This is the 17-point checklist from The Groundwork Advantage, our C-suite handbook on building AI that creates lasting value. It covers what an AI initiative depends on before the build starts, while it’s being built, and after it ships. You can work through it with your own leadership team and you’ll know whether your next use case is starting from a foundation or from scratch.
A checklist is only useful if a failed item tells you something. Every check below earns its place because of what a “no” reveals, and each one maps to a decision that sits with leadership rather than with a delivery team.
Ten minutes of honest answers here is worth more than a month of vendor evaluation, because most of what determines whether an AI initiative produces value is settled before anyone chooses a model.
Before anything is built
Five checks establish whether your organisation is in a state where any AI tool could succeed. Until these are true, the choice of tool barely matters.
1. A named business owner for AI decisions is in place, with the authority to prioritise and say no. A committee doesn’t count, and neither does IT by default. Everything downstream depends on someone being able to decline a use case, and a default mandate is how initiatives end up owned by nobody.
2. A current inventory of AI tools in use across the organisation exists, including unsanctioned ones. If that list would take a month to produce, the length of the delay is the measure of how much AI activity is running outside your oversight, and of how far your data has already travelled.
3. Data ownership is understood: you know where your data lives, who can access it, and what it flows into. Ownership determines whether you control your own AI trajectory or depend on a vendor’s pricing, roadmap and continued existence.
4. GDPR and AI Act requirements are treated as design inputs for any new system. Retrofitting compliance is expensive, and the requirements of GDPR and the AI Act, explainability, data lineage and human oversight, are the same properties that make a system trustworthy enough to scale. Deloitte’s 2026 research found the gap between piloting agentic AI at 38 per cent and running it in production at 11 per cent comes down to exactly this: governance and integration.
5. Existing systems of record are treated as assets to connect to. The ERP, the CRM and the document management system can usually stay where they are. What changes is the layer above them.
Choosing the first use case
Four checks govern what you build first and what you expect it to be worth. This is where most initiatives are won or lost, and it happens before a line of code is written.
6. Candidate use cases are scored on all three factors together: volume, friction and data readiness. Volume asks how often the process happens. Friction asks how much of the current effort is retrieval, formatting and chasing information rather than genuine judgement. Data readiness asks how much work is required before this specific use case can run on governed data. A use case strong on one factor and weak on the others is a trap.
7. The first use case scores well on data readiness, even if others score higher on ambition. The purpose of the first build is to prove the foundation works end to end, with a result the organisation can trust. That’s worth more at this stage than picking the highest-potential candidate and discovering the data behind it wasn’t ready.
8. A definition of success is agreed before the build starts. Agreeing it afterwards means agreeing it with the result already visible, which is how projects get graded on a curve.
9. Expected impact is estimated in writing across three separate categories. Direct efficiency, quality improvement and strategic option value. RAND’s research into AI project failures found leadership specifying the wrong problem among the most cited causes, and a written estimate is the cheapest way to find out early that you have.
Building it
Four checks determine whether the build produces a reusable foundation or another silo. Responsibility is the theme running through all of them.
10. The business owner defines the objective and validates the outcome, with IT and data teams responsible for what the foundation requires. Access, governance and integration with existing systems sit with IT. Whether the outcome matters commercially sits with the business.
11. The foundation is built to be reused: governance, access controls and data indexing that the next use case inherits. This is the single item that decides whether your fifth use case is cheaper than your first.
12. Any implementation partner is judged on how much of what they build is reusable, as well as on whether the first use case works. The value a partner provides is shifting away from writing the code and towards making sure what gets built is architected correctly, governed properly, and connected to something that supports what you build next.
13. The initiative is visible across the organisation: objective, expected impact and outcome communicated honestly, including if it underperforms. AI initiatives owned entirely by IT, with the business consulted once there’s something to demo, produce exactly the tool sprawl described in check two. If people don’t feel ownership of the sanctioned path, individual teams route around it.
Measuring and scaling
Four checks tell you whether the investment is compounding. They’re the ones most often skipped, because by this point the system is live and attention has moved on.
14. The outcome is measured directly against the written impact estimate. A general sense that things are going well doesn’t count. If the estimate was wrong, that’s useful calibration for the next one.
15. Results are reported before the next use case is chosen, including shortfalls. Quiet absorption of a disappointing result is how estimates never get calibrated, and how the same mistake gets funded twice.
16. Each new use case starts from the application layer, inheriting the foundation already in place. Gartner’s 2026 survey of 1,303 organisations found only 22 per cent have successfully scaled AI across multiple business units, which is what this check is designed to prevent.
17. The trend is checked over time: each use case should be cheaper and faster to deliver than the one before it. If it isn’t, the groundwork is being rebuilt somewhere. The same Gartner research found that high performers, who track returns and reallocate away from underperforming initiatives, reported positive returns on 81 per cent of their AI initiatives.
How to use the checklist
Treat the four stages as gates rather than as a scorecard, and don’t start the next stage until the current one has produced its output.
| Stage | The decision it settles | Who owns it | How you know it’s done |
|---|---|---|---|
| Before anything is built | Whether any AI tool could succeed here | Named business owner | Ownership, tool inventory and data ownership are all documented |
| Choosing the first use case | What to build first, and what it should be worth | Business owner, with the data team | A written, quantified case for the first build |
| Building it | How the foundation gets built and who is accountable | Business owner for the objective, IT and data for the foundation | Outcome measured against the written estimate |
| Measuring and scaling | Whether the investment compounds or restarts | Leadership team | Each new use case is a smaller project than the last |
If most items are easy to answer, you’re closer to ready than most organisations. If several are genuinely uncertain, that uncertainty is your starting point and the first thing to resolve.
Key takeaways
- A named business owner with the authority to say no is the first check, because every later decision depends on someone holding that authority.
- An inventory of the AI tools already in use, sanctioned or not, tells you how far your data has already travelled before you add anything new.
- The first use case should be chosen for the quality of its data rather than the scale of its ambition, because its job is to prove the foundation works end to end.
- Expected impact belongs in writing, in three separate categories, before the build starts, so there’s something concrete to measure against afterwards.
- If each new use case isn’t cheaper and faster to deliver than the last, the groundwork is being rebuilt somewhere and the investment isn’t compounding.
Get the full handbook
Each item on this checklist is explained in The Groundwork Advantage, our free C-suite handbook on building AI that creates lasting value. It covers why the first use case should be chosen for its data, why impact gets estimated in three categories instead of one, what each phase has to deliver before the next begins, and which decisions stay yours whoever builds the thing.
Download The Groundwork Advantage
If you’d like to pressure-test the checklist against your own situation, we run the same assessment with organisations preparing to scale AI. Get in touch and we’ll work through it with you.
