How to check AI readiness on one real process

Is our company ready for AI?

Ask the question about a process and the answer becomes concrete. Ready for AI in general means little: a company may have good IT and clear rules and still lack the data for the process that matters. Readiness is a list of conditions for one process, each either met or not, with a name next to every gap.

Data is where most checks find their gaps. In ServiceNow's survey published on 14 July 2026, 72% of German companies named data accuracy, data access and data management as the biggest hurdles, and only 16% had established test, audit and risk processes for AI. In Bitkom's survey published on 14 September 2026, 40% of companies that do not use AI named missing data, and 50% of companies that use it named data preparation as a large cost item.

Should a readiness check give a maturity score or follow a real case?

A maturity model places the whole company on a scale, usually from first experiments to AI across the business. That is useful for comparing yourself with others and for a first conversation in management. It does not tell you whether the invoices in your mailbox can be read, whether the ERP has an interface the system may use, or who will approve the results.

A sample case does. Take one real, recent case of the process you have in mind, an ordinary one rather than an exception, and follow it from the moment it arrives to the moment it is closed. Every system it touches, every document someone opens, every decision and every person involved becomes a line on the checklist. Then check the same lines on a set of past cases with known outcomes, because a single case can be unusual.

How do we run the check on one process?

1. 1

   Choose the process

   Pick a candidate with many similar cases and a result a person can check quickly. If several teams compete for the first place, score their proposals on the same questions first.

2. 2

   Pick a sample case and the people who know it

   One recent ordinary case, the person who handled it, the process owner and someone from IT who knows the systems. Do not start with the hardest exception.

3. 3

   Follow the case from start to finish

   Write down where it arrived, which documents and systems were used, what was looked up, what was decided and by whom, and where the result went.

4. 4

   Check each step against the checklist

   For every step, ask whether the data is digital, whether a system may read it, whether there is an interface, whether there is a rule for what the system may do, and who would check the result.

5. 5

   Collect past cases with known outcomes

   These become the test set later. If nobody can say what the right outcome of a past case was, quality cannot be measured yet, and that is a gap in its own right.

6. 6

   Write the gap list

   Each gap gets an owner and a decision: fix it first, work around it, or choose another process.

What belongs on an AI readiness checklist?

The checklist has five areas: skills, data, systems, rules and people. Answer each question for the sample case alone. A line only counts as met when someone can show it. Assuming it is true is not enough.

Some uses are high-risk under the AI Act, among them screening and ranking job applicants and credit scoring, both listed in Annex III. A company that deploys such a system takes on duties of its own, including human oversight by competent people, keeping the logs the system generates and informing workers' representatives before using it at the workplace. We do not automate such decisions without people. Whether a use falls under Annex III is for your company to decide, advised by its data protection officer and lawyers.

How good are our data quality and availability?

Data does not have to be perfect to be good enough for AI. It covers the cases the system will see, has an owner, can be reached with the right permissions, and lets you tell the current version from older ones. You do not need a data lake or a company-wide data project before the first process. You need the sources this process uses, in a usable state. Check four things on the sample case and the past cases:

• Coverage: do the past cases include the variants that occur, such as different suppliers, formats and languages, and the difficult cases as well as the easy ones?
• Correctness: where the outcome of a past case is recorded, is it right? A test set built on wrong outcomes measures the wrong thing.
• Currency: for questions answered from documents, which version is valid, and who removes outdated ones?
• Access: can a service account read the data with permissions that will still apply in production, rather than through a one-off export?

Finance teams feel this most. In KPMG's Global AI in Finance report of 22 May 2026, 36% of finance leaders named data quality as the biggest challenge for AI in finance. The guides on data readiness for AI projects and on preparing sources for RAG, where answers are drawn from your own documents with the source shown, go into each point.

Is there a free AI readiness check?

Yes. Mittelstand-Digital Zentren run free introductory AI consultations, and chambers of industry and commerce offer their own checks, for example the KI-Reifegrad-Check of IHK Schwaben and the KI-Check of IHK Stuttgart. Ask your chamber about the terms. They make a sensible start. They are neutral and structured, and they get management talking about AI in shared terms.

Know what such a check is designed to do. IHK Schwaben describes its check as a first structured assessment and says it is not consulting, legal advice or an audit. By design, a general check like this leaves out what only shows up on your own process:

• Whether your own cases can be handled, because no sample case is followed through your systems.
• Access and permissions in your ERP, CRM, document or ticket systems.
• Which infrastructure fits your data: a cloud service, your own cloud account or your own servers.
• How output quality will be measured, and how much review it will need.
• Building and running the system afterwards.

Use both: the chamber check for orientation, the sample-case check for the process you want to change. Our free 30-minute call with an engineer needs no access to your systems or data and can go through the checklist above for one process with you. If your staff already use AI tools, the free AI usage review adds one page on AI spend and quality, built from exports you choose to send.

What comes after the readiness check?

The gap list decides the next step, and it usually points to one of three outcomes:

• Ready, or close to it: measure today's process, agree acceptance criteria and start a pilot on your own cases.
• Ready after fixes: close the gaps that block a pilot first, usually access to a system, an owner for a data source or approval of the data flow, and keep the rest for the step into production.
• Not this process: if the data is missing or nobody can say what a right result looks like, choose the next candidate rather than forcing it.

The guide on introducing AI in your company describes the stages from here to a running system.

AI strategy and rolloutData integration for AIHeads of Digital and AI officers