12 questions to ask before implementing AI in your company

The decision to implement AI is often made faster than the answer to why. A competitor has implemented something, someone saw a presentation, licences are on offer. Below are twelve questions worth asking first. You do not need to know all the answers - but it is worth knowing which ones you do not know.

Goal

1. What problem are we solving - in one sentence? A good answer sounds like this: “preparing a reply to a request for a quote takes too long and clients wait two days”. A bad one: “we want to be modern”. If the problem cannot be named in one sentence, it will be hard to judge whether the solution helped.

2. How will we know it worked? This means a specific change in a specific task - shorter time, fewer corrections, fewer questions from clients. Without it, after a few weeks all you have is an impression.

Tasks

3. Which task exactly is supposed to change? Who does it, how often and with which materials. “Marketing” is not a task. “First draft of a service description, Anna, four times a month” is.

4. What does the task look like today and how long does it take? If nobody knows, the first step is measurement, not a tool. I describe how to do it without a reporting system in the article on measuring the effect of an AI implementation.

Data

5. What materials is the tool supposed to work on, and are they up to date? Price list, service descriptions, instructions, client history. A tool will not tidy up outdated documents - it will repeat their mistakes faster than a person.

6. Do these materials contain personal data or trade secrets? The answer determines which version of the tool you need, what contract must be in place and what must not be pasted in.

7. Who has access to this data today? Tools that work inside office suites use the user’s permissions. If permissions are too broad, the tool will show more than it should. More on this in the article on choosing an AI tool.

People

8. Who will use it, and what do they think about it? A team that fears the tool is meant to replace it rarely uses it well. It is worth talking before the implementation, not after.

9. Who is responsible for the implementation on the company side? One named person who has time for it. If you cannot name one, it is too early to implement - without an owner, projects fade out on their own.

10. How much time can the team spend learning? A new tool slows work down for the first few weeks before it speeds it up. If the team is in its busiest period of the year, it is better to postpone the start.

Risk

11. What happens if the tool makes a mistake? A mistake in a note for yourself costs a few minutes. A mistake in the price on a quote can cost a client. The answer determines how thoroughly results need checking. I describe splitting tasks by risk in the article on checking AI answers.

12. What do we not hand over to the tool at all? Pricing decisions, replies to complaints, assessing people - this list is worth writing down before anyone starts using the tool.

What to do with “we don’t know”

Every “we don’t know” points to a specific step:

• we don’t know what problem we are solving - talk to the people who do the work,
• we don’t know how long the task takes - a week of measurement,
• we don’t know whether the documents are up to date - tidy up the materials before choosing a tool,
• we don’t know who is responsible - a management decision before anything is bought,
• we don’t know what happens if it makes a mistake - split tasks by risk and set checking rules.

I answer these questions together with the team as part of an AI audit. It ends with a decision on which use case to start with and a card for a small pilot - described in the article AI pilot step by step.