AI Readiness Checklist: 12 Questions Before You Scale AI

Most organisations have now experimented with AI in some form. Staff use generative AI tools, a team has built a proof of concept, or a supplier has added AI features to a familiar product. Many of these experiments look promising. Far fewer turn into reliable, valuable services used across the organisation.

The gap between a successful pilot and lasting value is usually not the model. It is everything around it: the data, the skills, the infrastructure, the governance and the way the organisation adopts change. This checklist sets out twelve questions to ask before you scale.

Strategy and value

1. Do we know which problems AI should solve?

Scaling AI because competitors are doing it is not a strategy. The organisations that get value start with specific business problems, such as reducing time spent on routine queries, improving forecast accuracy or speeding up document review, and ask whether AI is the right tool.

Ready looks like: a short, prioritised list of use cases tied to measurable outcomes, owned by business leaders.

2. Do we know how we will measure value?

Pilots often succeed on enthusiasm. Scaling needs evidence. Agree before you start how you will measure benefit: time saved, errors reduced, revenue gained, customer satisfaction improved.

Ready looks like: a baseline and target for each use case, and a plan to measure after deployment.

3. Have we decided what to buy, adapt or build?

Many needs are best met by AI features in existing products. Some need adapting general models to your context. A few justify building something bespoke. Each option has different costs, risks and skills requirements.

Ready looks like: a clear sourcing approach for each use case, based on value, risk and capability.

Data

4. Is the data we need available and good enough?

AI depends on data. If the data needed for a use case is incomplete, inconsistent, inaccessible or poorly understood, the AI will struggle, however sophisticated the model.

Ready looks like: for each priority use case, the data has been identified, assessed for quality and is accessible under clear rules.

5. Do we know who owns the data and how it may be used?

Using data for AI raises questions of ownership, consent, privacy and intellectual property. Who can approve its use? Is personal data involved? Can it be shared with a supplier?

Ready looks like: named data owners, clear policies on permitted use and data protection reviews for uses involving personal data.

People and skills

6. Do we have the skills to build, run and use AI?

Scaling AI needs a mix of skills: people who understand the business problem, data and engineering skills, and the ability to evaluate whether AI outputs are reliable. Users also need skills to use AI tools well and to spot when they are wrong.

Ready looks like: a realistic view of skills gaps, with a plan to close them through training, hiring or partners.

7. Are people ready for the change?

AI changes how people work. Some will welcome it; others will worry about their jobs or distrust the outputs. Ignoring the people side is one of the most common reasons AI projects fail to deliver.

Ready looks like: honest communication about what AI will and will not do, involvement of affected staff in design, and support for new ways of working.

Technology and operations

8. Can we run AI reliably in production?

A model running on a data scientist's laptop is very different from a service used by thousands of people every day. Production AI needs reliable deployment, monitoring, version control and the ability to roll back when something goes wrong.

Ready looks like: repeatable pipelines for deploying and updating models, and monitoring for performance, cost and errors.

9. Will we notice when performance degrades?

AI models can drift as data and circumstances change. Outputs that were accurate at launch can become less reliable over time, often without anyone noticing.

Ready looks like: ongoing monitoring of accuracy and data quality, with clear thresholds and owners who act when they are breached.

Governance and trust

10. Do we have governance proportionate to risk?

Different uses carry different risks. Summarising internal documents is very different from screening job applicants. Governance should be proportionate: light for low-risk uses, thorough for high-risk ones.

Ready looks like: an inventory of AI uses, a simple risk-tiering approach, and named owners for higher-risk uses. Our guide to AI governance for mid-sized organisations sets out a practical starting point.

11. Can we explain and challenge AI-supported decisions?

Where AI affects customers or staff, people may need to understand how decisions were reached and to challenge them. Regulation increasingly expects this for significant decisions.

Ready looks like: human oversight for significant decisions, a way for people to query outcomes, and documentation of how each system works and was tested.

12. Are our suppliers' AI practices acceptable?

Much AI arrives through suppliers. Do you know how their AI features use your data, whether your data trains their models, and how outputs are tested?

Ready looks like: AI questions built into procurement and contract reviews.

How to use the checklist

Score each question honestly: not in place, partly in place, or in place. You do not need everything in place before doing anything. But gaps in data, production operations and governance tend to block scaling, so prioritise those for the use cases that matter most.

A fuller picture

The AI Readiness assessment covers these themes in more depth across six areas: strategy and vision, data foundations, talent and skills, technology and infrastructure, ethics and governance, and adoption and value.

Updated October 2026: for organisations building their own machine learning models, our new MLOps assessment looks at the engineering practices behind questions 8 and 9: reproducible experiments, data and feature management, deployment, production monitoring, model governance and platform skills.

Frequently asked questions

How long does it take to become AI-ready?

It depends on your starting point. Organisations with strong data foundations can scale selected use cases within months. Those with fragmented data and limited skills may need a year or more of foundational work.

Should we wait until we are fully ready?

No. Start with low-risk, high-value use cases that suit your current foundations, and build capability as you go. Use the checklist to avoid scaling uses that your foundations cannot yet support.

Who should own AI readiness?

A senior leader should own the overall approach, with business owners for each use case and support from technology, data, risk and HR.