Independent AI advisory
Most AI programmes have a strategy. Few make it to production.
They rarely fail because of the technology. They fail between strategy and delivery. That’s where I work: I advise leadership teams on AI strategy and governance and, where needed, help build the applications myself, mostly in knowledge-intensive, regulated, and EU-facing organisations.
01 / Services
How I can help
Three ways of working together, depending on where you stand.
Strategy
AI strategy and operating model
You want to decide where AI can help your organisation and what that requires. I work out with you where AI pays off and where other digital solutions fit better, clarify who owns what, and plan the rollout.
Typical outcomesPrioritised use cases with owners, a workable operating model with roles and guardrails, success metrics, and a roadmap with a clear sequence.
Architecture and delivery
From pilot to production
You have a use case or a first prototype. I work out how it becomes an application for daily use and take part in the technical build myself.
Typical outcomesReference architectures fitted to your system landscape, build-vs-buy assessments, production-grade implementations with quality and safety checks, and documentation that stands up to internal audit.
Sparring
Support for ongoing decisions
You need a second opinion on a vendor, an architecture, or a governance question. I examine the options, name the risks, and give a reasoned recommendation, as a single review or on an ongoing basis.
Common topicsDefining responsibilities and approvals for AI use, technical and commercial vendor due diligence, reducing delivery risk, and coaching for product and engineering leads.
02 / Examples
What I’ve been working on lately
Three anonymised excerpts from recent years. Happy to share more in conversation; references on request.
Eight AI applications, each built for one team’s workflow
For the agency I developed eight web applications that are in use today. One of them produces project references from completed contracts: previously, every reference was copied by hand from the contract and its annexes, with a backlog of around 100 projects. Today the application extracts the fields from the uploaded documents and produces a draft. Every field carries a source reference and a marker showing whether the evidence was found in the document. A manual review step remains mandatory.
An AI lab for in-house applications, with rules and gates
The group wanted to turn recurring workflows into AI applications of its own, without anything unreviewed reaching production. I designed and built the lab. The foundation is a group-wide AI policy with four usage tiers, which also governs the use of purchased AI tools. A small development team works in an isolated test environment. Fixed criteria govern the transitions from idea to prototype, to pilot, and into production, and the people who sign off for IT security, data protection, and AI Act review are named. A portal shows every application with its status and doubles as the register of AI systems. As of September 2026: 19 applications in the pipeline, 4 of them in production, aligned with the EU AI Act and ISO 42001.
Product concept and architecture for a research platform
The start-up wanted to turn raw AI output into market analyses that institutional investors and strategy consultancies in healthcare can rely on. I developed the product concept, methodology, and platform architecture: multi-stage processing with fixed human control points, and for every key claim a confidence score, a source chain, and an audit trail.
03 / Approach
How I work
Four steps I keep in almost every engagement.
Understand the starting point
How is the work done today, which systems are in use, where do handovers stall, which rules apply? Only then do I recommend anything.
Make decisions visible
I lay out open decisions with their options, risks, and consequences. That also makes clear who can take them.
Get something working
I test technical assumptions early against a prototype or an evaluation, in strategy engagements too. That cuts uncertainty faster than another workshop.
Your teams can run it themselves
Frameworks, decision templates, documentation, and handover are part of the job. Your teams should be able to run and develop the solution themselves afterwards.
04 / Tools
Frameworks and formats I use regularly
Four models from my own work, adapted to the context each time.
Lab model for in-house AI applications
A small team builds, IT operates, with a gate in between. Every application passes through four stages, idea, prototype, pilot, production, and has to earn the next one against fixed criteria. Sign-offs are tied to named roles for security, data protection, and the AI Act, not to committees. A prototype stopped early counts as a success.
Guiding questionWhich criterion would this application have to meet for IT to take it over?
Evidence Cards for verifiable AI output
With an AI-generated analysis, you need to be able to trace where every number comes from. An Evidence Card records, for every key claim, the source chain, a confidence score from one to five, and an audit trail: who extracted, who checked, with what result. The model estimates the score and gives its reasoning. Source-quality rules cap it, a person reviews it. Counter-arguments stay visible.
Guiding questionWhich of the core assumptions would have to be wrong for this recommendation to fall apart?
Five AI skills
For training, self-assessments, and role profiles, each skill with a guiding question and development steps:
- Strategic judgement: recognising which work is worth AI support and where AI has no place.
- Adaptive prompting: briefing a task well and steering in dialogue, instead of hunting for formulas.
- Workflow design: breaking a task into a verifiable sequence with human control points.
- Critical evaluation: not trusting a result just because it sounds convincing, and checking where it matters.
- Ethical practice: taking responsibility and disclosure seriously, and knowing what not to use AI for.
Guiding questionWhat would I flag if a junior colleague had handed this in?
Decision room for strategic choices
For leadership teams facing major decisions. AI widens the search to hundreds of options, fixed rules screen them, and every rejected option keeps its reason. A six-role red team, from the sceptical CFO to the pre-mortem, attacks the shortlist. Votes are cast blind first, and discussion starts where disagreement is greatest. Every decision records the assumptions it rests on and the threshold at which it gets a fresh look. The numbers come from data and fixed code, never from the language model. Based on Felipe Csaszar (Harvard Business Review, 2026).
Guiding questionWho in the room sees it differently and hasn’t said so yet?
05 / Industries
The environments I work in
Three environments where I have the most experience. I work elsewhere too, when the question fits.
EU institutions and institutional donors
Multilingual, stakeholder-intensive contexts with high transparency requirements, where new solutions have to be compliant from the start. Experience with procurement logic and donor rules.
International advisory and knowledge organisations
Consulting groups with distributed teams, in-house knowledge production, and several business units. Introducing AI where it changes day-to-day work, with clear roles, rules, and training.
Healthcare and Life Sciences
Manufacturers, providers, and investor-facing market intelligence. AI applications where evidence, traceability, and auditable documentation decide approval and market acceptance, and MDR, the GDPR, and the EU AI Act all apply at once.
06 / Fit check
When I’m not the right fit
If anything below applies, I’ll say so early and, where I can, suggest alternatives.
Requirements and limits
On your side, it takes someone who can take decisions and bring the teams involved on board. I take on a limited number of engagements at any one time. What I don’t offer:
- Pure staff augmentation or hourly capacity without a project behind it.
- Writing tender or grant submissions end to end without subject-matter input from your side.
- Research on models or algorithms.
Objections
Questions people rarely ask me directly
Do you actually have time for us alongside your other engagements?
I only take on an engagement when I have the time for it, which is why I take on only a few at once. Consulting days and fixed dates are committed in the proposal. If my capacity is booked at the moment, I tell you up front and give you a realistic start date.
What happens if you’re unavailable?
Everything I do is documented as I go and can be handed over at any time, so someone with the right skills can pick up straight away. If you wish, I share the documentation through a client portal or another channel. For cover and larger projects I work with a small number of regular partners I have known for many years.
Are there conflicts of interest with your other engagements?
Confidentiality is guaranteed in every engagement: I bring experience and my own methods, but no documents from other organisations, and what is created for you belongs to you. I don’t take on strategic engagements with competitors of the organisations I work for. In our first conversation I’ll tell you confidentially who they are, so you can check for yourself. Any other possible conflict I raise openly as soon as I see it, and we decide case by case whether and how it can work.
Which AI tools do you use with our documents?
Only those your requirements allow, and I comply with the EU AI Act and the GDPR. I also keep data to a minimum: client names or personal data only go in when the task really needs them, even in an approved tool.
Do you ever advise against AI?
Yes. In many cases the real issue is still digitalisation: an unclear process, missing data, or existing software that isn’t being used properly. Then I recommend solving that first. I have worked on digitalisation and technology transformation for over 20 years, and AI is one of several tools.
Why not just hire a large consultancy?
It depends on what your project needs right now.
Better with me
- A focused phase in which something should get built
- Finding the right problem, trying things out and putting what works into operation
- Few people involved and short decision paths
Better with a large consultancy
- Many stakeholders and long rounds of alignment
- Many teams that have to be coordinated in parallel
- The focus is on steering a large, complex programme
Roughly what does it cost?
I work at day rates that are usual for experienced independent AI advisors. With no large-firm overhead on top, that is as a rule well below what large consultancies charge. You’ll get a concrete proposal with scope and fee after our first conversation.
Check for yourself
Let your AI check whether I’m the right fit
Don’t just take my word for it. The prompt starts a short preliminary conversation: your AI asks you a few questions about your project, the way I would in a first call, and then tells you honestly whether I fit and what might fit better.
Your answers in the conversation go to the AI provider you choose, not to me. For anything confidential, copy the prompt into your organisation’s own AI tool.
View the prompt
I want to find out whether Martin Breitsprecher, an independent AI advisor, is a good fit for a project in my organisation. Have a short preliminary conversation with me, the way an experienced, independent advisor would. How to go about it: - Start straight away with the first question, without any preamble: "What would you like to achieve, and what is getting in the way right now?" - Ask only one question at a time and wait for my answer. - After that, ask only as much as you need for an assessment, at most five further questions. For example: which problem exactly, how the work is done today, which data and systems are involved, who decides, what the timeframe is. - You don't need confidential details; a rough description is enough. - If I have already told you a lot, skip the questions that are answered. Then give me an honest assessment in about 250 words at most, without sales language: 1. Is he a good fit, a partial fit or not really a fit, and why? 2. What should I clarify with him beforehand? 3. What might fit better, for example a larger consultancy, an internal team or a software vendor? 4. What could a manageable first step look like? Be critical, even if the result goes against him, and treat the profile as a self-description, not as verified fact. If he fits, offer to draft a short email to him of two or three sentences. Profile (self-description, as of October 2026): - Independent AI advisor based in Hamburg, working across Europe, freelance. More than 20 years of technology initiatives, advising on AI and machine learning since 2016. - Services: AI strategy and operating model; architecture and delivery from pilot to operation; sparring on vendors, architecture and governance. Builds applications himself where it makes sense, otherwise helps with make-or-buy and vendor selection, vendor-independent and without commissions. - Leads the AI lab of an international consulting group as an external advisor (AI policy, fixed gates, register; 19 applications in the pipeline, 4 of them in production). Eight AI applications in production for an EU communications agency. Five years as lead requirements engineer for a large order management system. - Focus areas: EU institutions and donors, consulting and knowledge organisations, healthcare. - Fits focused projects with short decision paths; sees large consultancies at an advantage with many stakeholders or large programmes. Takes on few engagements at a time, commits consulting days in the proposal and works with regular partners for cover. No pure staff augmentation. Also advises against AI when something else helps more. - Detailed profile: https://breitsprecher.de/llms.txt - Email: martin@breitsprecher.de
If your AI suggests an email to me at the end, feel free to send it as it is.
07 / About
About
I studied information systems in Münster and Brisbane and have worked on technology initiatives in organisations for over 20 years, from enterprise architecture and digital transformation to AI strategy and delivery in regulated environments. What interests me most is the step from strategy into operation.
Since 2025 I have led the AI lab of an international consulting group as an external advisor: strategy, governance, and in-house applications. Alongside that I work with other organisations across Europe. Where it makes sense, I build applications myself and support their operation. Not every project needs custom development, though: where an existing solution fits better, I help with the make-or-buy decision and vendor selection.
08 / Contact
Tell me what you’re trying to achieve.
Two or three sentences are enough: what you’re working on, what’s getting in the way. I’ll reply with a first assessment and a suggested next step.
Thank you, your message has arrived.
I’ll get back to you within a few days, usually sooner.
Prefer email? martin@breitsprecher.de