The AI AdoptionFree AI pre-audit

Answers your team can find
without asking you.

A week of questions that came to you · Illustration, fictional business

OP

OperationsWhere is the latest price list?

A shared folder

The AI answers, with a link to the file

NS

New starterHow do I log a return for a damaged item?

The returns procedure

The AI answers from the procedure

AC

AccountsWhat did we agree with the freight company about the delivery fee?

An email thread

The AI answers, quoting the email

SL

Sales leadWhat did the customer say about the colour on last week’s call?

The notes of the call

The AI answers from the meeting notes

OP

OperationsWho approves overtime this weekend?

Only in your head

Written once as a rule, then the AI answers

SL

Sales leadCan this customer have longer payment terms?

Your judgement, case by case

Comes to you, with the facts gathered

Our illustration: a fictional business and the tools such a business usually keeps its answers in.

Most of the questions your team brings you already have an answer somewhere: the price list in a shared folder, the returns procedure in a document, what was agreed with a supplier in an email, what a customer said in last week’s call. Others have an answer only you hold, such as who approves overtime or which customers get longer terms. Asking you is the quickest route for the person who needs the answer, and the work behind each question waits until you reply. This page explains which questions AI can answer from your own documents and rules, what it needs to do it, who may see what, and what still comes back to you.

Why the questions come to you

In an owner-led business the answers are usually spread across places that were never meant to be searched together: shared folders on Google Drive or SharePoint, the inboxes of the people who dealt with a supplier, the notes of meetings, and the habits of the people who have been there longest. Someone new does not know where any of it is, and someone experienced knows that asking you is faster than looking. Each question takes you a minute to answer, and the work behind it stops until you do.

The AI your team already uses does not change this on its own. In a chat window it knows only what someone pastes into the conversation, so a question about your price list or your returns process gets general advice, or an answer nobody can check. It starts answering from your business once it can read the places where your answers are kept, and once the rules you carry in your head are written down where it reads them.

Three kinds of question

The questions that reach an owner fall into three kinds, and each needs something different before AI can answer it. The first is where something is: the latest price list, the template for a quote, the contract with a supplier. The answer already exists in a file, and an AI that can search the right folders finds it and links to it.

The second is what was agreed: the delivery fee settled with the freight company, the colour a customer chose on a call, the date promised for an order. The answer sits in an email thread or in the notes of a meeting, and the AI finds it when those are kept with the customer or project they concern, which is how we set up the record of meetings in turning a meeting into actions with AI.

The third is how you decide: who approves overtime, which customers get longer terms, when a refund outside the policy is acceptable. These answers are often written nowhere, because you have always given them yourself, and only you can supply them. Once a rule is written down, the AI can answer the same way every time, and the cases the rule does not cover still come to you, prepared, as we describe in the decisions waiting on you, prepared by AI.

What the AI needs to answer them

Three things are set up before the team starts asking. The AI gets read access to the places where the answers are kept, and only to those. The agreements that matter, the important email threads and the decisions taken in meetings, are kept with the customer or project they concern. And the rules you hold are written down in plain words where the AI reads them at the start of every question: who decides what, the standard terms, the limits and the procedures your team follows.

Every answer then shows where it came from, the document, the email or the meeting notes, with a link, so the person who asked can check it before acting on it. When the sources do not contain the answer, the AI says so, and the question goes to the person who owns that subject, the operations manager for a return or the accounts lead for a supplier invoice, rather than to you by default. Their answer is added to the procedure, so the same question is answered from the source the next time.

In the assistant your team already uses

New starterA customer wants to return an item that arrived damaged. They bought it on sale last month. What do I do?

From the returns procedure:

  • Photograph the damage and log the return against the original order.
  • Offer a replacement, or a refund to the original payment method.

Returns procedureOperations folder, kept up to date by the operations manager

Not in the procedure: items bought on sale. Sent to the operations manager.

What happens to the gap

  1. Sent to the operations managerWho owns returns, not to you by default

  2. Answered onceTreat it like any other damaged item, and refund what they paid.

  3. Added to the returns procedureBy the person who keeps it up to date

  4. Answered from the procedure next timeFor anyone who asks the same question

Our illustration: a fictional business. The assistant answers from the documents it can read and shows where each answer came from.

Who may see what

An assistant connected to your business answers each person from what that person can already open. Microsoft states it for Copilot: it only surfaces organisational data that the individual user has at least permission to view, and Microsoft recommends using the permissions of Microsoft 365 services such as SharePoint so that the right people have access to the right content.

Anthropic states the same for the Microsoft 365 connector of Claude, the assistant we work in: Claude mirrors each user’s existing permissions, and members cannot reach anything through Claude that they could not already see in Microsoft 365. For its Google Workspace connectors, each person signs in with their own Google account, and Anthropic’s help page says Claude can only open files that account has access to.

That makes the permissions on your folders the setting that decides what the team can get from the AI. Anthropic’s help page adds that its Microsoft 365 connector searches SharePoint across the whole organisation with each user’s permissions, so a spreadsheet of pay rates left in a folder everyone can open can appear in an answer to anyone in the team. Before anything is connected, the files only some people should see, such as pay, contracts, personal records and the management accounts, are moved to folders open only to them.

Your folders, and who can open each

Shared drives or SharePoint sites

  • Price list and standard termsEveryone

  • ProceduresReturns, deliveries, onboarding

  • Customer projectsQuotes, emails, meeting notes

  • Supplier agreementsTerms, fees, contacts

  • Pay and contractsRestricted

  • Management accountsRestricted

New starter: what are the pay rates for the warehouse?

Nothing to answer from. That folder is not open to them, so the assistant cannot read it for them.

Our illustration, from what Microsoft and Anthropic state about their assistants: each person gets answers only from what they can already open.

Where that data goes when the AI reads it matters to a business in the UK, Australia or New Zealand alike, and our guide on what you can safely share with AI lists the questions to ask about training, retention and location. How access and approvals are set for each role is covered in adopting AI securely.

Keeping the answers right

An answer is only as current as the document it comes from. Each source has a person who keeps it up to date: the price list belongs to whoever sets prices, the returns procedure to operations, the standard terms to you. When something changes, that person edits the document, and the next answer uses the new version instead of the one somebody remembers.

The questions the AI could not answer show what to write next: each one points to a document that is missing or a rule that was never written down. What tells you it works is whether the same question still reaches you a few weeks later. Your team also needs to know how to ask well and how to check the source before acting on an answer, which we cover in AI adoption training for your team.

Where to start

Start with one area where the questions come back often and the answers rarely change: the questions of new starters, the returns or delivery procedures, the price list and the standard terms. The documents are quick to gather, the rules are few, and the questions that still reach you show what is missing.

What was agreed with customers and suppliers comes next, once the team trusts the first answers, because it depends on emails and meeting notes being kept with the right customer or project. The rules about how you decide grow over time: each time you answer a question that was not covered, the answer is written down once, and the question stops coming back.

How we start

Organising the business knowledge and project information your AI needs is part of how we set up AI around a business, with permissions that reflect each person’s role and clear points for human approval. Your team learns through its own day-to-day tasks, including how to work safely with business information, and the Sprint includes a skills library with reusable instructions, so nobody has to explain the same work from scratch. This structure is the one we describe in what an AI adoption consultant does.

The first step with us is a free AI pre-audit: 45 minutes on a call where we look at how you and your team use AI today and which work to hand over first. Bring the questions your team asked you last week and where their answers were kept. You leave with a clear next step, with or without us. See how we help, who we are and how we handle security. In the news today: Google Meet recaps now post in Google Chat.

Bring your questions.

Free AI pre-audit