Guide
Searching company knowledge: answers from procedures, contracts and emails instead of asking a colleague
In a company of 10 to 250 people, knowledge lives in a dozen places. Procedures on the shared drive, terms in contracts, agreements in emails, and most of it in the heads of the two people who have been there longest. A new hire asks, an experienced colleague answers, every day. This is what a function that lets you ask the system looks like.
Where knowledge lives in a company
On the shared drive, in folders whose structure is ten years old. In email inboxes, in threads where a customer agreed something with a salesperson two years ago. In PDF quotes, in contracts, in manufacturers’ manuals, in meeting notes. And in heads: the production manager knows how to raise a claim with this supplier, the bookkeeper knows how to account for that type of contract.
The result is that questions circulate. A new hire asks a senior colleague, who stops working and answers. A salesperson searches a contract for half an hour or phones the office. A field technician messages a colleague because they cannot remember the procedure.
What searching company knowledge looks like
A box where a staff member types a question in plain words: “what is the notice period in our contract with company X”, “how do I raise a claim with supplier Y”, “what did we promise customer Z about December deliveries”. AI searches the documents that staff member has access to and answers in a few sentences, pointing to the document and the place the answer came from.
Pointing to the source is mandatory. The staff member clicks and sees the passage from the contract, email or procedure. An answer without a source is guessing, and I do not allow that in company knowledge.
The same box can sit in the panel the team works in, in a portal for field technicians, or in an email inbox the staff member writes a question to and gets an answer from.
Permissions
Everyone sees only what they could open themselves. If a salesperson has no access to the employment contracts folder today, the search will not show those files or use them in an answer. We take permissions from how the drive and the system are divided today, and set them before launch.
The same applies to personal data and trade secrets. Before we start we agree which sets of documents go in at all, and where AI runs: with an outside provider whose terms do not allow them to use your data for their purposes, which I check before we start, or on a server we choose together.
Three cases where it works
A new hire. The first weeks bring hundreds of questions about how things are done in this company. Instead of interrupting colleagues, they ask the system and get the procedure with a pointer to where it lives. Experienced staff answer only the questions that have no document.
A salesperson looking for contract terms. A customer phones and asks about a discount, payment terms or the scope of service. The salesperson asks the system and has the answer from the contract in seconds, instead of promising to call back.
Service looking for a procedure. A technician at a machine on a customer’s site asks about the procedure for replacing a part, or what was agreed at the previous visit. The answer comes to their phone, from the manufacturer’s manual and the request history.
What it will not replace
The search answers from what is written down. Knowledge that exists only in the manager’s head will not appear in answers until someone writes it down. At launch it usually becomes clear which procedures are missing, and that is a good moment to write them.
It will not replace decisions either. The system says what the contract says, but it will not decide whether to grant a customer an exception. And it will not replace order in your documents, though it shows where that is missing, because you can see two contradictory answers to the same question from two files.
How we start
We pick one set of documents, the one staff ask about most often. Usually that is procedures or customer contracts. We set permissions and the place AI runs. I build the search as part of a module, meaning a self-contained part of the system the team already works in. After a month we look at how many questions the system handled and how many answers staff rated as accurate. Then we add further sets.
You pay after accepting a working function. Where AI makes sense in a company at all is on the main page of the guide. The other pages of the guide cover reading invoices and documents and answering customer questions.
Describe in the form below what staff most often ask each other and where the documents live. I will get back to you.
Questions and answers
Will every staff member have access to every document?
No. The search shows a staff member only the documents they could open themselves. A salesperson will not find employment contracts, and the warehouse will not see terms from a customer contract if they cannot see them today. We set permissions before launch, not after the first leak.
What if the documents are out of date or contradict each other?
They do in every company. The search shows the source and its date, so you can see that an answer comes from a procedure written three years ago. At launch it usually becomes clear which documents need updating or throwing out, and that is a tidy-up the company did not have before.
Do I have to tidy the shared drive first?
No. We start with one set of documents as it is. The tidying happens along the way, once you see what the system finds and what it does not. Sorting out the drive by hand before launch is weeks of work you can skip.
Contact
Describe what staff most often ask each other and I will tell you whether it can be searched
You do not need a specification. Describe how the work looks today and what should change. We will write the scope together.