Is AI worth it for your business? Check before you pay.

I want to find out whether AI is worth it for my business before we spend money on it. Interview me with the check below (from ai.teren.no) and write my plan at the end.

HOW TO RUN THE INTERVIEW
- Ask one question at a time, in plain words, and wait for my answer. Give a short example when I'm unsure.
- Use what you already know about me and my work: suggest an answer and let me confirm it, instead of asking what you already know.
- Reply in the language I write in. Use my currency (NOK if I'm in Norway).
- Be honest. "Not yet" or "not this task" is a good result when the numbers say so. Never invent numbers: if I don't know one, agree a range with me and mark it as an estimate.
- Keep the whole interview to about 10-15 minutes.

STEP 1 - READINESS
Ask these five yes/no questions. Anything I'm unsure about counts as no.
1. Is one named person responsible for how we use AI?
2. Does everyone know which information must never go into an AI tool?
3. Do we try a new AI tool on our own real tasks before the team relies on it?
4. Is there a short written note on what staff may and may not do with AI?
5. Can we tell whether a task got better, in time, quality or money?
4-5 yes: ready to pilot one task now. 2-3 yes: prepare first (a one-page AI policy, agree which information stays inside, show the team the basics). 0-1 yes: map the work first (the tasks that repeat every week, who does them, how long they take).

STEP 2 - THE FIRST TASK
Ask for up to three tasks I'd like help with; suggest some from what you know about my work. For each one ask:
a) Does it happen every week or more often?
b) Could I describe its steps in about ten minutes?
c) Is there someone who spots a bad result at a glance?
Also ask roughly how many hours a week it takes now.
Recommend the task with three yes answers and the most hours. If none has three, say so and say how to make the best one ready (write down the steps, name who checks). The most painful task is rarely the best first one: a quick, clear win comes first. Tasks where text or documents go in and text comes out are the easiest start. An assistant for our own staff is a safer start than one that talks to customers.

STEP 3 - COST
For the chosen task, ask for my figures, or estimate them with me:
One-off: setup and integration by a supplier; our own hours to set up, test and learn.
Monthly: fees (licences, AI model usage, hosting, support); hours per week spent checking the output; hours per month for the person who owns the task.
Use our real hourly cost: wage plus holiday pay and employer's tax (often about 1.3 times the wage).
Remind me that the subscription price is not the cost: checking time and the owner's time are real costs too.

STEP 4 - PAYBACK
Ask how many hours a week the task would save, and how many months pass before the gain starts (setup and pilot). Then calculate:
- Monthly gain = hours saved per week x 4.33 x hourly cost, plus any other gain I can actually measure.
- Monthly running cost = fees + checking time + owner's time. Checking time only counts once the gain starts.
- Cumulative result: start at minus the one-off costs, subtract fees and owner's time during the waiting months, then add (gain - running cost) for every month after that.
Show two cases over 12 months: BASE (my numbers) and CAUTIOUS (half the gain, twice the waiting time). Give the payback month for each, or say it doesn't pay back within 12 months. If only the base case pays back, say plainly that this is a bet on things going to plan, not a quick win.

STEP 5 - BASELINE, DATA AND THE HONEST NO
Ask:
- Which one number will show success? Prefer one that is about money: time from request to answer, minutes per task, or the share of results that had to be redone. Not "how many people use it".
- What is that number today? If I don't know, I measure it for two weeks before the pilot starts.
- Which information must never go into the tool (for example personal data about customers or staff, health information, prices and margins, contracts, passwords)?
- Who checks the results? They should check the final result, not every step.
Then go through the reasons to say no: a mistake would be costly and hard to undo, and checking would cost more than the gain; the task is rare or different every time; the information cannot leave the company; nobody owns the task; the team isn't ready for change. If any applies, recommend "not this task" with a date to look again.

FINALLY - MY ONE-PAGE PLAN
Write it with exactly these headings:
1. Verdict: Start a pilot / Start a small pilot / Prepare first / Make one task ready first / Map our work first / Not this task. One sentence why.
2. Readiness: score out of 5, and one action for every "no".
3. First task: the task and why it was chosen.
4. Pilot: one task, one assistant, six weeks. The number we measure and its value today (or "measure for two weeks from [date]"). What counts as success. Start date and stop date. Who checks.
5. Cost and payback: one-off cost, monthly cost, monthly gain, payback month in the base and cautious case.
6. Data rules: what never goes in. Company accounts only.
7. Risks we name ourselves: wrong answers and who checks; information and where it goes; decisions about people stay with a person; dependence on one supplier (can we switch the AI model and keep our data and instructions?).
8. If the answer is no: the reason and the date to look again.
Then offer a short version for my team or co-owner.