Most mid-size companies do not need a 60-page maturity framework to make this call. They need to know whether to start with a production use case, a contained pilot or groundwork on their data, and which of those fits their budget this year. Long frameworks describe every dimension of maturity, but they rarely say what to build first.
A company's AI readiness depends less on its technology than on four plain facts: whether it can name a measurable process to improve, whether the data behind that process is digital and accessible, whether someone owns the result and whether its rules say which data may be sent to which tools.
This article gives you an 18-question scored checklist, explains what each score band means, recommends a first project for each band and shows a worked example. It does not cover enterprise-wide AI strategy, and it does not rank AI vendors or tools.
Key takeaways
- Readiness rests on six areas: use case and ownership, data availability, data quality and access, systems and integration, security and governance, and people and budget.
- Score each of the 18 questions 0, 1 or 2. The total, out of 36, points to one of four starting points.
- Three knockout questions override the score: no owner, no digital data, or no rules on sharing data with external tools.
- The weakest area outweighs the total. A high score with weak data still needs data work first.
- Gartner predicts that through 2026, organizations will abandon 60% of AI projects that are not supported by AI-ready data.
- A first project that fits the score usually takes 4–12 weeks and a budget of $15,000–$80,000.
How to use the checklist
Answer each question for the process you most want to improve, not for the company as a whole. Readiness differs by process: invoice handling may be ready while customer support is not. Score 0 if the statement is false, 1 if it is partly true and 2 if it is fully true. Ask two or three people from different teams to score separately, then compare. Large differences in answers are a finding in themselves.
The checklist
1. Use case and ownership
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Can you name one process that is repetitive, time-consuming and done many times a month? | No | Several candidates, none chosen | Yes, one process is chosen |
| Do you know how you would measure success, such as hours saved, turnaround time or error rate? | No | Roughly | Yes, with a current baseline |
| Is there one person in the business who owns the result and has time to test it? | No | Someone, but without time | Yes, with time agreed |
2. Data availability
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Are the documents, records or messages behind the process stored digitally? | Mostly paper or scattered | Partly | Yes |
| Can you pull 100 real examples of the inputs and the correct outputs? | No | With effort | Yes, within a week |
| Is the history long enough to show normal variation, such as a full year of records? | No | A few months | A year or more |
3. Data quality and access
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Are key fields filled in consistently, with the same formats and categories? | No | In some systems | Yes |
| Can the data be accessed through an API, an export or a database connection? | No | Manual exports only | Yes, automated |
| Is there a single agreed source for each key record, such as customers or products? | No | Partly | Yes |
4. Systems and integration
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Do the systems involved have APIs or integration options? | No | Some | Yes |
| Can results be written back into the tool people already use, such as the CRM or ticketing system? | No | With custom work | Yes |
| Is there someone, internal or external, who can build and maintain integrations? | No | Occasionally | Yes |
5. Security and governance
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Do you have written rules on which data may be sent to external AI tools? | No | Informal rules | Yes, written and known |
| Do you know which regulations apply to the data, such as GDPR, HIPAA or sector rules? | No | Partly | Yes |
| Is there a process for a person to review AI outputs before they reach customers or decisions? | No | Ad hoc | Yes, defined |
6. People and budget
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Is there a budget for a first project, including running costs for a year? | No | Under discussion | Yes, approved |
| Will the people who do the work today take part in testing? | No | Some of them | Yes |
| Has leadership agreed how success will be judged and when? | No | Informally | Yes, with a date |
The three knockout questions
Some answers stop a project regardless of the total. If any of these scores 0, fix it before spending on development.
- No owner. Without someone accountable for the result, pilots stall after the demo.
- No digital data. A model cannot learn from or read records that exist only on paper or in people's heads. Digitization comes first.
- No data rules. Without rules on what may be shared with external tools, the first serious project will stop at the security review.
What your score means
| Score | Readiness | Recommended first step | Typical scope and budget |
|---|---|---|---|
| 0–12 | Foundations first | Data and process groundwork: choose one process, digitize its inputs, write data rules, measure the baseline | 3–6 weeks, $5,000–$20,000, or internal time |
| 13–20 | Ready for a contained pilot | A narrow pilot with an off-the-shelf tool or a simple custom assistant on one clean data source, measured against the baseline | 4–6 weeks, $10,000–$30,000 |
| 21–28 | Ready for a production use case | A proof of concept on real data, then a production build integrated with existing systems, such as document processing or a support assistant | 6–12 weeks, $30,000–$80,000 |
| 29–36 | Ready to scale | Several use cases on shared infrastructure, AI agents that act in business systems and a governance process for new projects | Ongoing team, from about $15,000 a month |
Then look at the area scores. Each area has a maximum of 6. Any area at 2 or below is the bottleneck, and the first project should either fix it or avoid depending on it. A company with a total of 24 but a data quality score of 2 is ready for a production use case only if the first project uses data that is already clean.
A worked example
A logistics company with 250 employees wants to automate the processing of delivery claims, which arrive as emails with photos and PDF documents. Three managers scored the checklist and agreed on these results.
| Area | Score | Notes |
|---|---|---|
| Use case and ownership | 6 | Claims process chosen, current handling time measured, claims manager owns it |
| Data availability | 5 | Three years of claims in the mailbox and the claims system |
| Data quality and access | 2 | Claim categories used inconsistently, no API to the claims system |
| Systems and integration | 3 | CRM has an API, the claims system does not |
| Security and governance | 3 | Informal rules, GDPR understood |
| People and budget | 5 | Budget approved, claims team will test |
| Total | 24 | Ready for a production use case, with a data bottleneck |
The total points to a production use case, and the data quality score shows where the risk sits. The recommended plan has three parts. First, two weeks to agree on claim categories and clean a sample of 500 historical claims. Second, a proof of concept that reads incoming claims, extracts the key fields and suggests a category. Third, a production build that writes results into the CRM, since that system has an API, while the claims system integration waits for the vendor. Writing data rules in parallel takes a few days and lifts the governance score before launch.
Why data readiness carries the most weight
Gartner predicts that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. In its survey of data management leaders, 63% of organizations either lacked the right data management practices for AI or were not sure whether they had them.
The broader results show the same pattern. In McKinsey's 2026 State of AI survey, nearly nine in ten organizations reported regular AI use, but only 37% attributed any EBIT impact to it, and about 6% qualified as high performers. The gap between using AI and earning from it usually sits in data, integration and ownership, which is what the checklist measures.
When data scores are low, the groundwork is often a data project rather than an AI project. Our articles on CRM data cleansing and medallion architecture show what that work involves.
Scored your company and not sure what to build first? We review your results, check the data behind your chosen process and propose a first project with a clear budget.
Explore AI servicesWhat first projects look like at each level
These .wrk projects show the kind of first step that fits each readiness level.
- Contained pilot on one clean source. A US cybersecurity media company needed a daily security newsletter assembled before a vendor subscription expired. The inputs were already digital, as RSS feeds and external articles, and editors owned the result. .wrk delivered a working AI service that classifies articles, generates briefs and drafts the newsletter in WordPress in 10 working days, with editorial review before sending. Read the Aesop editorial automation case study.
- Production use case with document data. Brandon J. Broderick, a US personal injury law firm, had a clear process, an owner and digital case files of about 250 pages each. .wrk built a tool with five specialized AI assistants that produces structured legal opinions for attorney review, at about $3–4 in model costs per case, and attorney throughput rose about 1.5 times. Read the BJB legal document processing case study.
- Data constraints decide the architecture. For a telecom operator whose call recordings could not leave its perimeter, the governance answers ruled out cloud speech APIs from the start, and the project was designed for on-premise deployment. Read the on-premise voice AI case study.
How readiness assessments go wrong
- Scoring the whole company. An average across departments hides the one process that is ready today.
- Letting one person score alone. IT, operations and the process owner often see the same situation very differently.
- Treating a low score as a no. A low score is a list of things to fix, and most of them take weeks, not years.
- Skipping the baseline. Without current numbers for time, cost or errors, nobody can show the project worked.
- Buying tools before scoring. A license bought before the data and rules are ready often goes unused.
For the next steps after the assessment, see our guide on how to implement AI in your business and our AI implementation checklist. When you are ready to test a use case on real data, our guide to running an AI proof of concept covers the process.
Frequently asked questions
What is an AI readiness assessment?
An AI readiness assessment reviews whether a company has a clear use case, accessible data, workable systems, data rules, an owner and a budget for AI. It ends with a score or rating that shows where to start and what to fix first.
How do you measure AI readiness?
Score a set of questions across use case and ownership, data availability, data quality, systems and integration, security and governance, and people and budget. Look at the total to choose a starting point, and at the weakest area to find the main risk.
What is data readiness for AI?
Data readiness means the data behind a process is digital, consistent, accessible through an API or export, available in enough volume to show normal variation and covered by rules on how it may be used. It is the area where most AI projects run into trouble.
How long does an AI readiness assessment take?
A self-assessment with this checklist takes an hour or two per process. An external assessment that checks the data, systems and security rules in detail usually takes one to three weeks.
What should a company with a low AI readiness score do?
Choose one process, digitize its inputs, write rules on which data may be shared with AI tools, measure the current baseline and name an owner. These steps usually take a few weeks and move most companies into the pilot range.
To sum up
An AI readiness assessment should end with a decision, not a maturity label. Score the process you want to improve across six areas, check the three knockout questions and let the total and the weakest area choose between groundwork, a contained pilot, a production use case or scaling.
.wrk helps companies in the US and Europe assess their readiness, choose a first AI use case and build it into production, with data work, integrations and governance included.
Want a second opinion on your readiness score? Send us your results and the process you want to improve, and we will recommend a first project with scope, timeline and budget.
Explore AI services