Five Questions to Answer Before AI Adoption

"Should we use AI?" is too broad for any answer to be useful. Here are five better ones.

"Should our company use AI?" — this question is too large. Any answer to it is empty. Replace it with the following five specific questions, answerable in one internal meeting. When done, you will clearly know: whether to proceed, which process to start with, and how to describe your needs to a vendor.

Question 1: Which Process Consumes the Most Repetitive Labour?

Get to the action level: "a colleague spends one hour every morning opening emails one by one, downloading attachments, and copying figures into a spreadsheet."

If you can produce that kind of sentence, you have a candidate process. If the answer is a feeling-level description like "overall efficiency seems low," go observe for a week first and record actual actions — automation transforms actions, not feelings.

Question 2: Can the Inputs and Outputs of This Process Be Described Clearly in Words?

"Input is the PDF reports received daily, output is a summary pushed to the group" — yes, this process is suitable for automation.

"Input depends on the situation, output depends on what the boss wants" — a process that cannot be described has no standard method even for humans; machines cannot learn it. Standardise the process first, then discuss automation.

Question 3: What Is the Cost of an Error?

AI systems carry probabilistic characteristics — occasional errors are a feature (see AI project failure patterns, pattern four). So: if this process occasionally produces an error, are the consequences "a colleague fixes it during review, five minutes," or "client receives incorrect document, reputational damage"?

The first case: automate boldly. The second: still automatable, but the process design must include human review checkpoints — and that affects efficiency calculations, because review itself is also labour.

Question 4: Where Is the Data and Who Can Access It?

The raw material for the candidate process — emails, documents, historical records — where does it currently live? One system, or scattered? Consistent formats? Sensitive content?

The answer to this question directly determines project timeline and quote (see cost structure, layer three). The more centralised and organised the data, the faster and cheaper the project. The vaguer the answer, the more important a data inventory before signing.

Question 5: After Delivery, Who Is Responsible for Watching It?

An automation tool after go-live needs a "responsible person": monitoring that it is running normally day-to-day, and notifying the maintenance party when anomalies occur. The workload is small — but a name must be assigned.

Organisations that cannot name someone will abandon the tool after the first undetected failure. This is the quietest form of adoption failure.

After Answering All Five

All five questions with concrete answers: you are ready, and your first description to a vendor will be very clear — directly saving back-and-forth in the interview stage (see project workflow). Three or more unanswerable: fill in those answers internally first. The process of filling them in is free, and regardless of whether you proceed with AI, your understanding of your own processes improves.

FAQ

Q: All five are answerable, but the process is small. Is it worth doing?

Small processes are the ideal starting point: small scope, low quote, capped risk — and the lowest-cost education for your organisation on "how to work with AI vendors." The real output of the first project is half tool, half experience.

Q: The process management wants to do and the process the five questions surface are different. Whose takes priority?

Use question three to compare: do the one with lower error cost first. The first project's primary goal is building confidence and methodology — pick the battle with higher fault tolerance.