Value
Use case and business value
We document the current workflow, its cost and the impact of errors. An AI idea only becomes a viable use case once a measurable outcome has been defined.
- Value hypothesis
- Baseline
- KPI
AI consulting
A viable use case first. Then the right AI.
We define the problem, quality criteria and acceptable failure modes before selecting a model. We then compare operating options, build a pilot in the real workflow and can take responsibility for integration, adoption and operation.
Starting point identifiedA viable use case first. Then the right AI.
01
Use case and business value
02
Data and knowledge access
03
Cloud and local models
04
Integration and automation
05
AI Act, privacy and governance
06
Adoption and ongoing operation
Your starting point
Signs that action is needed
We start with the problem, its consequences and the person who will own the outcome, not with a tool.
01
Many ideas, no viable use case
Teams test tools, but nobody can name the value, impact of errors and accountable owner in concrete terms.
02
Knowledge is scattered
Documents, permissions and domain expertise are not organised in a way that lets a model use them reliably.
03
Cost and quality remain invisible
Answers look convincing, but neither domain quality nor the economics of a completed case are measured.
04
AI is already used without control
Employees use different services without agreed data paths, approvals, roles or a shared response to failures.
Technical coverage
From an idea to controlled AI operations.
We connect the business workflow, data, model, integration and accountability. The result is not an isolated chat, but an auditable workflow with explicit boundaries.
Knowledge
Data and knowledge access
Sources, freshness, permissions and deletion paths are designed together. In RAG systems, every answer must be traceable to content the user is actually allowed to access.
- RAG
- Permissions
- Sources
Technology
Model and operating decision
A cloud API, a European provider and a local model are measured against the same test set. Quality, latency, data paths and cost determine the choice together.
- Cloud API
- Local AI
- Model evaluation
Workflow
Integration and automation
We connect the model to business systems and explicit approval points. Logging, safe refusal and a manual fallback are designed into the workflow.
- Interfaces
- Workflows
- Human in the loop
Accountability
Governance, privacy and the AI Act
An application inventory, roles, risk classification, transparency and suppliers are documented before production. Obligations become operating routines, not archived paperwork.
- EU AI Act
- GDPR
- Governance
Adoption
Enablement and ongoing operation
Business teams, management and administrators receive role-specific rules and exercises. After launch we monitor quality, usage, cost and changes to models and data.
- AI literacy
- Training
- Monitoring
AI pilot in detail
Prove it before production.
A pilot is not a demo with three successful questions. We use real cases to determine whether quality, data access, effort and operating risk are viable together.
Phase 1 Problem and baseline
Record time, quality, impact of errors and accountability in the current workflow.
Phase 2 Test cases and data
Assemble representative cases, permitted sources and an explicit evaluation rubric.
Phase 3 Compare models
Evaluate several operating options against the same cases, constraints and cost assumptions.
Phase 4 Test in real work
Use a limited user group, real workflow steps, human approval and complete logs.
Outcome Make a deliberate decision
Choose production, further work or a stop based on the agreed criteria.
Decision criteria
Approval never depends on one impressive answer. All four dimensions need sufficient evidence for the specific workflow.
Domain qualityMinimum quality agreed before the pilot
TraceabilitySources and refusals can be reviewed
EconomicsCost per completed case is visible
Operational readinessPermissions, oversight and failure paths are defined
What you receive
- Documented use case with explicit non-goals
- Representative test set and evaluation rubric
- Model comparison including a local operating option
- Assessment of data paths, permissions and AI Act duties
- Pilot in the real workflow with human approval
- Production decision with cost and next steps
Market and decision
What is changing and what to decide first
We assess developments by whether they change the outcome, risk, effort or operating model.
Development
Adoption is growing, maturity varies
In 2025, 20.0% of EU enterprises with at least ten employees used AI, an increase of 6.5 percentage points from 2024. The next constraint is often controlled adoption in a real workflow.
Development
Regulation is already an operating concern
Substantial provisions of the EU AI Act have applied since 2 August 2026, while some high-risk areas have later transition dates. Inventory, roles, literacy and classification should already be part of a pilot.
Development
Evaluation matters as much as implementation
The NIST profile for generative AI highlights governance, content provenance, pre-deployment testing and incident disclosure. A capable model alone is not a reliable service.
Clarify before commissioning
- Define value, baseline and impact of errors before selecting a model
- Document data access, permissions, storage and deletion paths
- Measure quality with real cases and a fixed evaluation rubric
- Calculate cost per completed case, not only cost per token
- Design human approval, safe refusal and a manual fallback
- Assign ownership for changes to models, data and workflows
From problem to outcome
A defensible path to the next decision
- 01
Define the problem and baseline
We record the current workflow, roles, time, quality and impact of errors. This establishes the goal and explicit non-goals.
- 02
Prepare data, permissions and test cases
Representative cases, permitted sources, evaluation criteria and data paths are agreed before the first model test.
- 03
Compare models and operating options
Cloud APIs, European services and local models process the same test set. Quality, speed, cost and sensitivity are reviewed together.
- 04
Integrate a pilot into real work
A limited user group works with explicit approvals, logs and a manual fallback. Deviations are recorded and addressed.
- 05
Decide and operate
The agreed criteria determine production, further work or a stop. A launch adds monitoring, training and controlled change management.
Outcome
What changes for your team
1
Evidence before investment
Test cases and approval thresholds show whether the use case is viable. A well-founded stop is also a useful outcome.
2
Measurable domain quality
Answers are assessed against a documented rubric and representative cases, not by their first impression.
3
Controlled data paths
Sources, permissions, providers, storage and deletion paths are known and aligned with the workflow.
4
A team ready to operate
Owners understand boundaries, approvals and failure paths. The workflow does not depend on one prompt specialist.
FAQ
Common questions about AI consulting
Concise answers for an initial assessment.
Do we need to know our AI use case already?
No. We can begin with the current workflow and identify where a measurable outcome is realistic.
Does every project start with a large platform?
No. A limited pilot with explicit quality thresholds is usually the better first step.
How do you address privacy?
We clarify data types, legal basis, providers, storage locations, permissions and deletion paths before production integration.
Does the model have to run in the cloud?
No. We compare cloud services, European providers and local models using the same domain cases. Sensitivity, quality, latency, operating effort and total cost determine the right option.
Do you only advise or do you also implement?
We can cover the complete path from workflow assessment and evaluation to pilot, integration and operation. A strategy-only engagement is possible, but it is not our only deliverable.
What should we prepare for the AI Act?
Requirements depend on your role, purpose and risk classification. A practical baseline covers an inventory, accountable owners, data, providers, human oversight, transparency and AI literacy. This does not replace formal legal advice.
Further depth
Continue with evidence and practical guidance
AI strategy
Connect goals, workflows and ownership before selecting technology.
Explore insightsModel selection
Compare cloud and local models against the same domain cases.
Explore insightsPrivacy and governance
Make data paths, responsibilities and controls part of the operating model.
Explore insightsAI pilot
Use a representative test set to make a conscious production decision.
Explore insightsNext step
Discuss your use case Assess your AI initiative with evidence
We clarify the problem, data, quality criteria and the smallest useful pilot. You will know what is viable and what is not.