AI & Digital Tools
AI that makes you productive - not dependent.
We advise mid-market companies, family offices and energy firms on AI strategy, agentic workflows and knowledge-graph buildout. Local-first, GDPR-compliant, vendor-lock-in-free. We do not sell SaaS - we deliver architecture.
- AI strategy reviews with clear use-case inventory instead of buzzword bingo
- Agentic workflow architecture (multi-persona, self-learning, compliance veto)
- RAG and knowledge-graph buildout for internal knowledge bases
- Local-first LLM setup (Ollama, qwen2.5, llama3.3) for GDPR-sensitive data
The AI & Tech Dossier.
Institutional sovereignty is achieved through access to validated sector coupling pathways and rigorous strategic modeling.
Strategic Value Levers
Agentic Due Diligence
Targeted control and evidence-backed execution to maximize returns and mitigate portfolio risks.
Local-AI Sovereignty
Targeted control and evidence-backed execution to maximize returns and mitigate portfolio risks.
Data Integrity Gates
Targeted control and evidence-backed execution to maximize returns and mitigate portfolio risks.
Decision Scenarios
"Automated document analysis via cloud AI (anonymized)."
Ready for a confidential Strategic Audit?
AME principals invite you to a structured evaluation. Let us assess tangible value levers and decarbonization pathways for your portfolio.
Foresight &
Risk Remediation.
Heightened liability risks when using third-party AI for confidential transaction data.
Migration to locally operated AI models to keep data sovereignty with the client.
Non-validated AI models lead to governance gaps and audit risks.
Set up validation and sign-off gates with a documented audit trail to secure outputs.
Topics we manage.
AI Strategy - Use-Case Inventory
We scan your organization for concrete AI levers: which processes, which data, which ROI. Output: inventory with P0/P1/P2 prioritization instead of Excel wishlist.
Agentic Workflows
Multi-persona systems with compliance veto and self-learning loop. We build the architecture - you operate it yourself. Vendor-lock-in free.
RAG & Knowledge Graph
Retrieval-Augmented Generation on your data. Knowledge chunks with source tracking, conflict resolution, uniqueness audit. Hallucination-resistant architecture.
Local-First LLM Setup
Ollama-based local stacks for GDPR-extreme data (clients, whistleblowers, wealth). Cloud LLMs only for aggregated, anonymized outputs.
Tool-Routing Strategy
Which tool for which task? Claude Code for logic, Cursor for inline edits, Lovable for UI, Ollama for sensitive. We deliver the routing matrix.
Compliance & Audit Trail
Every LLM call documented. Compliance-veto system as hard gate. GDPR-tier routing (low/medium/high/strict). Audit logs for board and regulator reviews.
Four stages from first use-case to agentic operating system.
Discovery & Use-Case Inventory (Weeks 1-3)
Workshop-based exploration across processes and pain-points. We deliver an inventory of 10-20 use-cases, ranked by ROI, complexity and GDPR tier.
Architecture Decision (Weeks 4-6)
Which LLMs local, which cloud? Which workflow patterns? Which tools? We deliver a decision memo with three viable architectures and one recommendation.
Pilot Build (Weeks 7-14)
We build the first use-case end-to-end: ingestion, RAG, persona, compliance veto, audit trail. You have a working pilot you can run yourself.
Scaling & Knowledge Transfer (Weeks 15+)
Your team takes over. We remain strategic sparring partner, document architecture, train operators, audit quarterly health checks.
Run the numbers on your AI and tech case.
Every recommendation on this page has a tool behind it.
AI Data Center: GPU Cluster Calculator
Model the CapEx, utilization and unit economics of your GPU cluster for AI training and agentic workloads.
Data Center Sustainability
Assess PUE, energy demand and carbon footprint of your data center under a growing AI load.
Immersion Cooling Calculator
Compare the investment and operating cost of immersion cooling against classic air cooling at high rack density.
Post-Quantum Crypto Migration
Estimate the effort and timeline of moving to quantum-safe encryption - relevant for GDPR and long-term data retention.
Quantum Key Distribution (QKD)
Model the range, key rate and cost of optical QKD links for highly secure data communication.
Executive Scenario Simulator
Play through digitalization and AI transformation scenarios with value and cash impact.
Scope
What this service covers
The AI & Digitalization service shapes the use of AI and data in companies. Technology companies as a market belong to the Tech & Growth sector. The service covers process and data analysis, use case selection, automation, integration into existing systems, rules and responsibilities, and secured operation. Data centres, GPU clusters, cooling technology and quantum-safe encryption are explored as market and investment questions in the Tech & Growth sector. Legal assessment, for example under data protection law or the European AI Act, is provided by legal and data protection advisers; this service creates the business and organisational foundation for it.
Starting point
Many companies test AI without first clarifying the task and the data situation. The result is pilots that run technically but make no process faster, cheaper or more reliable, and tools that sit alongside the actual workflows.
Review use cases and data maturity
Which task should measurably improve, and which data may be used for it? Without these two answers, every project lacks the baseline against which success or failure can later be read.
Prioritise solutions by benefit and risk
There are usually more ideas competing for budget and specialists than a company can implement at once. Ranking them by benefit, effort, data risk and feasibility prevents the most visible rather than the most effective project from winning.
Integrate processes and interfaces
Automation only works once its output reaches the leading system, such as accounting, customer management or document filing. Stand-alone solutions next to them create duplicate data entry and new sources of error.
Secure quality, operation and fallbacks
Where needed, human review and an independent fallback remain in place. Models, data and interfaces change constantly; without monitoring, result quality declines without anyone noticing.
Data is often the real bottleneck
Master data has grown over years, departments maintain the same data differently, and spreadsheets serve as a hidden database. Data quality, meaning consistency, timeliness, accessibility and documented origin, is therefore usually the first task, not the last.
Business cases are calculated too optimistically
Vendor calculations often show licence costs and best-case assumptions. Integration, data cleansing, training, monitoring, retraining and process change are missing, and time saved is booked as profit although it has not yet been redeployed.
Responsibility and rules are missing
Business units procure tools without involving IT, data protection or management. There is no register of who uses which application with which data, and when errors occur it is unclear who is accountable.
Waiting has a cost too
Repetitive manual work continues, experience is lost when staff change, and separate data sets become more costly to merge every year. These costs are less visible than a project budget, but they accrue every day.
From use case to operation
- Stage 01
Survey
Record processes, manual steps, error cases, data sources and tools already in use; capture processing time, rework and cost per case as the baseline.
Levers
- Prioritised inventory
- Baseline values per case
- Overview of tools used without rules
- Stage 02
Assess and select
Rank use cases by benefit, effort, data risk and feasibility; review the solution without AI on equal terms for each case and set termination criteria.
Levers
- Ranking by impact
- Alternative without AI reviewed
- Termination criteria before the start
- Stage 03
Decide
Define architecture and data paths: usage rights, where models run, interfaces, logging and fallbacks.
Levers
- Data rights clarified
- Vendor dependence limited
- Integration effort known
- Stage 04
Pilot
Build one use case end to end, with real data, human review and measurement against the baseline.
Levers
- Comparison baseline
- Measurable result quality
- Approval based on measured values
- Stage 05
Integrate and introduce
Embed the solution in leading systems and workflows, train roles, and anchor checkpoints and escalation paths.
Levers
- No duplicate data entry
- Acceptance in business units
- Clear responsibility
- Stage 06
Operate
Tests, monitoring, logging and fallbacks; review quality and cost at a fixed interval.
Levers
- Recoverability
- Cost per completed case
- Early detection of quality loss
Typical decisions
Which task is tackled first?
The inventory usually contains more ideas than budget and specialists can support. The choice of the first use case decides whether the pilot builds or uses up trust.
Options
- Largest measurable time or cost lever
- Lowest data and implementation risk as a learning case
- Bottleneck with the highest error or liability risk
AI, conventional automation or a change to the process?
Many tasks can be solved more reliably and cheaply with rules, forms, interfaces or changed responsibilities than with a language model. AI pays off above all where inputs are unstructured and cases vary.
Options
- Simplify the process without new technology
- Rule-based automation or standard software
- AI-supported solution with human review
Which models run locally, which in the cloud?
Data protection, cost and quality set the framework. Particularly sensitive data, for example from client, personnel or wealth matters, argues for local operation; compute-intensive tasks with anonymised data tend to suit cloud models.
Options
- Local operation
- Cloud models for anonymised data
- Combination
Build, buy or combine?
Building in-house creates control and internal knowledge but ties up specialists permanently. Standard products start faster but make the company dependent on the vendor's pricing and product decisions.
Options
- In-house build with an internal operations team
- Standard software with configuration
- Core logic in-house, individual components bought in
How much decision authority does the system receive?
The greater the consequences of an error, the closer human oversight must be. The chosen level determines review effort, liability risk and the achievable relief.
Options
- Proposal, a person decides
- Automatic execution with sample checks
- Automatic execution only below defined thresholds
Roll out, improve or end the pilot?
After the pilot, a comparison with the starting point is available. Without criteria set in advance, a weak pilot is often continued out of habit and a good one is not rolled out consistently.
Options
- Roll out to further areas
- Improve with a new review date
- End and document the lessons learned
Value levers
Processing time
- Metric
- Processing time and rework rate: minutes per case and share of cases reworked in per cent, each against the measured baseline
- Effect
- Shows whether a process is actually relieved. Time freed up only affects earnings once it is redeployed.
Result quality
- Metric
- Measurable result quality against a baseline: error rate in per cent and share of results approved without correction
- Effect
- Separates technical from business success and shows how much human review is still needed.
EBITDA effect
- Metric
- Change in EBITDA in EUR per year: costs saved or avoided and additional cases handled, less ongoing licence, usage, operating and monitoring costs
- Effect
- Shows whether an earnings contribution remains after all ongoing costs, instead of showing gross savings only.
Cost per completed case
- Metric
- Total process cost in EUR per case completed in business terms, before and after the change, including staff, technology and review costs
- Effect
- Links technology and staff costs to the actual output and prevents only partial costs from being compared.
Investment need and payback
- Metric
- One-off costs in EUR for build, integration, data cleansing and training, and payback period in months until offset by the monthly net effect
- Effect
- Sets the budget frame and shows under which assumptions a project does not pay for itself.
Cash flow and capital tied up
- Metric
- Change in operating cash flow in EUR per month and in capital tied up in days, for example the time from invoicing to payment receipt or the coverage of receivables and inventory
- Effect
- Automated invoice checking, dunning or demand planning often affect liquidity more strongly than earnings.
Time to implementation
- Metric
- Weeks from approval to productive use and to the first measured effect
- Effect
- Every week of delay postpones the benefit while costs are already incurred.
Risk concentration
- Metric
- Share of critical work steps in per cent that depend on a single vendor, model or person, and number of steps without a tested fallback
- Effect
- Reveals dependencies that endanger operations in the event of price changes, outages or staff turnover.
Controllability
- Metric
- Share of applications in the inventory in per cent that have a named responsible person, documented checkpoints and logging
- Effect
- Makes measurable whether management and supervisory bodies can oversee use and intervene when errors occur.
Availability and recoverability
- Metric
- Availability in per cent of agreed operating time and recovery time in hours after an outage, including the switch to the fallback
- Effect
- Ensures that business operations continue when a model or service is unavailable.
Risks and early indicators
| Risk | Early indicator | Countermeasure |
|---|---|---|
| RiskFaulty results without review | Early indicatorNo defined human control; rising share of results adopted unchecked | CountermeasureDefine checkpoints and logging, grade sample checks by the consequences of the case |
| RiskDependence on one vendor | Early indicatorNo alternative for critical steps; prices or terms of use change at short notice | CountermeasurePlan an architecture with fallbacks, keep data, configuration and test cases exportable |
| RiskPilot without transition to operation | Early indicatorNo roll-out or termination criteria before the start; no responsible operating unit named | CountermeasureSet decision criteria, an operating budget and operational responsibility before the pilot starts |
| RiskInsufficient data quality | Early indicatorHigh share of duplicates, inconsistent definitions, manual corrections after every run | CountermeasureReview data before the build, name data owners, automate quality checks |
| RiskUnclear usage rights and confidentiality of data | Early indicatorSensitive data in external services without documented approval; tools procured without the knowledge of IT and data protection | CountermeasureClassify data, introduce an approval process, process sensitive data locally and have legal questions reviewed by specialist advisers |
| RiskUnderestimated total cost | Early indicatorUsage-based costs rise faster than the number of cases; integration and monitoring effort not budgeted | CountermeasureFull-cost calculation per case, cost ceilings and a monthly comparison with the plan |
| RiskGradual loss of quality in operation | Early indicatorFalling share of results approved without correction, more queries from business units, model versions changed unnoticed | CountermeasureOngoing measurement against fixed test cases, version control and a set review interval |
| RiskLack of acceptance in business units | Early indicatorLists kept in parallel, low usage rate, new workflows bypassed | CountermeasureInvolve business units early, train on their own cases, act visibly on feedback |
| RiskRegulatory requirements recognised too late | Early indicatorNo register of applications in use, no risk classification, no documentation of purpose and oversight | CountermeasureSet up the register and a proposed classification early, obtain legal assessment from legal advisers |
Links to sectors and services
- Go to topic: Real EstateAsset data and operating processes: tenant lists, consumption data and technical documents become usable for portfolio analysis, valuation indications, building control and predictive maintenance, provided meters, sensors and maintenance history are available.
- Go to topic: EnergyAsset and metering data in operation: forecasts for procurement, load and self-consumption as well as automated evidence require high-resolution metering data and clean contract data.
- Go to topic: Tech & GrowthTechnology companies as clients, without equating the sector with the service. Data centres, GPU clusters and quantum-safe encryption are explored there as market and investment topics.
- Go to topic: HealthcareAdministrative processes and data: scheduling, documentation and billing can be supported, while treatment and care decisions remain with the responsible professionals. Health data requires particularly strict access, logging and deletion rules.
- Go to topic: EducationLearning administration and data processes: enrolment, attendance records, examination organisation and billing often sit in several systems and can be brought together and partly automated.
- Go to topic: Strategy & TransformationAI changes roles, workflows and steering. So that pilots do not remain isolated, they need to be anchored in the company's operating model and governance.
- Go to topic: M&A & SuccessionDue diligence and integration involve large volumes of documents and divergent data sets. Document analysis and data harmonisation speed up the work, while valuation remains a human task.
Approach
- Step 1
Survey and use case inventory
Workshop-based survey of processes, weak points, data sources and tools already in use. Maturity is assessed across data quality, process clarity, governance, skills and infrastructure, and baseline values for processing time, rework and cost per case are measured or estimated with reasons.
Result: Prioritised use case inventory with maturity and baseline values
- Step 2
Assessment and business case
Each prioritised case is calculated with full costs, conservatively estimated benefits and three scenarios. The solution without AI is reviewed on equal terms, and roll-out and termination criteria are set.
Result: Business case per use case with decision criteria
- Step 3
Architecture decision
Which models run locally, which in the cloud, which workflow patterns and interfaces are used, and which data paths are permitted? Several viable options are compared by cost, risk and dependencies.
Result: Decision paper with options and a recommendation, data and integration concept
- Step 4
Pilot with a baseline
One use case is built end to end: data intake, processing, human review, logging and fallback. It is measured against the baseline values from the survey.
Result: Working pilot with a measurement report and a decision template
- Step 5
Introduction, operation and knowledge transfer
Your own team or the company's IT service provider takes over operation. Architecture, checkpoints and fallbacks are documented, responsible staff are trained, and quality and cost are reviewed at a fixed interval; on request, AME remains a strategic sparring partner.
Result: Operation, test and fallback plan with documented responsibilities
Scope of service
Scope limits
- No blanket AI return and no full automation of critical decisions.
- No legal advice on data protection or AI regulation.
- No software sold on subscription.
- No commitment to specific savings, timelines or result quality; effects are measured in the pilot.
- No ongoing operation or hosting of solutions by AME; operation is the responsibility of the company or its IT service provider.
- No assurance of legal compliance and no information security audit or certification.
- Data centre, cooling and quantum technology as market and investment questions belong to the Tech & Growth sector.
Decision rights
- The company decides on use, data and operation.
- AME recommends architectures and supports the build.
- Management, or a body it appoints, decides on rolling out a pilot based on the measurement report.
- Data approvals are granted by the company's data owners; legal assessments come from its legal and data protection advisers.
- In reviewed processes, the final business decision stays with the responsible staff unless the company specifies otherwise.
- A project can be ended at any time; the criteria for this are agreed before the start.
Information needed
- Process description and manual steps
- Data sources and usage rights
- System and interface overview
- Error cases, quality requirements and operating costs
- Available baseline values: case volumes, processing times, error and rework rates
- AI and automation tools already in use, with contracts and costs
- Requirements from IT security, data protection and internal policies
- Budget frame, time window and available internal contacts
Deliverables
- Prioritised use case inventory
- Data and integration concept
- Pilot with a defined baseline
- Operation, test and fallback plan
- Maturity assessment across five dimensions
- Business case per use case with three scenarios and decision criteria
- Architecture decision paper with options and a recommendation
- Register of AI applications in use with owners and checkpoints
- Pilot measurement report against the baseline
- Architecture documentation and training concept for your own team
Automated incoming invoice checking: what the earnings effect depends on
Hypothetical example with freely chosen, rounded model values and no link to any company or mandate
- Method
- Static, undiscounted calculation: working time saved multiplied by the fully loaded hourly rate, less ongoing costs, gives the annual net effect. The payback period is the one-off investment divided by the monthly net effect. A sensitivity calculation shows the case in which only half of the time freed up is redeployed.
- Period
- 24 months from productive start; the effect applies evenly from the first month, without a pilot phase and without a ramp-up curve.
| Item | Value | Unit |
|---|---|---|
| Processing time saved (24,000 × 7 minutes) | 2,800 | hours per year |
| Gross effect with full redeployment (2,800 × 60) | 168,000 | EUR per year |
| Ongoing costs | 24,000 | EUR per year |
| Net effect on EBITDA | 144,000 | EUR per year |
| Equivalent per month | 12,000 | EUR per month |
| One-off costs | 90,000 | EUR one-off |
| Payback period (90,000 ÷ 12,000) | 7.5 | months |
| Cumulative net effect after one-off costs (24 × 12,000 minus 90,000) | 198,000 | EUR in 24 months |
| Sensitivity: only half of the time is redeployed, net effect | 60,000 | EUR per year |
| Payback period in this case (90,000 ÷ 5,000) | 18 | months |
| Cumulative net effect in this case (24 × 5,000 minus 90,000) | 30,000 | EUR in 24 months |
Assumptions
- 24,000 incoming invoices per year, evenly distributed
- Processing time today 12 minutes per invoice, after the change 5 minutes including human review
- Fully loaded internal hourly rate of EUR 60
- One-off costs of EUR 90,000 for integration, data cleansing and training
- Ongoing costs of EUR 24,000 per year for licences, usage, monitoring and maintenance
- Base case: the time freed up is fully redeployed to other tasks or avoids additional hiring
- Error costs, quality differences, taxes, financing and early payment discounts are not considered
Limits
The example shows the calculation logic, not the expected effect of any project. Pilot duration, ramp-up, error costs and quality differences are missing, as are taxes and financing. The sensitivity shows that the earnings effect depends above all on whether freed-up time is actually redeployed; without that, there is relief but no earnings contribution. A solution without AI, such as rule-based checking or a simplified approval process, should be calculated with the same method and may be cheaper.
Further topics
The existing pages explore individual work steps in more depth and remain accessible on their own. Related topics are assigned here to the step they belong to: the readiness page and readiness check to the survey, data quality and the data quality assessment to the data review, the ROI page and ROI calculator to the business case, governance to operation. Tools only return results based on the assumptions entered.
- AI consultingAdvice on AI strategy and use: maturity levels, methodology with review gates, sector navigator and notes on the AI Act as a detailed page alongside this hub.Open page
- AI readiness assessmentMaturity across five dimensions: data quality, process clarity, governance, skills and infrastructure. The basis of the first work step.Open page
- AI readiness checkSelf-assessment tool for maturity; it complements the assessment and does not replace it.Open page
- AI data qualityData quality as a prerequisite: consistency, timeliness, accessibility and documentation, plus cleansing, rules and ongoing monitoring.Open page
- Data quality assessmentTool for a structured review of whether the data is sufficient for a use case.Open page
- AI ROI calculationComplete cost blocks, conservatively estimated benefits, three scenarios and payback.Open page
- AI ROI calculatorCalculation tool for an initial estimate of a pilot; the result depends entirely on the assumptions entered.Open page
- AI governanceRules and responsibility in AI use: register, risk classification, roles, escalation paths and review cycle.Open page
- AI use cases in real estateValuation indication, portfolio analysis, building control and predictive maintenance with their data requirements.Open page
- AI use cases in energyProcurement forecasts, load management, forecasting of PV self-consumption and automated evidence.Open page
- AI sector checkGuided assessment of AI readiness by industry.Open page
- AI and strategyAnchoring AI in the operating model; interface with the Strategy & Transformation service.Open page
- AI and M&AAI-supported document analysis in due diligence and integration; interface with the M&A & Succession service.Open page
- For AI decision-makersAudience page for CTOs, CDOs and AI leads with maturity levels and sector use cases.Open page
- Tech & GrowthSector hub for technology companies as a market, including data centre, GPU and quantum topics.Open page
- AI & technology FAQFurther answers on maturity, use cases, governance, data quality, business case and implementation.Open page
Method and evidence
Only content already published. Evidence of effect appears only with proof. The linked pages describe methodology, architecture and tools; they are not proof of effect for individual projects. General market figures on individual subpages say nothing about the effect of a specific project, and calculation tools only return results based on the assumptions entered.
Frequently asked questions
Is AI & Digitalization the same as Tech & Growth?
No. AI & Digitalization is a service for companies in all sectors, Tech & Growth is a client market. A company that digitalises its workflows remains in its own sector; Tech & Growth concerns companies whose product is itself software, a platform or a data-based service.
Does every task need an AI solution?
No. Simplifying the process, a fixed rule, a form or an interface is often more reliable and cheaper. The solution without AI is therefore reviewed on equal terms for every use case and calculated with the same method.
How is the business case for a project calculated?
With full costs for build, integration, data cleansing, training, operation, monitoring and retraining, conservatively estimated benefits and three scenarios. Only measurable effects enter the calculation; qualitative benefits are documented separately. The conditions under which a project will not be implemented or will be ended are set in advance.
How quickly are first results possible?
A first productive pilot is often achievable within a few weeks to a few months. The pace depends on the data situation, process clarity, approvals and internal capacity; no commitment on timing is given in advance.
Does our company need its own developers?
Not necessarily. AME handles conception, tool selection and the pilot build and trains your team. For ongoing operation, however, you need named internal owners or an IT service provider, because AME does not operate the solutions itself.
What happens if our data is not clean?
That is the normal case. Data quality is assessed during the survey; where needed, a cleansing plan with responsibilities is drawn up. A one-off clean-up is not enough, because without rules and ongoing checks quality declines again.
Do the models run locally or in the cloud?
That depends on the need for protection, cost and required quality. Particularly sensitive data argues for local operation, while anonymised or aggregated data can be processed in cloud models; a combination often makes sense. The options are compared in the decision paper.
Does AME advise on data protection and the AI Act?
AME creates the organisational foundation: a register of applications, a proposed classification, roles, checkpoints and documentation. Legal assessment is provided by your legal and data protection advisers. AME gives no assurance of legal compliance.
Which industries is the service suited to?
Companies in all five sectors. In real estate it concerns asset data and property operations, in energy asset and metering data, in healthcare administrative processes such as documentation, billing and reporting, and in education learning administration and data processes. Technology companies can be clients as well.
How much may a system decide on its own?
The company sets this for each process. The greater the consequences of an error, the closer the human oversight: from a mere proposal through sample checks to automatic execution below defined thresholds. Critical decisions are not fully automated.
When is a pilot ended?
When it does not meet the criteria set in advance, for example on result quality, cost per case or acceptance, and improvement is not promising. The lessons are documented so that they inform the next selection.
Let us probe your AI use-case in a structured conversation.
60 minutes under NDA. We hear which processes you want to improve and name three viable paths - at no cost.