Ask three vendors what an AI initiative costs and you will get three answers that differ by a factor of five. Ask an internal team and you will get a number that covers the obvious parts and misses the rest. The result: budgets that are either so padded they never get approved, or so lean they run out mid-project.

After delivering 25+ DACH enterprise engagements, we can provide a more honest answer. Not a range so wide it means nothing, but a structural breakdown of where the money actually goes — and which cost categories organisations consistently underestimate.

The cost anatomy

Every AI initiative has five cost layers. Most budget proposals cover two of them.

Layer 1: Discovery and scoping (€8,000–25,000)

Before any engineering begins, you need to establish workflow readiness, validate data accessibility, and define success metrics. This is the work that determines whether the initiative should proceed at all.

Skipping discovery is the most expensive mistake in AI deployment. A €15,000 discovery engagement that kills a doomed initiative saves €80,000. A discovery engagement that sharpens scope and de-risks the build pays for itself in the first sprint.

Organisations that skip discovery and go straight to build typically spend 40–60% more overall, because they discover scope problems, data problems, and integration problems during engineering — when changes are expensive.

Layer 2: Engineering and build (€20,000–120,000)

This is the cost everyone budgets for. It covers model selection and configuration, prompt engineering, workflow orchestration, API development, and testing.

The range is wide because it depends on complexity. A single-workflow Level 1 deployment with clean data and modern APIs: €20,000–40,000. The same workflow with legacy integration, custom data pipelines, and complex business logic: €50,000–80,000. A multi-workflow Level 2 deployment with cross-system orchestration: €80,000–120,000.

The key driver is not AI complexity — it is integration complexity. The model configuration for most Mittelstand use cases is straightforward. The engineering effort goes into getting data from legacy systems, handling edge cases in business logic, and building reliable pipelines. This is why data accessibility is the strongest predictor of engineering cost.

Layer 3: Infrastructure (€200–3,000/month)

Cloud infrastructure for AI workloads is cheaper than most organisations expect. A typical Level 1 deployment running on managed cloud services: €500–1,500 per month. This covers compute, storage, API calls to model providers, monitoring, and logging.

The cost scales with volume, not complexity. A workflow processing 1,000 units per week costs roughly the same as one processing 100 units per week — the per-unit cost at AI-native scale is negligible compared to the human cost it replaces.

Where infrastructure costs surprise organisations: data storage and transfer from legacy systems. If the data pipeline requires significant staging, transformation, or historical data loading, infrastructure costs can spike during the build phase.

Layer 4: Integration and data access (€5,000–40,000)

This is the layer that blows budgets. Integration work is the bridge between the AI workflow and the systems it needs to read from and write to. For organisations with modern, API-first architectures: minimal cost, included in the engineering layer. For organisations with legacy systems: this becomes a project within the project.

Common integration scenarios and their costs: API wrapper around an existing database view (€5,000–10,000). Data pipeline from a legacy ERP with batch exports (€10,000–20,000). Real-time event pipeline from a system without native streaming (€15,000–30,000). Multi-system integration with data reconciliation (€25,000–40,000).

The critical insight: integration cost is knowable before the project starts. A readiness assessment reveals the integration landscape within days, not weeks. Organisations that discover integration complexity during the build phase pay 2–3x what they would have paid with upfront planning.

Layer 5: Change management (€5,000–20,000)

The cost that almost no one budgets for — and the reason many technically successful deployments fail in practice. Change management covers team training, process documentation, KPI redefinition, and the ongoing support required to transition a team from a manual workflow to an AI-assisted one.

Without it, expect the pattern described in operating model clarity: the AI system works technically but is unused within 60 days because no one redefined how the team operates.

Total cost by deployment type

Combining all five layers:

Level 1 — single workflow, modern stack: €40,000–80,000 total, €500–1,500/month ongoing. Payback typically 4–8 months.

Level 1 — single workflow, legacy integration: €70,000–150,000 total, €800–2,000/month ongoing. Payback typically 8–14 months. The premium is almost entirely integration cost.

Level 2 — multi-workflow, cross-system: €150,000–300,000 total, €1,500–3,000/month ongoing. Payback typically 10–18 months. Should only be attempted after a successful Level 1 deployment.

Where organisations overspend

Over-engineering the AI layer. The model and prompt engineering for most Mittelstand use cases costs €10,000–25,000. If you are spending more, you are either solving a problem that does not need AI, or building custom models when pre-trained ones suffice. See build vs. buy for the decision framework.

Under-investing in discovery. The correlation is consistent: organisations that invest 10–15% of total budget in discovery and scoping spend 20–30% less overall than those that skip it. Discovery prevents the most expensive kind of rework — changing direction mid-build.

Ignoring change management. A €100,000 technical deployment with €0 change management budget delivers less value than a €80,000 deployment with €15,000 of structured adoption work. The technology only generates value when people use it.

How to budget

Start with the diagnostic to assess your readiness across all six dimensions. The dimension scores predict which cost layers will be significant:

Strong workflow readiness and data accessibility → budget toward the lower end of engineering, minimal integration cost. Weak data accessibility → budget 30–40% of total for integration work. No operating model clarity → explicitly budget for change management.

Then build the business case using the template structure — with Phase 1 gates that limit capital at risk.

The goal is not to minimise cost. It is to spend knowingly, with clear expectations of what each layer delivers and when the investment pays back.