NSolutions
EN PL

Free checklist

Executive AI Readiness Checklist

10 questions to answer before you commit budget to an AI project. Plan about 45 minutes with the right people in the room.

Start the checklist

For CTOs, CDOs, COOs and CEOs. A self-assessment, not an audit. Updated October 2026.

How to use it

  • WhoFill it in together: the business owner of the use case, the architecture or IT lead, and whoever owns the data. Three people, about 45 minutes.
  • ScopeAnswer for one concrete use case, or for the programme you are about to fund. Answers for “the company in general” hide the gaps.
  • EvidenceMark Ready only if you can point to a document, dataset, contract or named person. “We are working on it” is Partly.
  • ScoringNot started 0, Partly 1, Ready 2. Maximum 20. If question 1, 3 or 4 scores 0, do not start building, whatever the total.

Question 1 · Strategy Gate: a score of 0 stops the build

Can you state the business outcome in one sentence, with a number and an owner?

Why it matters
AI projects without a measured outcome end as demos. A target number is what lets you decide later whether to scale, change direction or stop.
Ready when
You have one sentence such as “Cut invoice processing from 9 days to 3 by the end of Q3, owned by the CFO.” The baseline is measured today, and a named executive has signed off the target.
Red flags
The goal is “use AI” or “be innovative”. The baseline is a guess. The sponsor is the IT department.
First move
Write a one-page case: the process, its current cost, time and error rate, the target, the owner and the date. If you cannot get the baseline from your own data, measuring it is the first project.
Your answer

Question 2 · Strategy

Have you compared candidate use cases and picked one or two to start with?

Why it matters
Most companies have a dozen ideas and capacity for one or two. Choosing by enthusiasm instead of evidence spreads the budget thin.
Ready when
A list of 10 to 20 candidates scored on business value, data availability, effort to change the process and risk. The first one or two are chosen, and the rest are parked on purpose.
Red flags
Each department runs its own pilot. Nobody can say which ideas were rejected and why. The first use case was picked because a vendor demoed it.
First move
Run a half-day workshop with business and IT. Score each idea 1 to 5 on value and on feasibility, and start with one that is high on both and can show a result within 90 days.
Your answer

Question 3 · Data Gate: a score of 0 stops the build

Does the data for your first use case exist, can you reach it, and do you know its quality?

Why it matters
Data work is usually the largest and least visible part of the effort. A pilot built on data nobody has inspected fails in production, not in the demo.
Ready when
A short inventory exists for the use case: where the data lives, who owns it, how it is accessed. A real sample (a few hundred to a few thousand records) has been profiled for completeness, duplicates, errors and, where needed, labels.
Red flags
“The data is in several spreadsheets and the ERP.” Nobody has opened it. Access depends on one person. The records needed to answer the question are not stored at all.
First move
Pull a real sample and profile it before you choose a vendor or a model. Count missing fields, duplicates and conflicting values. The result tells you whether to proceed, clean first, or pick another use case.
Your answer

Question 4 · Data and security Gate: a score of 0 stops the build

Are you allowed to use this data with AI, and are the security risks covered?

Why it matters
Customer, employee and contract data carry legal limits (GDPR, contracts, sector rules) and new attack routes, such as data leaking through prompts or hostile instructions hidden in documents.
Ready when
Data is classified (public, internal, confidential, personal, special category). The legal basis for using it is documented, a data protection impact assessment is done where required, and where the data is processed (country, vendor, sub-processors) is known and acceptable. Access is limited by role and use is logged.
Red flags
Staff paste customer data into public chat tools. Nobody has read the vendor’s terms on training with your data. The security review is planned for “after the pilot”.
First move
Issue a one-page rule today on which data classes may go into which tools. Then ask your data protection officer and security lead to review the first use case before any pilot starts.
Your answer

Question 5 · Architecture

Will the solution connect to where the work happens, and does it fit your target architecture?

Why it matters
A tool that people must copy and paste into does not change the process. Integration, identity and logging often cost more than the AI component itself.
Ready when
The process is mapped end to end, the systems it touches are listed (CRM, ERP, document store, ticketing), and the interfaces exist or have an owner and an estimate. Single sign-on, event logging and the security review path are agreed with IT.
Red flags
A demo that runs on exported files. “Integration comes in phase two.” Nobody from the team that runs your core systems has seen the design.
First move
Draw the process with every system and hand-off on one page, mark where the AI result must land, and ask the owner of each system what it takes to connect. Treat the answers as scope.
Your answer

Question 6 · Vendors and cost

Do you know what to build, buy or partner for, and what it costs at production volume?

Why it matters
Demo prices rarely survive production volumes, and switching vendors later is expensive if your data, prompts and evaluation work are locked in.
Ready when
There is a decision per component (build, buy, partner) with reasons. The cost at expected production volume is modelled, including usage-based fees. The contract lets you export your data and evaluation sets and replace the model or the vendor.
Red flags
Pricing is based on the pilot. The choice rests on one vendor’s presentation. There are no exit terms, no data portability clause, and no second option was considered.
First move
Put three things into the next RFP or vendor call: a price at your real production volume, export of your data and evaluation sets, and how you would switch model or vendor. Compare at least two options.
Your answer

Question 7 · Operations

Who will run it after go-live, and what happens when it gets worse?

Why it matters
AI results change when the data, the users’ behaviour or the underlying model changes. Without an owner and monitoring, quality drops silently.
Ready when
A named service owner and a support route. Monitoring of quality, cost and usage with alert thresholds. A procedure for updating or retraining, and a way to switch the feature off or roll back. “Done” means running in production, not demo accepted.
Red flags
The project team disbands at go-live. There is no budget line for running costs. Nobody checks outputs after launch.
First move
Before building, write the skeleton of the run-book: owner, metrics, alert levels, what to do on failure, who pays for running costs. If you cannot fill it in, the use case is not ready for production.
Your answer

Question 8 · Quality and risk

How will you measure whether the output is good enough, and do you know which AI Act rules apply?

Why it matters
“It looks good” is not a quality standard. And regulation depends on what the system does, not on the technology behind it.
Ready when
A test set of real cases (start with 50 to 200) with agreed correct answers and a minimum pass rate set before the pilot. Human review wherever a wrong answer is costly. Each use case is classified under the AI Act, and staff who use AI have been trained.
AI Act, position in October 2026
Prohibited practices and the duty to support staff AI literacy have applied since 2 February 2025. Transparency duties, such as telling people they are dealing with an AI system, apply from 2 August 2026. Obligations for high-risk systems (for example credit scoring of individuals or recruitment) apply from 2 December 2027, and for AI built into regulated products such as medical devices from 2 August 2028, following the AI omnibus adopted in June 2026. Confirm the current position with legal counsel.
Red flags
Quality is judged by whoever demoed it. Nobody has checked whether the use case touches recruitment, credit, insurance pricing, health or another regulated area. Outputs are not logged.
First move
Build the test set from your own past cases before the pilot and fix the pass mark. Then ask legal to classify each use case and list what applies.
Your answer

Question 9 · People

Will the people whose work changes actually use it?

Why it matters
Adoption, not accuracy, decides whether the benefits show up. People avoid tools they were not asked about, that add steps, or that threaten their role.
Ready when
At least five future users were interviewed or observed before design. The managers of the affected teams sponsor the change. It is clear what changes in daily work, what stops, and how performance is judged. Training and a champion in each team are planned.
Red flags
The tool is announced after it is built. Staff worry about their jobs and nobody has addressed it. Success is measured by logins, not by the business metric.
First move
Spend a day with the team that will use it. Write down each step they take today and what changes. Name one champion per team and agree what the saved time will be used for.
Your answer

Question 10 · Funding

Is the budget sized for the whole lifecycle, with clear points to scale or stop?

Why it matters
The pilot is the cheap part. Data preparation, integration, security, change management and running costs often exceed it.
Ready when
A budget that covers the pilot, the production build and 2 to 3 years of operation, including data preparation, integration, licences or usage fees, support and training. Funding is released in stages against three gates: pilot result, production readiness, scale-up. Stop criteria are written down.
Red flags
Only the pilot is funded. The business case uses the vendor’s savings claims. Nobody has said what result would make you stop.
First move
Set three gates, each with a metric and a date: pilot to production, production to scale, and a stop rule. Ask finance to review the total cost of ownership, not just the build.
Your answer

Your result

0 / 20

Book a discovery call

Want a second opinion on your result?

Bring your score and the three weakest answers to a free 30-minute discovery call. I will tell you what I would do first.

Book a discovery call

This checklist is a self-assessment, not an audit or legal advice. AI Act dates reflect the position in October 2026; confirm the current rules with legal counsel. Your answers stay in your browser and are not sent anywhere.

0 / 20 0 / 10 answered See result