An AI answers.
How do you know it answers correctly?
The problem
An AI can produce an extremely convincing answer that is incomplete, badly sourced or plainly wrong. Without measurement, nobody on the team knows where to trust it.
You build an assistant on a document set, put together a bank of questions, test its answers, check its sources and write down its limits.
A bounded document set, questions the team genuinely asks, and the internal clearance to use those documents.
The lawyers who field the recurring questions, and whoever can authorise the use of the documents.
The protocol
The document set
You draw the line around what the assistant may read, and nothing else. A vague set cannot be measured.
The questions
You write a bank of questions the team genuinely asks, with the expected answer for each.
The assistant
You build it, inside the environment you already have.
The test
You ask the questions, record the answers and trace every cited source back to the document.
The measurement
You count: right, partial, wrong, invented. You end up with a number, not an impression.
The limits
You write down what the assistant must not be used for, and that page stays with the object.
What you leave with
- A prototype you can test.
- A reference bank of questions.
- A scoring grid you can reuse.
- The ability to judge the next tool someone offers you.
- An AI that is right every time.
- A move into production.
- Guaranteed compliance.
The details
- Format
- In-house on your document set, or open on an Academy set.
- Length
- One day.
- Group
- 4 to 10 people.
- Prerequisites
- You have used a conversational assistant at least once. No technical skill required.
- Environment
- Microsoft 365 or Google Workspace. The assistant is built with what you already have.
- Afterwards
- A one-hour review, thirty days later.
Price
- The scope of the document set.
- The number of people.
- Preparing the document set and the bank of questions.
- The technical environment to prepare.
The open-session price is the same for everyone in a given edition. In-house sessions are quoted after the scoping call, in writing.
Interested in this workshop?
On your document set, with your team. One hour to settle the documents and the environment.
On an Academy document set, signing up on your own. We let you know when the next edition opens.