# Is Your Business Data Ready for AI? A Ten-Question Test

Canonical URL: https://business.theaieducator.io/posts/is-your-business-data-ready-for-ai
Publication: Dan Fitzpatrick Insights
Author: Dan Fitzpatrick
Topic: AI Readiness and Results
Published: 2026-10-19T15:43:00.000Z
Modified: 2026-10-19T15:43:20.064Z

A ten-question test to find out whether your business files are ready for AI: one named document, one owner and one check date per question, with a sheet to copy and a move for every result.

## In brief

Your business data is ready for AI when ten questions customers and staff ask every week each have one named document, one named owner and a check date from the last twelve months. Run the Single Source Test this week, fix competing or missing answers first, and let AI read only the documents that pass.

## Key takeaways

- AI cannot tell which of your documents is right, so competing versions of a price list, contract terms or checklist produce confident wrong answers.
- In the UK government's AI Adoption Research (28 January 2026), 21% of 3,500 businesses with five or more employees said data complexity had prevented AI adoption.
- In Moffatt v. Air Canada (14 February 2024), a Canadian tribunal held an airline responsible for a chatbot answer that contradicted the airline's own web page.
- The Single Source Test is ten questions, each needing one document, one named owner and a date it was last checked, not just last saved.
- A question that fails the test can still use AI for drafting when a person supplies the facts; it fails only where AI must read your files and answer alone.
- Each failing row has one move: pick the right version, write five plain lines, check or retire the stale file, or dictate what only one person knows.
- You do not need to fix your whole archive: test only the documents behind the first AI job you chose.

Your business data is ready for AI when ten questions that customers and staff ask every week each have one named document that answers them, one named owner, and a check date within the last twelve months. Where a question has two competing answers, or none, fix that before AI touches it. AI repeats whatever it finds, and it repeats it fast.

You do not need a data warehouse, a consultant or a spring clean of every folder to find out. You need an afternoon, a pen and ten honest questions. This article gives you the test, a sheet to copy and a move for each result.

## Why does AI struggle with a small business's files?

AI cannot tell which of your documents is right. Ask a 20-person firm for the current price list and you may be handed a PDF from March, a spreadsheet from June and an email from the owner that contradicts both. A person knows to ask Sam. AI takes whichever version it reaches first.

The UK government's largest recent survey of firms puts a number on the trouble. In [AI Adoption Research](https://www.gov.uk/government/publications/ai-adoption-research), published on 28 January 2026 by the Department for Science, Innovation and Technology, 21% of 3,500 UK private sector businesses with five or more employees said they had "too much data complexity (both lack of data and unstructured data)" and that it had prevented them from adopting AI. Among the 700 businesses already using AI, 23% said the same thing was holding back wider use. The telephone survey ran from 12 February to 2 May 2025. The report does not give a separate figure for firms under 50 employees, so read the numbers as a signal about UK firms in general, not a measure of yours.

"Unstructured data" sounds like a problem for a bank. In a small firm it means the quote template saved under five names, the terms in a PDF nobody has opened since 2024, and the reply that only the office manager remembers how to write. That is my reading of the term, not the report's.

What happens when AI speaks for the business from documents that disagree? On 14 February 2024, the Civil Resolution Tribunal of British Columbia decided [Moffatt v. Air Canada](https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do). The airline's chatbot told a customer he could claim a reduced bereavement fare after he had traveled. The airline's own bereavement page, which the chatbot linked to, said the policy did not cover requests made after travel. The tribunal found that Air Canada "did not take reasonable care to ensure its chatbot was accurate" and held it responsible for what the chatbot said. The chatbot did not malfunction. The business had two answers, and the customer was given the wrong one. This is a single Canadian decision and does not bind courts in the UK or the US, but the lesson travels. The customer service side of the case is covered in [AI for customer service in a small business](https://business.theaieducator.io/posts/ai-for-customer-service-small-business).

## What is the Single Source Test?

The Single Source Test is ten questions your business gets asked every week. For each, you name the one document that holds the right answer, the person who owns it, and the date it was last checked. A question without all three is not ready for AI to answer.

That is my suggested way of thinking, not research. Each part earns its place:

- **One document.** Two people who do not own the answer should be able to find the same document in under a minute. If they find different ones, you have two answers.
- **One owner.** A name, not "the office" or "the team". The owner is the person who corrects the document when the answer changes.
- **A check date.** The date someone last confirmed the answer is still right, which is not the date the file was last saved. A document saved yesterday may have been checked two years ago.

A question that fails the test can still use AI for drafting, as long as a person supplies the facts. It fails only where AI is expected to read your files and answer for itself.

## How do you run the Single Source Test?

Start by collecting ten questions from places where they leave a trail: your last thirty sent emails to customers, the questions new hires asked in their first month, the calls the office answers, and the answers you give more than once a week. If you already listed ten customer questions for your busiest season, as in [AI for small retail and hospitality before the holiday rush](https://business.theaieducator.io/posts/ai-for-retail-and-hospitality-small-business), begin with that list. Aim for three from customers, two from quotes and sales, two about how work gets done, one about people (leave, expenses, equipment) and two of your choosing.

Then fill in the sheet. Copy this into a spreadsheet and add rows to ten:

| # | Question | The one document | Owner | Last checked | Verdict |
|---|---|---|---|---|---|
| 1 | What is our returns window? | Returns policy, shared drive, Policies folder | Priya | 3 September 2026 | Pass |
| 2 | Who signs off a quote over our usual limit? | (none written down) | (nobody) | (never) | None |
| 3 | What does the new-starter checklist cover? | Two versions, one marked "final v2" | Unclear | 2024 | Two documents |

Finally, ask two people who do not own the answers to find each document, and time them. Where they land on different files, or give up, the sheet has told you something your own memory could not.

## What do you do with each result?

Give every row one of five verdicts, and take the move beside it this week.

| Verdict | What it means | Your move | Ready for AI to read? |
|---|---|---|---|
| Pass | One document, a named owner, checked in the last twelve months | Diary a recheck for three months ahead | Yes, with a person checking what customers will see |
| Two documents | Versions disagree | Pick the right one. Move the others to a folder called Old, outside the shared drive | Not yet |
| None | Only people can answer it | The owner writes five plain lines. That is the document | Not yet |
| Stale | One document, not checked in a year | The owner checks it, dates it, or retires it | Not until checked |
| In a head | One person knows and nothing is written | Ask that person to dictate the answer. Have AI turn the dictation into a first draft, then have the person correct it | After the person signs it off |

The last row is the quiet gift in all this. AI answers badly from a mess and helps well in clearing one.

When a customer will see what AI produces from a passing document, apply the Loop Test. The Loop Test is three questions I ask before accepting "a human checks it" as an answer: Would that person notice a wrong result without being told to look for it? Would they notice in time to stop it mattering? And if they missed it, could the work be put back? Oversight that survives all three is a safeguard. Oversight that fails any of them is a signature.

## What does it look like in a real business?

Take a hypothetical 22-person commercial cleaning company. The office manager writes down ten questions from the last month of client emails and new-cleaner questions, then fills in the sheet with the owner.

Three rows pass. Three have two versions: the standard contract terms, the notice period for canceling a contract, and the supply request form. Two have no document at all: what a cleaner should do when a colleague is sick and a site is uncovered, and what to do when a client reports a missing item. One is stale: the induction checklist was last confirmed two years ago. One lives in the head of the operations manager: which client sites need a keyholder.

The fix takes the office manager about ninety minutes to settle the three versions, plus short slots across a week to write the two missing documents. The owner decides that until the other rows are fixed, AI may draft replies to client emails using only the three passing documents. The test did not slow AI down. It told them where to begin.

## Does a small business need to fix its data before using AI at all?

No, and waiting for a perfect archive is a good way never to start. When you paste the facts into the chat yourself, you are the source, and the test does not apply. It starts to matter the moment AI reads your files, searches your shared drive, answers customers, or helps a new hire find what to do.

So test only the documents behind the first job you chose. If that job is replying to customer emails, ten questions about customers are enough. [How should a small business start using AI](https://business.theaieducator.io/posts/how-should-a-small-business-start-using-ai) covers how to pick that job. If your team already pastes client files into public tools, that is a separate risk, covered in [shadow AI in a small business](https://business.theaieducator.io/posts/shadow-ai-small-business).

## What I See When I Train Business Teams

Across the business teams I train, the question "where is the current version?" is almost never answered with a location. It is answered with a name: "ask Sam." That works until Sam is on holiday. It stops working when the thing being asked is AI, because AI cannot ask Sam.

The second pattern is the best document in the company. It is usually a proposal or a standard reply that someone copied from a job two years ago, improved, and never saved as the master. It is the one everyone wants AI to learn from, and it sits in one person's sent folder. Moving it to a shared place with an owner and a date is the cheapest readiness work most firms can do.

## What should you do this week?

1. **Monday:** write your ten questions. Use your sent emails, not your imagination.
2. **Tuesday:** fill in the sheet, and put a name beside every document you can find.
3. **Wednesday:** ask two people to find each document, and time them.
4. **Thursday:** settle every Two documents and Stale row. Move the losers to Old.
5. **Friday:** pick the one AI job that uses only passing rows, and put a three-month recheck in the diary.

From now on, put "last checked" and the owner's name at the top of every document you keep.

## Where to go next

If you want to know where your business stands across the whole picture before you spend another month experimenting, take the free [Business AI Readiness Scorecard](https://theaieducator.io/business-ai-readiness?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=is-your-business-data-ready-for-ai), which takes about five minutes. If your team will be working from these documents through AI and nobody has shown them how to do it well, that is what my [AI training for teams](https://theaieducator.io/ai-training-for-business?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=is-your-business-data-ready-for-ai) is built for. I have worked with organizations that include Welsh Water, Hyve Group and Ealing Council.

For the wider picture of readiness, see [how ready is your business for AI](https://business.theaieducator.io/posts/how-ready-is-your-business-for-ai), where tidy files are one of seven things to check, and [does a small business need an AI policy](https://business.theaieducator.io/posts/does-a-small-business-need-an-ai-policy) for the rules about what goes into AI tools.

## Sources and further reading

- Department for Science, Innovation and Technology (research by IFF Research and Technopolis Group), "AI Adoption Research", 28 January 2026: [gov.uk](https://www.gov.uk/government/publications/ai-adoption-research)
- Civil Resolution Tribunal of British Columbia, "Moffatt v. Air Canada, 2024 BCCRT 149", 14 February 2024: [decisions.civilresolutionbc.ca](https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do)

*Dan Fitzpatrick, The AI Educator, is a Forbes contributor, international keynote speaker and bestselling author who has trained more than 150,000 people across 30+ countries. [About Dan](https://theaieducator.io/about?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=is-your-business-data-ready-for-ai)*


## Frequently asked questions

### How do I know if my business data is ready for AI?

Write ten questions your customers and staff ask every week. For each, name the one document that answers it, its owner and the date it was last checked. If every question has all three, AI can read those documents safely. Any gap tells you what to fix first.

### Do I need to clean all my files before using AI?

No. If you paste the facts into the chat yourself, you are the source and no clean-up is needed. Tidying matters when AI reads your files, searches your drive or answers customers. Then test only the documents behind the one job you chose first.

### What does AI-ready data mean for a small business?

For a small business it means ordinary documents, not a data warehouse: one current version of each price list, policy, template and standard reply, stored where your team can find it, with a named owner and a recent check date.

### What happens if AI answers customers from out-of-date documents?

The business is likely to be held to the answer. In Moffatt v. Air Canada (British Columbia, 14 February 2024), a tribunal made the airline responsible for a chatbot answer that contradicted its own web page. It is one Canadian decision, not binding in the UK or US.

### Who should own a document that AI reads?

One named person, not a team or an office. The owner corrects the document when the answer changes and confirms it is still right on a set date. If nobody can be named as the owner, the question fails the test and should not be answered by AI.

### How often should I recheck documents that AI uses?

Every three months is a sensible default for anything customers see, such as terms, returns or delivery. Slower-moving documents can wait up to twelve months. Put the recheck date in a diary, and write the last checked date at the top of the document.

### Can AI help me fix messy business documents?

Yes, once a person has decided what is true. Ask the person who knows an answer to dictate it, have AI turn the dictation into a first draft, then have that person correct and sign it off. AI helps you write the document; the owner stays responsible for it.

---
Source: [Is Your Business Data Ready for AI? A Ten-Question Test](https://business.theaieducator.io/posts/is-your-business-data-ready-for-ai)
Publisher: [The AI Educator](https://theaieducator.io)