AI and knowledge base
Four weeks to bring your documentation together into a single, versioned source of truth that an LLM can use, and to set the rules for AI use in the organisation.
- Duration
- 4 weeks
- Format
- Fixed scope
- Deliverable
- Structured, versioned corpus, rules for AI use, first automations
- Who it is for
- Organisations whose information contradicts itself from one channel to the next
The website says one thing, the brochure another, the listing on the neighbouring tourist office's site a third, and the person at the front desk answers from memory. Each of those versions was right one day. Nobody knows which one still is, and correcting an opening time means remembering the seven places it appears. When an AI tool joins the team, it reads those seven places and reproduces the contradiction with confidence.
The engagement fixes the problem at its source. All the information about your activity is gathered into a single base, written in full sentences, where each fact appears once and carries a date. The other channels become copies of it: when the base changes, they change. An assistant that relies on that base answers in your words, and only in yours.
The rules for AI use are written with you, not imposed. They say what the team may hand to an AI tool, what stays internal, which data never leaves, and who proofreads before a text goes out. They are short because they have to be read. The automations put into service at the end are chosen for what they save the team the following week, not for the demonstration.
What you keep at the end: the base, its history, the rules for AI use, the automations in working order and the upkeep procedure, all in your tools and in your name. The person trained keeps the base on their own; if the AI tool changes, the base stays and is reconnected. The price is set in the written note that follows the first call.
What you get
- An inventory of everything that holds information about your activity: website, brochures, listings, spreadsheets, email, social networks, business software, with, for each source, what it says more or differently than the others.
- A single knowledge base, organised by subject, written in full sentences and dated, where each fact exists only once. Opening hours, prices, conditions and descriptions are written there once and reused everywhere else.
- A change history: who changed what, when, and what the previous version was. A mistake is corrected in a minute and can always be traced.
- Two pages of rules for AI use: what the team may hand to it, what it never hands to it, which data leaves the organisation and which does not, and who proofreads before publication.
- Two or three automations built on that base: for example, answers to visitors' recurring questions, updates to your listings on partner websites, or a newsletter prepared from the month's news.
- A one-page upkeep procedure so that the base stays accurate without me: who updates it, how often, and how to check that a channel has not drifted.
How it runs
Week 1: collection and contradictions
Sources listed and read against each other. You receive the list of contradictions found between your channels, with, for each one, the version you decide to keep. You make the call, not the tool.
Weeks 2 and 3: the base
The base written and organised from the versions you kept, the change history set up, first tests with an assistant to check that it answers correctly from your texts and from nothing else.
Week 4: rules and automations
The rules for AI use written with you, the chosen automations put into service, a half-day of training for the person who will keep the base. You leave with a base the team is already using.
Guides linked to this offer
All guidesNo linked guide yet.
A digital problem to put on the table?
Thirty minutes on video. You set out the situation, I say what I understand of it and whether I can help.