Report May 2026 Read time ~10–12 min

AX Readiness 2026 —
where houses stand today.

First read-out from CRAiD diagnostic sessions across the German-speaking mid-market and enterprise: three patterns, one blind spot. Where the move from CX to AX structurally breaks — and where it already runs.

Evidence base: Bitkom · Mittelstandsbund/BIDT · Gartner · MIT · KPMG · Capgemini · Cognizant · ASAPP · NICE · Anthropic Economic Index. Full source list ↓

00 Intro

CX was journey work. AX is structural work.

The DACH economy answered the adoption question in 2025/26. 36 % of German companies use AI — a doubling versus 2024 (Bitkom 2025). 47 % plan or discuss adoption; only 17 % still consider AI "not relevant" (down from 41 % in 2024). A success story.

What stays open is the scaling question. And the logic collapses there: only 9.5 % of the German Mittelstand have fully implemented AI (Mittelstandsbund/BIDT). Only ~12 % of all AI pilots ever reach production (Gartner). And Gartner predicts that more than 40 % of agentic AI projects will be cancelled by end of 2027.

In DACH, AX rarely fails because of models, vendors or "the AI". It fails because nobody builds the line in which an agent can live.

Across recent CRAiD diagnostic sessions, three recurring patterns keep that gap open — plus one blind spot that amplifies it. Both are well documented by market data. The report shows where the build breaks today, what to rebuild in 30/60/90 days, and which eight questions you can use to do a quick health check yourself.

01 Where the market stands

Adoption is solved.
Scaling is the open wound.

The last 24 months were an adoption wave. The next 24 are an operating-model question. Steering on adoption KPIs in 2026 means measuring a battle already won — and missing the one currently being lost.

Adoption
36 %
of German companies use AI actively (2024: 20 %).
Doubled within 12 months.
Bitkom AI 2025 · n=604
Scaling
9.5 %
of the Mittelstand have fully implemented AI. 24 % in pilot, 43 % no plan.
Mittelstandsbund / BIDT 2025
Pilot → Production
~12 %
of AI pilots reach production (4 of 33). 88 % vanish in pilot limbo.
Gartner via Gondlach 2025
Cancellation
40 %
of agentic AI projects will be cancelled by end of 2027 — cost, value, risk.
Gartner June 2025

The global view supports the picture. MIT Media Lab documents in "State of AI in Business 2025": 95 % of generative AI pilots deliver no measurable ROI — despite $30–40 billion in investment. The 5 % that succeed don't differ by better models, they differ by better build: they integrate AI into existing workflows instead of bolting it on.

In DACH, Cognizant reports: 71 % of companies feel they aren't moving fast enough on their GenAI strategy. Enterprise (>500 FTE) sits above 60 % adoption, the Mittelstand significantly below (Bitkom AI 2026). Both halves wrestle with the same structural bottleneck — from opposite sides.

02 Three patterns

Three patterns —
one shared bottleneck.

These are not maturity stages. They are structural constellations. Once you see them, you can rebuild on purpose — without buying the next tool first.

Pattern 01

Pilot islands without a load-bearing structure

Symptom

Use cases work locally but don't scale across teams or journeys. Steering meetings discuss tools, not accountabilities.

Root cause

Decision rights, ownership and standards are missing. AX runs as a project pilot, not as a line function. After the pilot, no one is operationally accountable.

Signal

Who can decide — without escalating to steering — whether agent X goes live on journey Y? "No one" = the trap.

Move

Before the next pilot, name a line role that stays accountable after the pilot — with budget, mandate, outcome targets.

Evidence: Only ~12 % of AI pilots reach production (Gartner via Gondlach). Cognizant: 71 % of DACH companies say they're not moving fast enough.
Pattern 02

Source of truth is missing

Symptom

Teams discuss models and prompts, but agent answers are inconsistent or non-auditable. "Did the agent get that right?" is the most common internal question.

Root cause

No consolidated source of truth per customer journey, no data paths between CRM, service tooling, knowledge base and agent layer. The EU AI Act has enforced auditability since August 2025.

Signal

Which three data sources are "source of truth" for your top-2 journeys? If the answer triggers a debate, the foundation is fundamentally tilted.

Move

Document a source of truth per top journey before building the next agent. Without it, every investment is symptom-fighting.

Evidence: 76 % of SMEs report data quality issues, 71 % wrestle with data silos, 83 % have no data strategy (Maximal.digital 2025).
Pattern 03

Containment instead of outcome

Symptom

Steering KPIs are usage, adoption rate, containment rate. Nobody can say whether customer effort is going up or down. Risk guardrails are not defined.

Root cause

No shared outcome definition between service, product, data and compliance. Vendor scorecards measure what the vendor wants to sell.

Signal

Which three outcome metrics would justify killing a pilot because they got worse? "We don't measure that" = output theatre.

Move

Define three outcome metrics before every pilot: customer effort, quality (e.g. CSAT delta), risk (e.g. hallucination rate). Containment is operations, not steering.

Evidence: ASAPP: containment can rise while NPS and customer effort fall. LatentView: 30–50 % cost-to-serve reduction only with a clear outcome framework.
03 Blind spot

The blind spot —
it's called Steward.

Not the model. Not the data volume. Not the budget. A line role that's almost everywhere thought about — and almost nowhere actually built.

Core point

No one is building the structural learning loops. Who watches the agent? Who decides on interventions? Who writes outcomes back into the journey? As long as "Agent Steward" is a side task of tool admins instead of a role, every platform investment is structurally still a pilot.

An agent isn't a tool — it's an employee without a manager. A human in the same position would have: an onboarding plan, a steward (mentor/lead), performance reviews, escalation paths, data-protection training, clear decision rights. The agent gets: an API key.

The research is unambiguous here. KPMG "AI governance for the agentic AI era" (2025) makes the point: classical AI governance is built for predict/generate models. Agentic systems take actions — and need their own structures for audit trails, identity, liability. Most organisations simply don't have that layer. Mayer Brown adds the legal angle: under the EU AI Act, agent identity, authorisation scopes and audit trails are no longer best practice — they're required.

The market reaction confirms the gap. Vendors like SAS bake stewardship layers into their platforms (SAS Viya, April 2026) — because the line role is missing in most companies. That's symptom, not cure.

"The primary barrier to widespread adoption is no longer a lack of capability — it is a lack of trust."a21.ai · From Ignore to Execute (2025)

And the DACH practitioner voice says the same. Thomas Maxeiner (Beyondbuzzwords, Feb 2026): "Mittelstand companies need an agentic governance council." Roover.de puts it more bluntly: "The biggest hurdle is rarely the model — it's operations."

This isn't hype. It's the direct consequence of 9.5 % at scale, 95 % pilot failure, and the 40 % cancellation forecast. The companies that anchor stewardship as a line function are the 5 % that will still be running in 2027.

04 The House model

One house, five components —
that's how AX becomes diagnosable.

The House model is a quick map: where do you stand today — foundation, frame, rooms, supply, roof? Maturity isn't "everything perfect", it's "nothing critical missing".

Component Typical gap Intervention (smallest step) Owner
Foundation
Data & process clarity
Source of truth unclear; data silos (71 %), no data strategy (83 %) Per top journey define 3 SoTs, update cadence + audit path Data/Platform + Journey Owner
Frame
Operating Model
Pilot instead of line; decisions diffuse; no one accountable post-pilot 1 AX owner with budget · 1 stop/go decision logic · 1 standard for inputs/outputs/QA Transformation / COO / CX Lead
Rooms
Journeys / use cases
Use cases instead of journeys; "nice to have" instead of business-critical Prioritise top-2 journeys; document outcome & risk profile Journey Owner
Supply
Enablement & stewardship
Steward role doesn't exist as a profile; enablement is ad hoc; guardrails not operational Steward mandate as 0.2 FTE in the line; onboarding + escalation paths; EU AI Act training HR/L&D + Ops + Legal
Roof
Outcome measurement
Containment / adoption used as steering KPI; no outcome logic; no risk telemetry Per journey: customer effort + quality + risk; demote containment to operations Finance/Analytics + Risk
05 Self-assessment

Eight questions —
for a quick health check.

If you can't answer these clearly, you're missing components. If you can, you have a roadmap — even without buying a new tool.

AX Readiness — Self-Assessment

  1. Which decisions can the AX team take today without escalating to steering?If the answer is "none", AX lives as a project, not as a line function.
  2. Which three data sources are "source of truth" for your top-2 journeys?If the answer triggers a debate, the foundation is tilted.
  3. Which three outcome metrics would justify ending a pilot — because they got worse?"We don't measure that" means output theatre, not steering.
  4. Who is operationally accountable after the next pilot — by name, with a budget?Without a named owner, the pilot is a 90-day limbo with an expiry date.
  5. Which role in your org has "Agent Steward" as a profile — not as a side task?Steward = mentor + auditor + outcome owner for every productive agent.
  6. Which customer journey is EU AI Act-auditable today?Mandatory since August 2025. Fines up to 7 % of revenue.
  7. How much time passes between an agent output and outcome feedback back into journey data?Without a learning loop, the agent doesn't get better. It just gets older.
  8. When was an AI pilot last deliberately stopped because an outcome metric got worse?"Never" is a diagnosis, not a success.
06 30/60/90 days

A plan that builds —
without overwhelming you.

The plan follows the House logic: stability first, scale second. Not the other way round.

30 days

Diagnose & setup

  • List your top-3 AI initiatives — for each, clarify: who is accountable after the pilot ends?
  • Pick the top-2 journeys (owner + outcome definition)
  • Document 3 source-of-truth data points per journey (owner, update frequency, audit path)
  • Define the outcome triplet per pilot: customer effort · quality · risk
  • Sketch the steward mandate (0.2 FTE in the line, not as add-on)
60 days

Pilot(s) + guardrails

  • Move one initiative from pilot mode into a line role — with budget & mandate
  • Source of truth for top-1 journey complete, EU AI Act-aligned
  • Outcome dashboard live: customer effort + quality + risk visualised (not containment-only)
  • Steward in place; onboarding plan and escalation paths defined
  • Stop/go drill: deliberately test a pilot against outcome metrics
90 days

Scale & operating model

  • AX operating model v0.5: roles, decision rights, standards — pressure-tested in 2 areas
  • Source-of-truth standard for all top journeys; MCP/integration strategy defined
  • Outcome logic embedded in steering KPIs; containment-only reports retired
  • Stewardship as a regular line function — no longer "a new initiative"
  • Scale decision: onboard further journeys based on demonstrated impact
07 Sources

Evidence & sources.

Cross-validation: every core claim is backed by at least two independent sources, including at least one with a DACH/EU lens. No PR-solo citing.

  1. Bitkom Research — "Künstliche Intelligenz 2025" · n=604 companies with 20+ FTE, representative. 36 % adoption (2024: 20 %), top barriers: legal 53 %, know-how 53 %, data protection 48 %, traceability 38 %, missing data 24 %. bitkom-research.de
  2. Mittelstandsbund / BIDT KI-Index Mittelstand 2025 · 9.5 % fully implemented, 24 % in pilot, 43 % no plan. ~10 % with agentic experience. bidt.digital
  3. Maximal.digital — KI im Mittelstand 2025 · 76 % data quality issues, 71 % data silos, 83 % no data strategy, 69 % don't know which data is relevant. maximal.digital
  4. Gartner — "Over 40 % of Agentic AI Projects Will Be Canceled by 2027" · June 2025. Anushree Verma: "agent washing" (~130 real out of thousands of vendors). gartner.com
  5. MIT Media Lab — "State of AI in Business 2025" · 95 % of GenAI pilots without measurable ROI. Coverage: Forbes / Fortune. forbes.com
  6. Kai Gondlach — "From pilot to rollout: AI operating model" · ~12 % pilot-to-production rate (Gartner reference: 4 of 33). kaigondlach.de
  7. Cognizant — "Gen AI conquers DACH companies" · 71 % of DACH companies feel they aren't moving fast enough. cognizant.com
  8. Capgemini Research Institute — World Quality Report 2025 · 15 % enterprise-scale deployments at ~90 % active experimentation. capgemini.com
  9. ASAPP — "Moving beyond containment" · Containment rate is the most common but most misleading metric in conversational AI. asapp.com
  10. NICE — "Essential KPIs for Agentic AI CX" · Outcome framework: customer effort, agent experience, adaptive intelligence, compliance. nice.com
  11. LatentView — "Agentic AI in Customer Service" · 30–50 % cost-to-serve reduction with a clear outcome framework. latentview.com
  12. KPMG International — "AI governance for the agentic AI era" (2025) · Agentic governance needs its own structures (audit trail, identity, liability). kpmg.com
  13. Mayer Brown — "Governance of Agentic Artificial Intelligence Systems" (Feb 2026) · Agent identity, authorisation scopes, audit trails as legal duty. mayerbrown.com
  14. a21.ai — "From Ignore to Execute: Measuring Trust in Agentic AI" · "The primary barrier is no longer capability — it is trust." a21.ai
  15. Beyondbuzzwords (Maxeiner, Feb 2026) · "Mittelstand companies need an agentic governance council." beyondbuzzwords.de
  16. SAS — "SAS Viya: secure AI assistants & agentic AI features" (April 2026) · DACH market signal: stewardship as a platform layer. sas.com
  17. EU AI Act compliance — Mittelstand 2025/26 · Transparency and training duties since August 2025. Fines up to 7 % of revenue. openpr.de
  18. Codana — "MCP – Model Context Protocol" · The N×M integration problem; MCP as an emerging standard since 2025. codana.de
  19. Roover.de — "AI-first organisation in the Mittelstand" · "The biggest hurdle is rarely the model — it's operations." roover.de
  20. Bitkom AI Study 2026 · 41 % active use (2026), 48 % planning. Enterprise (>500 FTE): >60 % adoption; Mittelstand significantly lower.
← Back to Insights