TL;DR & Quick Summary
Readiness isn't a company-wide score. It's whether one specific project can succeed. Answer these 20 questions for the workflow you have in mind. Each "no" is something to fix before spending money, and most can be fixed in days.
Problem (Q1–4): can you name the workflow, its owner and a measure?
Data (Q5–9): does the data exist, can you reach it, and is it good enough?
Systems (Q10–13): can an AI read from and write to the systems involved?
People (Q14–17): who will use it, own it and support it?
Risk (Q18–20): what may the AI do, and what happens if it's wrong?
Key Takeaway: Most "not ready" answers are organisational, not technical: no agreed measure, no owner, data nobody can export. Name them and they're usually quick to fix.
Get Started: Want help with the answers? Schedule a strategy call with Cogniq AI or see our custom AI development services.
How to Use This Checklist
Pick one workflow, not "AI for the business". Gather the person who runs it, someone who knows the systems, and whoever controls the budget. Answer each question yes, partly or no, and write down the evidence. A "yes" without evidence counts as "partly".
At the end, count the "no" answers in each section. The section with the most gaps is your first job.
1. The Problem
Q1. Can we name the specific workflow and the step that should change? "Improve customer service" fails. "Answer order-status emails without an agent" passes.
Q2. Do we know how the workflow performs today? Volume per week, time per case, error rate, response time. Without a baseline, nobody can say whether the project worked.
Q3. Have we agreed what success looks like, in numbers? A target such as "resolve 30% of order-status emails automatically within three months" turns an idea into a project.
Q4. Is the value worth the effort? Multiply volume by time per case by cost per hour. If the answer is small, choose a different workflow. Our guide to how long AI automation takes to pay back shows how to estimate it.
2. The Data
Q5. Do we know which data the workflow depends on? List the records, documents and fields a person uses to do the job today.
Q6. Can we actually access it? Can someone export it, query it or reach it through an API this week? Data that exists but can't be reached is the most common blocker.
Q7. Is it accurate enough for this use? Spot-check 50 records. Duplicates, missing fields and outdated entries all matter more than volume.
Q8. Do we have real examples to test against? Collect 50–100 real cases with the right answer. These become the test set that decides when the AI is ready.
Q9. Do we know what we're allowed to do with the data? Personal and customer data may have restrictions on where it can be processed and by whom. Find out before the build, not after.
For forecasting projects specifically, our data-readiness guide for demand forecasting goes deeper.
3. The Systems
Q10. Do we know which systems the AI must read from? CRM, help desk, ERP, shared drives, email.
Q11. Does it need to change anything, or only read? Read-only projects are faster, cheaper and safer. Start there if you can.
Q12. Do those systems have usable APIs? If not, access may need database exports or other workarounds, which add cost. See AI integration for legacy systems.
Q13. Can we test without touching live data? A test environment or a safe copy avoids learning lessons on real customers.
4. The People
Q14. Who will use the result day to day? Involve them early. They know where the edge cases are, and they decide whether the tool gets used.
Q15. Who owns it after launch? One named person accountable for how it behaves and whether it's improving. Projects without an owner decay.
Q16. Who will maintain it technically? Internal staff, the original builder, or a support agreement. Decide before the build. Our guide to partner vs in-house teams covers the trade-offs.
Q17. Is leadership agreed on the priority? Competing priorities stall projects halfway. If leaders disagree on which project comes first, settle that before building anything.
5. Risk and Rules
Q18. Have we decided what the AI may do on its own? Answering questions is low risk. Sending emails, changing records or issuing refunds needs limits and approvals. See human-in-the-loop approval gates.
Q19. Do we know what happens when it's wrong? How will mistakes be noticed, corrected, and prevented from repeating? Who does a customer contact?
Q20. Do any rules or regulations apply? Sector rules, data protection and contractual obligations can all shape the design. Check early, and take advice where needed. For integrations with older systems, what AI governance costs covers the controls involved.
Reading Your Results
| Gaps | What it means | What to do |
|---|---|---|
| 0–3 "no" answers | Ready for a first project | Scope it and get quotes |
| 4–8 "no" answers | Ready soon | Fix the gaps first; most take days to weeks |
| 9+ "no" answers, or any "no" on Q1, Q6 or Q15 | Not ready for this project | Pick a different workflow, or fix the basics first |
Three questions carry extra weight. Q1 (a named workflow), Q6 (accessible data) and Q15 (an owner) are the most common causes of failed projects. A "no" on any of them is a stop sign, whatever the total.
A Worked Example
For illustration, a distributor considers automating replies to "where's my order?" emails.
- Problem: yes on all four. About 400 emails a week, 4 minutes each, a target of resolving 30%.
- Data: partly. Order and shipment data exists, but tracking numbers sit in a carrier portal with no export (Q6 partly), and nobody has collected example emails yet (Q8 no).
- Systems: read-only (Q11 yes). The ERP has an API; the carrier portal doesn't (Q12 partly).
- People: the support lead will own it (Q15 yes). IT can't commit maintenance time (Q16 no).
- Risk: replies only, no refunds (Q18 yes). No plan yet for wrong answers (Q19 no).
Result: four gaps, so ready soon. The fixes are to collect 100 example emails, confirm the carrier's data feed, agree a support arrangement, and write the escalation rule. That's about two weeks of work. Then scope the build.
Turning Gaps Into a Two-Week Plan
Each "no" or "partly" answer should become a task with an owner and a date. Typical fixes and how long they take:
| Gap | Typical fix | Typical time |
|---|---|---|
| No baseline (Q2) | Count a week of cases and time a sample | 1 week |
| No test set (Q8) | Collect 50–100 real examples with the right answer | 1–2 weeks |
| Data hard to reach (Q6) | Agree an export or API access with IT or the vendor | 1–3 weeks |
| No owner (Q15) | Name one person and give them time in their objectives | Days |
| No rules on actions (Q18) | Write down what the AI may and may not do | Days |
Run the fixes in parallel. Most organisations can close four or five gaps in two weeks, which turns "ready soon" into "ready to scope".
Who Should Answer Which Questions
Answering alone tends to produce optimistic results. Split the questions:
- The person who does the work answers the problem and people questions. They know the real volumes and edge cases.
- Someone from IT answers the data and systems questions, and confirms access in writing.
- The budget holder answers on value, priority and ownership.
- Legal or compliance, where relevant, answers the risk questions.
Disagreements between them are useful. They show exactly where the project would have stalled.
Common Mistakes
- Assessing the whole company instead of one workflow. Company-wide maturity scores rarely change a decision.
- Answering yes without evidence. "Our data is fine" usually means nobody has checked.
- Skipping the people questions. Most stalled projects had the data and lacked an owner.
- Treating "not ready" as "never". Most gaps take days or weeks to fix.
- Buying a platform before answering Q1. Tools don't create a use case.
When to Get Outside Help
This checklist works as a self-assessment for one workflow. Outside help earns its cost when you have several candidate projects and need to rank them, when the data situation is genuinely unknown, or when a board needs an independent view. Our guide to AI consulting vs in-house development explains when an assessment is worth paying for.
Conclusion
AI readiness comes down to one workflow, a measure, accessible data, an owner, and clear rules on what the AI may do. Answer the 20 questions honestly, treat the three critical ones as stop signs, and fix the gaps before spending on a build. Most organisations are closer to ready than they think. They just haven't written the answers down.
Schedule a strategy call with Cogniq AI to work through the checklist for your first project, or explore our custom AI development services.



