Which AI Tools Does Your Company Use? Build a Register in One Afternoon

Picture a normal Monday at three small companies.

At an accounting office, the office manager renews the cyber insurance. The form has a new question: “List the AI tools that your staff use.” She writes down three. Then she asks each team. By Friday, her list has 14 tools. Two of them run on personal accounts. One is a browser extension that reads every web page that the staff open.

At an online store, the help desk software installs an update overnight. In the morning, every agent sees a new button: “Draft reply with AI.” Nobody asked for it. Nobody knows what the vendor does with the customer emails that the AI reads.

At a design agency, a client joins a video call. A fourth guest is already there: an AI notetaker. A colleague connected it to her calendar months ago, and now it joins every meeting. The client asks, “Who invited the robot?” Nobody has a good answer.

Nobody in these stories did anything wrong on purpose. The problem is simple: nobody kept a list.

That list has a name: an AI tool register. Today I am sharing a template for it, in four versions, with an Excel file that does the math for you.


What an AI tool register is

An AI tool register is a list of every AI tool that your company uses. For each tool, it says what the tool is for, who is responsible for it, and how risky it is. It answers four questions:

  1. Which AI tools do we use, and for what?
  2. Who is responsible for each one?
  3. How risky is each use?
  4. Is each use approved, approved with conditions, or not allowed?

Think of a restaurant kitchen. A good kitchen keeps a list of every appliance: the oven, the fryer, the meat slicer. For each one, the chef knows who cleans it, when someone last checked it, and which ones need extra care. Nobody calls that list bureaucracy. It is how the kitchen passes the health inspection and avoids a fire.

Your AI tools are your appliances. The register is your list.

Why you need one now

There are three reasons.

  1. People ask. Customers, insurers, and auditors ask “Which AI do you use?” more often every year. You want to answer in minutes, not in a week.
  2. AI arrives without asking. A software update can switch on an AI feature overnight, like the help desk button in the second story. You may already use AI that nobody chose.
  3. Your AI policy depends on it. Our AI acceptable-use policy tells staff to use only approved tools. Somebody must write down what “approved” means.

From the template, section 1:

Our AI Acceptable-Use Policy tells staff to use only approved tools. This register is where we write down “approved”. When a customer, an auditor, or an insurer asks “which AI do you use?”, this register is the answer.

One row for each use, not for each tool

Most people start with a list of products. The template asks for something a little different: one row for each use.

Think of the family car. It is perfect for the school run. You would not use it to carry barrels of chemicals. It is the same car, but a different job brings a different risk.

AI tools work the same way:

From the template, section 2.1:

The same tool can be safe for one task and risky for another. A chat assistant that drafts internal emails is low risk. The same chat assistant that drafts answers to customers about their accounts is higher risk. Give each use its own row.

What goes on the list

Most people think of chatbots first. The template asks for much more. Here is part of the list:

From the template, section 2.2:

– AI features inside software that we already use, for example email, documents, CRM, phone systems, and design tools

– meeting recorders and AI notetakers

– browser extensions and AI browsers

– AI agents that can take actions, and AI tools connected to our systems

– free or personal AI accounts that staff use for work, with the status “Prohibited”, so that the decision is on record

Look at the last item. Why write down a tool that you do not allow? Because a decision that nobody wrote down is easy to forget and easy to argue about. When a new employee asks, “Can I use my free chatbot account?”, the register already says no, and it says why.

How the risk score works

Think of a weather forecast. It tells you two things: the chance of rain, and how much rain will fall. A 90% chance of light drizzle is not a problem. A 20% chance of a hurricane is a big problem. You need both numbers to decide what to do.

The register works the same way. Each use gets two numbers from 1 to 5:

  • Likelihood: how probable is it that harm happens in the next 12 months?
  • Impact: how bad is the worst case that could credibly happen?

Multiply them, and you get a risk score from 1 to 25. The Excel file does the multiplication for you. The score then puts the use into one of four levels:

LevelScoreReview every
Low1–412 months
Medium5–96 months
High10–153 months
Critical16–253 months

Who decides at each level:

  • Low: the tool owner or a manager approves it.
  • Medium: the approver sets written conditions.
  • High: the approver sets conditions and a plan to lower the score.
  • Critical: nobody approves it until the score goes down.

Here is an example. Your support team uses a chat assistant, on a business plan, to draft replies to customer emails. A person reads every reply before it goes out.

  • Likelihood: 2. Harm is unlikely, because a person checks each reply.
  • Impact: 3. A wrong answer could reach a customer, and you would need to correct it.
  • Score: 2 × 3 = 6. Level: Medium. The Excel file sets the next review for 6 months later.

Now change one thing. An AI agent sends the replies by itself, and nobody checks them first. The likelihood goes up to 4, because nobody stops the mistakes. The impact goes up to 4, because the agent can also make promises to customers. The score is now 16: Critical. It is the same AI, but a different setup gives a very different risk.

Four rules that keep the scores honest

Scores only help if people score honestly. So the template has a few simple rules:

From the template, section 6.5:

1. Score the use as it runs today, with the controls that are really in place. A control that is only planned does not count yet.

2. If you are not sure between two numbers, choose the higher one.

3. If the use affects decisions about people, the impact is at least 4.

4. If the tool sees Restricted data and the vendor terms are “Unknown”, the impact is at least 4.

Rule 1 is my favorite. A smoke alarm that is still in its box does not protect your kitchen. A safety step that you plan for next month does not protect your data today.

Rule 3 matters a lot. If AI helps to decide who gets hired, who gets a loan, or which patient gets a call first, a mistake hurts a real person. The template never lets those uses look small.

The Excel file does the math

You do not need to be an Excel expert. The file does the work:

  • It calculates the risk score and the risk level for each row.
  • It colors each row by level: green, yellow, orange, or red.
  • It calculates the next review date from the decision date and the level.
  • It gives you drop-down lists, so everyone uses the same words.
  • It has a “Scoring guide” sheet with both scales, so nobody has to guess what “3” means.
  • It has a “Change log” sheet, so you can see who changed what, and when.

The template also includes a one-page request form. When someone wants a new AI tool, they fill it in. The form asks the questions that you need for the score. For example: which data will the tool see, and can it take actions by itself?

Four versions: choose the one closest to you

Every industry has its own risks. So the template comes in four versions:

  • Generic, for any company: placeholders that you fill in, and 5 example rows.
  • Healthcare clinic group, for clinics and medical groups: patient data rules, a higher floor for clinical uses, and the medical director in every clinical decision.
  • B2B AI software company, for companies that sell an AI product: separate rows for product AI, and extra steps for a new model provider.
  • Consulting firm, for consultants and professional services: each client’s AI choice comes first, and a simple approval chain for a small team.

The three industry versions use fictional companies, so you can see real rules in a realistic business. For example, the healthcare version has a strict rule about patient data. PHI means protected health information: anything that identifies a patient and relates to their health or care. A BAA (Business Associate Agreement) is the contract that US health law requires before a vendor can handle PHI.

From the healthcare version, section 6.3:

5. If the use touches PHI and no signed BAA covers it, the status is “Prohibited”. We do not score it until a BAA exists.

In plain words: no contract, no patient data, no discussion.

Keep the list alive

A register is like the emergency phone list on your fridge. If nobody updates it, it is worse than no list, because people trust it.

So the template sets a review date for every row. The riskier the use, the sooner the review. Some events also trigger an early review:

From the template, section 8.2:

– the vendor changes its model, its terms, or its owner

– the tool starts to see a higher data tier

– people outside the company start to see the output or talk to the tool

– an agent gets new permissions

– an incident or a near miss happens

Every three months, the template also asks you to check for new tools. Ask each team what they use. Look at the AI features that your software added since the last check.

Start in one afternoon

  1. Download the Excel file and the guide. Choose the version that is closest to your company.
  2. Write down the AI tools that you already know about. Include the AI features inside your email, office, and customer software.
  3. Ask each team one question: “Which AI tools did you use for work this month?” Add the answers as they arrive.
  4. Add one row for each use. If one tool has two different jobs, give it two rows.
  5. Score each row with the “Scoring guide” sheet. When you are not sure, choose the higher number.
  6. Decide each row: approve it, approve it with conditions, or prohibit it.
  7. Put the first review date in your calendar. The Excel file tells you when.

Most small companies can do steps 1, 2, and 4 to 7 in one afternoon. Step 3 takes a few days, because people need time to answer.

Download the template

The templates are free on GitHub. You can use and adapt them, also for commercial work.

The healthcare, software, and consulting versions are in the same collection.

A list you can show anyone

Go back to the three stories. The office manager now knows about all 14 tools, and she can answer the insurer in minutes. The online store sees the new AI button in the register before staff use it. The design agency switched off automatic join for the notetaker, and now it asks the client first.

A register does not stop your team from using AI. It turns “I think we use a few AI tools” into a list that you can show to anyone who asks.

This is the second of four templates. Next come an AI vendor security questionnaire and an AI incident-response playbook.

In my next post, I go deeper. I explain the design decisions behind this register, the alternatives that I rejected, and the frameworks behind each rule. That post is for IT, security, and compliance people.

Have a question or feedback about the register? Contact NosisTech.

Keep reading: AI governance templates

Next article: Inside My AI Tool Register: 10 Design Decisions and the Alternatives I Rejected

Template 1: Your Team Already Uses AI. Here Is a Policy That Makes It Safe

Template 3: What Does Your AI Vendor Do With Your Data? Ask Before You Sign

Topic hub: AI Governance

All articles: NosisTech Writing

This post is general information, not legal advice. The opening stories and the example companies are fictional. Any resemblance to real people or companies is a coincidence. Laws about AI change often and differ by state and country. Ask your legal counsel before you rely on any template.

Follow NosisTech on LinkedIn