AI readiness assessment: 12 questions before you build

Most AI readiness assessments score the whole company: data maturity, leadership, skills, culture. You get a number out of five and no idea what to build on Monday.

Readiness is easier to judge one job at a time. A business can be ready to automate its quotes this month and its month-end close next year. So the questions below are asked about one task, the one you would most like to stop doing by hand. Answer them for your top three tasks and you have your shortlist.

The process

1. How often does the task happen?

Count it. Enquiries answered, invoices keyed in, reports assembled, per week.

Ready when the task repeats twenty or more times a week, or takes a full day each time it comes round. Volume is what turns a small saving per task into a number worth funding.

2. Can someone write down what "done" looks like?

One sentence a new hire could check their work against. "The quote matches the price list and goes out within an hour." "Every invoice line is in the ledger with the right cost code."

Ready when that sentence exists and two people on the team agree with it. That sentence becomes the system's acceptance test.

3. Where do the exceptions go today?

Every process has cases that do not fit: the unusual order, the missing document, the angry customer. Someone handles them now.

Ready when you can name that person. The system handles the routine cases and hands the rest to them, with the context attached.

The data

4. Where do the answers live?

A system, a shared drive, a spreadsheet, one person's head. Write it down for each step.

Ready when the answers sit in something a system can read, even if that something is a spreadsheet. Where they sit in paper or scanned files, document processing turns them into records first, and the rest follows.

5. Can those systems be connected?

Most business tools allow it: CRMs, accounting packages, inboxes, booking systems, databases. Some only allow an export.

Ready when you know which tools hold the data and who holds the admin login for each. Checking what each one allows is a short, early step in any good assessment.

6. Which numbers must be exact?

Prices, balances, quantities, dates. Anything a customer or a board will rely on.

Ready when those numbers are listed, so the build computes them in code and the model only writes the words around them. That split is how AI stops getting your numbers wrong.

The people

7. Who will own it after launch?

Not the vendor. Someone inside the business who can change the rules, read the log and decide when it is working.

Ready when that person is named and has an hour a week for it.

8. Have the people doing the work been asked?

The people describing a process and the people doing it often describe two different processes. The gap between them is usually where the hours go.

Ready when the team doing the task has walked someone through it, step by step, with the real files open.

The risk

9. What does a mistake cost?

A wrong internal summary costs a correction. A wrong quote, a wrong medical instruction or a wrong payment costs much more.

Ready when each step is marked low or high cost. High-cost steps get a person's approval before anything goes out. The system prepares, a named person releases.

10. What must it never do?

Promise a delivery date, quote a discount, give advice it is not qualified to give, email a customer unread.

Ready when that list is written. It becomes the system's written boundary and its test set: the questions it passes only by refusing.

The money

11. What does the task cost today?

Hours per week, times the loaded cost of the people doing it, plus what slow replies and errors cost you in lost work.

Ready when you have that figure. It is the baseline every saving is measured against. Our automation payback calculator works it out from four numbers.

12. What will it cost to run each month?

A system has a build cost and a running cost: model usage, hosting, monitoring and upkeep. Plain code often does part of the job for less every month. We have swapped a model for plain code five times in our own systems, each time to lower the monthly bill.

Ready when the monthly figure is estimated next to the saving, so the decision is a comparison of two numbers. Our free agent-or-automation tool tells you which kind of build a task needs from three questions.

How to read your answers

Score each task by how many of the twelve you could answer.

• Ten to twelve: the task is ready to scope. Go straight to a build conversation.
• Six to nine: the task is a strong candidate, and the gaps are the first part of the work. Usually that means writing down "done", naming the owner and costing the baseline.
• Five or fewer: the opportunity is real and the groundwork comes first. Mapping the process and connecting the data is the project that makes the next one pay.

Every score has a next step. The order matters more than the score.

What a full assessment adds

A checklist gets you a shortlist. A structured assessment turns it into a decision you can fund. Ours runs two weeks at a fixed fee: interviews with the people doing the work, a map of where the time goes, the process costed as it runs today, and a shortlist of automations with effort, running cost and risk for each. The fee is credited in full against any build that follows.

If you have answered these questions for your top task, bring the answers to the first call. Book the two-week AI assessment, or read how to choose an AI development company in Dubai if you are comparing providers.