St. Louis · Kansas City

AI Data Readiness Assessment Template You Can Fill In Today

Before you buy an AI tool, it helps to know what it would be reading. Most owners have a rough picture: customers in one system, tickets in another, and the numbers that matter in a spreadsheet someone updates on Friday afternoons. This template gives that rough picture a score you can act on.
An AI data readiness assessment template is a set of plain questions that checks whether your company data can support one specific AI use case. This one covers five areas: where the data lives, who owns it, who can see what, how current it is, and how it connects. Score each answer from 0 to 3, add the points, and divide by the maximum to get a percentage that shows where to start.
In this article
What you need before you start
You need three things: one use case, a list of your systems, and two or three people who know them. Readiness only means something against a specific job, so pick one first. Good starting use cases are narrow and frequent, such as answering questions about upcoming renewals or about open support tickets.
Write down every system the use case would touch, then book an hour with the people who use those systems every day. That might be an office manager, a service lead and whoever handles billing. You do not need a data team or an enterprise AI budget to fill this in. You need honest answers from the people closest to the work.
The AI data readiness assessment template
The template has five question sets with five questions each, 25 in total. For each question, write a short answer, give it a score from 0 to 3, and name the person who owns the follow-up. The scoring rules are in the next section, so read them before you put numbers in the boxes.
1. Where the data lives
| Question | Your answer | Score (0 to 3) | Owner | Note |
|---|---|---|---|---|
| Can you list every system this use case needs, including accounting, ticketing, customer records and contracts? | ||||
| Which spreadsheets hold business information that is not in any system? | ||||
| Which shared folders or SharePoint sites hold documents the use case would need? | ||||
| For each type of record, such as the customer list, which system is the source of truth? | ||||
| Is any important information kept only in someone’s inbox, personal drive or head? |
2. Who owns it
This set covers governance in its simplest form: a named person for each system, and clear rules for who changes what.
| Question | Your answer | Score (0 to 3) | Owner | Note |
|---|---|---|---|---|
| Does each system have a named person who is accountable for it? | ||||
| When a record is wrong, who fixes it, and how does anyone report it? | ||||
| Who decides who gets access to each system? | ||||
| Do people agree on what key terms mean, such as “active client” or “open ticket”? | ||||
| If the owner of a system left tomorrow, would someone else know how it works? |
3. Who can see what
This set matters most if you plan to use Microsoft Copilot or any AI assistant or agent that answers from your files. Microsoft’s own guidance on preparing for Copilot starts with finding and fixing oversharing, and these questions put its checks into plain words.
| Question | Your answer | Score (0 to 3) | Owner | Note |
|---|---|---|---|---|
| Does each person see only what their role needs, such as payroll for payroll staff only? | ||||
| Have former employees and contractors lost their access? | ||||
| Are any sites, folders or files shared with the whole company or with “anyone with the link”? | ||||
| Does every shared site or team folder have an owner who still works here? | ||||
| Are folders that hold sensitive data, such as HR files or client contracts, limited to the right people? |
4. How current it is
This set checks freshness and completeness, the two problems that most often make an answer wrong.
| Question | Your answer | Score (0 to 3) | Owner | Note |
|---|---|---|---|---|
| When was the data in each system last updated? | ||||
| How long does a change in one system take to show up across the others? | ||||
| How many duplicate records are there, such as the same client entered twice? | ||||
| Which important fields are often blank, such as renewal dates or contact names? | ||||
| Are old or obsolete files still sitting next to current ones? |
5. How it connects
Each connection between two systems is a small data pipeline, the route information takes from one place to another. Fewer manual steps on that route means fewer chances for errors.
| Question | Your answer | Score (0 to 3) | Owner | Note |
|---|---|---|---|---|
| Can each system connect to others, or export its data in a standard format? | ||||
| Do your systems share an identifier for the same customer, such as an account number? | ||||
| How much information moves between systems by manual export, retyping or reports built by hand? | ||||
| When two systems report the same number, such as monthly revenue, do they agree? | ||||
| Does someone have the admin access needed to set up a connection? |
How to score your answers
Score each question on evidence, then turn the total into a percentage. Use these four levels:
- 0: No, or nobody knows.
- 1: Partly, and the answer comes from memory.
- 2: Mostly, and someone can point to where it lives.
- 3: Yes, and you can show proof, such as a report, a permissions list or a screen.
One rule keeps the score honest. A confident answer with nothing to show for it counts as a 1 or a 2, never a 3. A 3 should mean someone opened the system and checked.
Add up your points and divide by the maximum, which is 3 times the number of questions. With all 25 questions, the maximum is 75. Then look at each section’s subtotal out of 15, because the weakest section is where you learn what to fix first.
Here is a worked example for a company testing a renewals use case:
| Section | Score |
|---|---|
| Where the data lives | 11 of 15 |
| Who owns it | 9 of 15 |
| Who can see what | 7 of 15 |
| How current it is | 10 of 15 |
| How it connects | 8 of 15 |
| Total | 45 of 75 (60%) |
At 60%, this company is partly ready, and permissions are the first thing to fix. If you skip questions that do not apply to your use case, divide by 3 times the number you actually answered.
What your score means
Your percentage places you in one of four bands. They are a common convention among practitioners, and the 70% line in particular is a rule of thumb from client work, so treat the bands as a guide. The stakes are real either way: Gartner predicts that “through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data.”
Under 40%: not ready for this use case yet. Start with the basics. Name an owner for every system, write down the source of truth for each type of record, and close access for anyone who has left.
40 to 69%: partly ready. You can start, with care. Pick your lowest section and fix the two or three answers that would change the result for your use case. Leave the rest on a written list.
70 to 89%: ready for this use case. Run a small pilot on the use case you scored. Keep the template and check it again before you add a second use case, since a new job touches new data.
90% and above: ready for production. Your data readiness is strong enough to run this use case day to day. Keep it there with regular reviews of access and freshness, and score the next use case before you expand.
Do you need to clean your data first?
No, you can start before everything is clean, and your data is probably good enough to start on one narrow job. An AI model answers from the data it is given, so what matters is the data the first use case touches. Most practitioners advise starting with data that is good enough for one narrow use case, fixing the problems that would change the answer, and writing down the rest. Google Cloud’s guidance on preparing data points to the same basics: handle errors and missing values, keep units consistent, and build automatic data quality tests into the pipelines that feed it.
A full cleanup before you start can run long and still miss what the first project actually needs. Higher-risk uses, such as anything touching finances or personal records, deserve stricter checks. For most first projects, the template’s weakest section is the cleanup list. As ThrottleNet puts it, “AI without unified data is just a faster search bar,” so connecting the data usually comes before polishing it.
What a self-check can’t tell you
A self-check shows what your team believes about its data. That is useful and directional, but nobody has opened the systems to confirm it. Permissions are the clearest example. Finding every folder shared with “anyone with the link” usually takes reports run inside the systems themselves, which a meeting around a table cannot do.
A written assessment tests the answers against your real systems, permissions and data, and ends with a prioritized list of fixes. ThrottleNet’s version is a free written feasibility study. We deliver it in writing within five business days of a 30-minute discovery call. It includes a map of your current systems, the three to five best automation candidates with time and dollar estimates, and a custom ROI report. You keep it whether or not you work with us, and you can read what a feasibility study actually covers before you book the call.
If your template score came out under 70%, or one section scored much lower than the others, a written study is a sensible next step. If you scored high, run your pilot and keep the template for the next use case.
Frequently Asked Questions
How do you score a data readiness assessment?
There is no single industry standard. The 0 to 3 scale used here is one common choice, and other frameworks use 0 to 4 or 1 to 5. The arithmetic works the same on any scale: total points divided by the highest possible total. Pick one scale and stick with it, so you can compare scores from one quarter to the next.
Why does Microsoft Copilot readiness start with permissions?
Copilot answers using data each person already has permission to open. Microsoft says it can deliver accurate, secure responses when data is “well governed, current, and appropriately shared.” If a payroll folder is open to everyone, Copilot can surface it to anyone who asks the right question. Microsoft also advises removing inactive or obsolete files, since they lower answer quality.
How current does data have to be?
It depends on the decision the data feeds. There is no universal refresh interval. Set a maximum acceptable age for each important dataset, such as same-day for open tickets and monthly for financial summaries, and name who checks it. Some practitioners suggest monthly reviews by the person who looks after the data and a yearly review of who owns what, though no standard requires it.
What is the difference between a data owner and a data steward?
The data owner is the accountable person who decides who may use the data and for what. The data steward looks after quality and definitions day to day. In a smaller business, one person often does both jobs, and that is fine as long as the name is written down.

