Most AI readiness assessments ask the wrong questions. They measure how many data scientists you employ, whether you have a "data strategy," and how your cloud infrastructure scores on some maturity model. Then they produce a 40-page PDF that sits in a SharePoint folder.

That is not readiness. That is inventory.

After 25+ AI engagements with DACH enterprises — insurance, e-mobility, industrial, retail — we have a clearer picture of what actually predicts whether an AI initiative will reach production. It is not what most frameworks measure.

The readiness gap nobody talks about

The typical Mittelstand company in our pipeline looks like this: €30–200M revenue, 200–2,000 employees, one or two AI pilots that produced impressive demos but never reached production. The CTO has a proof of concept. The board has a mandate. And somewhere in between, the initiative stalled.

The gap is not technical. The models work. The APIs exist. The cloud is provisioned. The gap is operational: the organisation cannot absorb AI into its daily workflows without breaking something — a process, a compliance requirement, a team's trust.

This is what readiness actually means: the organisation's capacity to put AI into production and keep it running.

Not "are we advanced enough for AI?" but "can our organisation actually operationalise a model in a real workflow, this quarter, without a two-year transformation programme?"

That is a very different question. And it produces very different answers.

Why most readiness frameworks fail the Mittelstand

Enterprise readiness frameworks — McKinsey's AI maturity model, Gartner's assessments, the various consulting scorecards — were designed for DAX-40 companies with dedicated AI teams and eight-figure transformation budgets. They measure capabilities that are useful at scale but irrelevant for a company trying to deploy its first production AI workflow.

They ask: do you have a centralised data platform? Do you have MLOps infrastructure? Do you have an AI Centre of Excellence?

For a €50M industrial supplier with 400 employees, the honest answer to all three is "no." But that does not mean they are not ready. It means they need a different framework — one that measures operational readiness rather than organisational maturity.

The distinction matters. Maturity is about where you are on a five-point scale. Readiness is about whether you can execute a specific initiative in the next 90 days. The first produces reports. The second produces results.

Six dimensions of operational AI readiness

In The AI Operating System, we define readiness across six dimensions. Not because six is a magic number, but because these are the six areas where we have seen engagements fail or succeed.

1. Workflow readiness

Can you name, in writing, the three workflows with the highest AI-addressable volume? Not "we could use AI in customer service" — but "our claims triage team processes 1,200 cases per week, 40% of which follow a pattern that a classification model could handle with >90% accuracy."

If you cannot write that sentence for at least one workflow, you are not ready to build. You are ready for Discovery.

Workflow readiness is the foundation because it determines everything downstream: what data you need, what compliance applies, who needs to be involved, and what "success" looks like in measurable terms.

The most common mistake we see: companies define their target workflow too broadly. "Automate customer service" is not a workflow. "Classify incoming support tickets by urgency and route to the correct department" is a workflow. The narrower and more specific, the higher the probability of production deployment.

What to check: documented process maps, measurable throughput, a clear definition of "good output" that a model can be evaluated against.

2. Data accessibility

Not data quality — data accessibility. The question is not "is your data clean?" (it is not; nobody's is). The question is: can you get the data from where it lives to where a model needs it, in under two weeks, without a six-month integration project?

Most Mittelstand companies have their critical data locked in ERP systems (SAP, Microsoft Dynamics, Sage), document management systems, or — more often than anyone admits — Excel files on shared drives. The readiness question is whether there is a viable path to pipe that data into an AI workflow without rebuilding your entire data infrastructure first.

This is where many engagements die silently. The workflow is clear, the sponsor is committed, the team is available — and then IT estimates eight months to build the data pipeline. The project loses momentum and quietly disappears from the quarterly review.

The antidote is pragmatism. For a first AI workflow, you do not need a data lake. You need a CSV export that runs nightly, an API endpoint that returns the last 90 days of transactions, or a document folder that a retrieval pipeline can index. Perfect data architecture is a goal for year two. Functional data access is a requirement for week one.

What to check: API availability on core systems, data export capabilities, existence of a staging or sandbox environment where data can flow without touching production.

3. Decision authority

Who can say "yes, deploy this to production"? In our experience, this is the single strongest predictor of success. If the answer is "the board, after a three-month approval process," you will spend more time in committee than in development.

AI readiness requires an exec sponsor with budget authority and operational mandate. Not a committee. Not a working group. One person who can allocate €30–150K, assign team members, and approve production deployment within days, not quarters.

In Mittelstand companies, this is often the Geschäftsführer personally — which is actually an advantage. The shorter the decision chain, the faster the deployment. A €80M family business where the CEO says "let's do this" will outperform a €500M corporate where AI initiatives need sign-off from four directors and a steering committee.

What to check: is there an exec sponsor named, with budget pre-approved? Can they make go/no-go decisions without escalation?

4. Compliance posture

Under DSGVO, the EU AI Act, and sector-specific regulation (BaFin, insurance supervisory authorities), AI in production is not a technical decision — it is a compliance decision. The readiness question is not "are we compliant?" but "do we know what compliance requires for our specific use case?"

Most companies discover compliance requirements after they have built the model. This is backwards. A 2-hour session with your DPO and legal team, using a DPIA template, will tell you whether your target workflow is high-risk, limited-risk, or minimal-risk under the AI Act — and what documentation you need before deployment.

The good news: for most Mittelstand AI workflows — document classification, process automation, internal knowledge retrieval — the compliance burden is manageable. The EU AI Act reserves its heaviest requirements for high-risk use cases (hiring decisions, credit scoring, critical infrastructure). A workflow that classifies incoming invoices is not high-risk. But you need to know that, documented, before you build.

What to check: has a Data Protection Impact Assessment been started for the target workflow? Does your DPO know this project exists? Has someone mapped the use case against EU AI Act risk categories?

5. Team capacity

Not "do you have AI engineers?" — but "can you free up 2–3 people from their current work for 6–13 weeks?" AI implementation is not something that happens alongside the day job. It requires dedicated time from the people who understand the workflow being automated.

The DIWM (Do It With Me) model works because your team does the implementation with coaching — but they need actual time allocated for it. The DIFM (Do It For Me) model requires less team time but still needs subject-matter experts available for requirements, testing, and validation.

We have seen technically excellent projects fail because the domain expert was simultaneously running three other projects and could only join weekly check-ins. AI needs feedback loops that run in days, not weeks. That only works if the people providing feedback have the bandwidth to provide it.

What to check: can you name the 2–3 team members who will be involved? Has their manager agreed to reduce their other workload? Is this officially part of their objectives for the quarter?

6. Operating model clarity

This is the dimension most frameworks miss entirely. Once the AI workflow is in production, who operates it? Who monitors for drift? Who handles edge cases? Who decides when to retrain?

AI is not a feature you ship and forget. It is an operational capability that needs ongoing attention — not daily, but weekly. If you do not have clarity on who will own the AI workflow post-deployment, you will build something that decays within three months.

The good news: for most Mittelstand use cases, the operating burden is modest. A weekly check of model accuracy, a monthly review of edge cases, and a quarterly decision on whether to retrain or expand scope. This can be done by the same team that owns the underlying business process — it does not require a dedicated ML engineering team.

But someone needs to be named. If nobody owns it, nobody maintains it. And unmaintained AI is worse than no AI, because it degrades silently.

What to check: is there a named owner for the target workflow today? Will they own the AI-enhanced version too? Do they understand what "operating an AI workflow" entails in practice?

The readiness threshold: what you actually need

You do not need a perfect score on all six dimensions. In our experience, the threshold for a successful first engagement is:

  • Dimensions 1 + 3 are non-negotiable. You need at least one named workflow with measurable volume, and an exec sponsor with budget authority. Without these, nothing else matters.
  • Dimensions 2 + 4 need a viable path, not a finished state. Data does not need to be clean. Compliance does not need to be complete. But both need a path that can be walked in weeks, not years.
  • Dimensions 5 + 6 can be developed during the engagement, especially with Discovery (2 weeks, €10K), which is specifically designed to establish these foundations.

This threshold is deliberately lower than what most frameworks suggest. We have found that companies over-prepare and under-execute. The goal is not to be ready for every possible AI use case. The goal is to be ready for one — the one that will teach your organisation how AI works in practice.

Three levels of readiness — and what to do at each

Based on the six dimensions, companies typically fall into one of three levels:

Level 1: Not yet ready

You have interest but no specific workflow, no exec sponsor with budget authority, or fundamental blockers in data access or compliance. What to do: start with education and alignment. Read The AI Operating System. Run an internal workshop to identify candidate workflows. Get your DPO involved early. Come back when you have a named workflow and a named sponsor.

Level 2: Ready for Discovery

You have a target workflow and an exec sponsor, but need to validate feasibility, define scope, and establish the operational foundations. What to do: Discovery is a 2-week engagement (€10K) that produces a validated scope, a technical feasibility assessment, and a deployment plan. It is the lowest-risk way to move from "we think this could work" to "here is exactly what we will build, how long it will take, and what it will cost."

Level 3: Ready to Build

You have a validated workflow, accessible data, exec authority, compliance clarity, team capacity, and operational ownership. What to do: move to Accelerator (6 weeks, €30K) or OS Build (13 weeks, €75–150K), depending on complexity.

What if you are not ready?

Then you know exactly what to fix. That is the value of an honest assessment: it turns "we should do something with AI" into "we need to solve X, Y, and Z before we can build."

The worst outcome is not failing the readiness check. The worst outcome is skipping the check, spending €150K on an OS Build, and discovering at week 8 that nobody has authority to deploy to production.

We would rather tell you "come back in three months" than take your money for an engagement that will stall. That is not altruism — it is self-interest. Failed engagements do not produce case studies, referrals, or expansion revenue. Successful ones do.

The 10-minute self-assessment

If you want a structured self-assessment, our AI Operating Diagnostic scores your organisation across all six dimensions in about 10 minutes. It is free, requires no login, and gives you a personalised next-step recommendation.

For a deeper assessment with our team, book a Fit Call. We will tell you honestly whether you are ready for Discovery, Accelerator, or OS Build — or whether you need to do groundwork first. No pitch deck. No pressure. Just clarity on where you stand and what comes next.

This article is based on the methodology in The AI Operating System by Andreas Anding. The framework has been tested across 25+ DACH enterprise engagements in insurance, e-mobility, industrial manufacturing, and retail.