88% use AI. 5.5% earn money with it. The gap is what we close.
Half of your AI list falls away. The rest gets a business outcome — when you sort before the pilot, clear data readiness up front, and build adoption in parallel. AI in daily operations for industry and wholesale in the DACH mid-market. EU AI Act classification per use case, plus the matching architecture.
The jump from "AI is in the building" to "AI moves EBIT" fails for three quarters of all companies. The bottleneck is rarely the model — it sits between pilot and daily use.
At least one business function running AI — mostly isolated, without connection to core workflows.
Source: McKinsey · State of AI 2025Two thirds fail at the leap from pilot to broad adoption — the use case stays in demo mode.
Source: McKinsey · State of AI 2025Only this minority can demonstrate financial returns from AI — the rest runs without showing up in the business result.
Source: McKinsey · State of AI 2025Between "running" and "delivering EBIT" lies an 82.5 percentage-point gap. That's exactly where we operate — before and after the model.
Across mandates we see three recurring bottlenecks — all three sit outside the model. Use-Case Hygiene before the pilot, data readiness parallel to the pilot, adoption after the pilot. Two of the three determine whether it becomes EBIT-relevant at all.
Before the pilot, the list gets sorted — what's actually worth it, what's hype, what's a board member's favorite demo. ROI reality check per use case, with data readiness as the hard filter. We don't build demos without a connection to a real workflow. We cut half the list — the remaining half has a chance. What we cut is burnt pilot budget, before it gets spent.
Data quality, access, lineage, lock-ins, master data hygiene — before the model. In most B2B companies, data is distributed across ERP, CRM, PIM, Excel, and field sales notes. We map the inventory honestly, identify the data pools that are genuinely AI-ready, and prioritize use cases accordingly.
The best model without adoption is useless. We build adoption parallel to the pilot — sales champions, training, feedback loops, visible wins in daily operations. Field reps need to feel that the recommendation makes their day easier, not harder. Otherwise the demo stays pretty — and the pilot investment sits idle.
Before any production use case: which EU AI Act class applies, and which architecture carries your data sovereignty? Both answered in detail on the tech sub-pages.
Four risk classes — unacceptable, high, limited, minimal. High-risk applications need audit trails, documentation, and human oversight from day one. Non-compliance penalties: up to €35M or 7% of worldwide annual turnover (Art. 99). We classify per use case and build compliance parallel to development — written report, no workshop mood.
AI Readiness Audit in detail →Frontier API (Anthropic, OpenAI, Google), local model (Llama, Mistral), or hybrid with routing — decision per use case based on data sensitivity, latency, cost, and vendor lock-in. Default: data stays in-house. Cloud only when controlled.
AI Orchestration in detail →This page: C-level view. Tech depth (model architecture, AI Act matrix, audit pillars) on AI Orchestration. Architecture principles as reference: Sovereign by Design — five principles.
Bitkom 09/2025. Sovereignty does not hold through brand loyalty — it holds through architecture, even when vendors consolidate. Read the five architecture principles →
We build with you, not for you. Engagements are workbenches, not theatre. Delivery in short loops. Clean handover, then your team carries it alone.
Method in detail
25 years in IT, 14 of them in B2B commerce. Architecture mandates with large enterprises, building and steering distributed expert teams, vendor-neutral project rescue. Focus areas: platform architecture, project rescue, team operations.
Chris Zepernick
We measure use cases against the daily reality of the people who are supposed to use them. If the field rep ignores the recommendation, the model has no value — regardless of how well it benchmarks.
How we workAI use case in the pipeline? A few minutes on the phone. We sort signal from hype.
