 spends 15 minutes a day looking for information that already exists, that's 150 minutes a day for the team. A task nobody put on their schedule.
Mistakes, next. When you can't find the procedure, you improvise. When you can't find the last quote, you start from scratch and the price changes from one customer to the next. And the same customer question gets a slightly different answer depending on who takes it. Customers end up noticing.
Dependence, above all. When the knowledge is in one person's head, the process only works when they're there. The day they retire, they take with them years of small decisions that were never written down.
What is an AI assistant connected to your documents?
It's an assistant that answers your questions from your company's documents, and shows where each answer comes from. You ask the question in your own words. It searches your procedures, your quotes, your emails and your files, then writes the answer with its sources: the document, the page, the date.
The difference with a general assistant matters. Claude, ChatGPT, Copilot or Gemini, used as is, know a lot about the world. They don't know your return policy or the price of your last project. Connected to your documents, they do.
And when the information isn't in your documents, a good assistant says so instead of making something up. An answer without a source is a rumour. An answer with its source is information you can check.
How do we handle a product return for a business customer?
Returns are accepted within 30 days if the packaging is intact. For a business customer, we issue a credit note rather than a refund. Ville du Lac has a special agreement: Name approves it.
Can an AI assistant make mistakes?
Yes, and that's exactly why sources matter. It can quote an outdated procedure, mix up 2 customers with similar names or summarize a long document a little too quickly.
A good assistant is designed so these mistakes show. It shows its sources, it says when it finds nothing and it flags documents that contradict each other. The person reading the answer can check it in a few seconds.
For anything that commits the company, like a price, a clause or a sensitive reply to a customer, a person checks before acting. The assistant finds and prepares. People decide.
What is an AI assistant for in an SMB, concretely?
Answering faster, training faster and quoting more accurately. A few concrete uses:
For quotes, the assistant finds similar projects, what was charged and what went over. The person preparing the quote decides the price, but they start from a history rather than a blank page.
For onboarding, the new person dares to ask the assistant all their questions, even the ones they wouldn't ask a busy colleague 3 times. Experienced people keep their time for the real questions.
Examples
- Training a new person: they ask the assistant their questions before bothering someone
- Internal support: "how do we process a return?", "where is the vacation form?"
- Answers to customers: finding the already approved answer about the warranty or lead times
- Quotes: similar projects, their prices and what went over
- Decision support: what was decided last time, and why
What does it take to make it work?
2 things above all: up-to-date documents and respected access rights. The rest can be worked out.
An assistant is never better than the documents it reads. If it finds 3 versions of the same procedure, it can quote the wrong one. An up-to-date guide is worth more than one more assistant.
For access, the rule is simple: the assistant sees what the person asking the question is allowed to see. That's the principle behind the assistants we build, and the one we use ourselves: your access, nothing more.
Up-to-date documents
Every important document has a person responsible and a review date. Old versions are archived, not left next to the new ones.
The right access
Payroll, employee files and confidential agreements stay closed to the people who don't have access to them today. The assistant doesn't open any extra doors.
Visible sources
Every answer shows where it comes from. You check in one click, and you quickly spot a document that is no longer right.
A person who decides
When 2 documents contradict each other, the assistant flags it. A person decides, then fixes the document.
3-year agreement, prices revised every January, payment in 30 days.
This document isn't part of your access. To see it, check with management.
Do you need to document everything before you start?
No. Start with the questions that come up most often, not with a big documentation project.
Pick one area: customer service, quotes or onboarding new people. Gather the documents that answer those questions, remove the old versions and write down what only exists in people's heads. Often, a few well-chosen pages are enough for a first try. Then measure: how many questions the team asks the assistant, how often it finds the right answer and what people think of it.
For many SMBs, the business versions of Copilot, Gemini, ChatGPT or Claude, which already search some of your files, are a good first step. When answers need to come from several systems, with each person's access, we build the assistant for you and connect it to the tool your team already uses. A prototype takes a few days instead of a few months.
Bonus: documenting for the assistant is also documenting for humans. The person starting next Monday will thank you.
Frequently asked questions.
What is an AI assistant for internal documents?
It's an assistant that answers your team's questions from the company's documents: procedures, quotes, emails, files. Every answer shows its source, so it can be checked. When the information isn't there, it says so instead of making something up.
How is it different from ChatGPT or Copilot used as is?
Used as is, a general assistant knows a lot about the world, but not your return policy or your past quotes. Connected to your documents, with each person's access, it answers from what is true at your company.
Can the assistant see confidential documents?
Only if the person asking the question already has access to them. A properly configured assistant respects each person's rights: payroll, employee files or confidential agreements stay closed to everyone else.
Can an AI assistant make mistakes?
Yes. It can quote an outdated document or summarize a long text poorly. That's why it must show its sources, say when it finds nothing and let a person check anything that commits the company, like a price or a clause.
Do our documents need to be perfect to start?
No, but they need to be up to date in the area you choose. Start with the questions that come up most often, remove the old versions and write down what only exists in people's heads. An assistant is never better than the documents it reads.
How long does it take to try an assistant on our documents?
A prototype for a specific area takes a few days instead of a few months. It lets you see with your team whether the answers are good before going further.
Your company's knowledge is worth a lot. For now, it's scattered across emails, folders and the heads of a few people.
An assistant that answers with its sources puts it back in the hands of the whole team, without waiting for the right person to come back from vacation.
Read next
An idea in mind?
Let’s chat!
:quality(80))
AI and automationCan you put confidential data into ChatGPT?
AI and automationWill AI replace my employees?
AI and automationWhere should an SMB start with AI?