AI Consulting & Implementation
From idea to production-ready solution — structured, honest, no hype.
In short: I help Swiss companies make honest decisions about AI — and then deliver in production. That covers structured discovery workshops, 2-week PoC sprints, independent architecture reviews before large investments, and concrete vendor and hosting strategies for Swiss FADP/EU compliance. Four concrete offers, each with a clear deliverable, a clear timeframe, and a written outcome. No default retainer model. I say no when AI is the wrong answer for the question.
The four concrete offers
Clearly scoped, individually bookable, with a written deliverable.
AI Readiness Assessment (2–3 days)
Discovery workshops with stakeholders, mapping of existing data and processes, identification of 3–5 prioritised use cases with rough effort/value estimates. Deliverable: a written report with ranking, risks, and next steps — not a slide show, but a decision document you can keep working with even without me.
PoC Sprint (2 weeks)
One clearly scoped use case, built end-to-end: data wired in, model chosen, eval setup, runnable prototype. Includes handover documentation. The goal is not a demo but an honest verdict on whether the use case is viable in production with your current data — and what a production implementation would realistically cost.
Architecture Review & Second Opinion
Independent review of an existing AI system or a planned design before a large investment. Tradeoffs, risks, blind spots, concrete recommendations. Deliverable: a written review with prioritised findings. Typical triggers: before a vendor decision, before a go-live, or when a running project isn't delivering the expected results.
Vendor, Hosting & Governance Strategy
Which LLM vendors fit your compliance and cost requirements? Which hosting (Azure Switzerland, AWS Bedrock EU, on-premises)? Which guardrails are needed before going to production? Deliverable: a written decision record with reasoning — so an audit or a successor can later understand why this choice was made.
How I consult
I say no when AI is wrong here. Not every use case benefits from an LLM. A clean SQL query, a classical rules engine, or a simple workflow is the right answer in many cases. If that surfaces in the readiness assessment or in the review, I say so openly — even when it means a smaller follow-on engagement. An honest no is worth more in the long run than a well-packaged maybe.
Honest effort estimates, not aspirational ones. Estimates are calculated against what I have actually built — not against what looked possible in a demo video. If a PoC sprint needs two weeks, I say two weeks. If data quality has to be resolved first, I say that too — and don't push it into a follow-on contract.
No default retainer model. Every offer above has a clear deliverable and a clear endpoint. If ongoing support makes sense afterwards, a retainer can be added — but only if you want one, not because it's bundled into the default package. I don't manufacture vendor lock-in.
Written deliverables, not slideware. Every engagement ends with a document you can keep and reuse: report, decision record, architecture review, handover docs. So you can keep working without me — and so whoever comes after me finds a real foundation rather than a vibe.
Engagement Models
Flexible, transparent, and matched to the consulting need.
Project-based
Fixed price for a clearly scoped deliverable — readiness assessment, PoC sprint, architecture review. Ideal when scope and timeframe are clear.
Hourly/Daily rate
Flexible billing by effort. Fits exploratory discovery, ongoing accompaniment, or sparring sessions on concrete decisions.
Retainer (optional)
Monthly allocation for continued accompaniment after a project — only on request, not by default. Clear hours, clear notice period.
All prices are calculated individually based on scope and complexity. Concrete numbers come after a non-binding discovery call.
Request a quoteFrequently asked questions
When does an AI Readiness Assessment make sense?
When you know AI has to become a topic, but it's still unclear where the first concrete step lies. Typical triggers: a management mandate ("we need an AI strategy"), a vendor pitch you can't quite place, or a list of ten possible use cases without prioritisation. The readiness assessment delivers in 2–3 days a prioritised ranking of the 3–5 most promising cases, sorted honestly by effort and value — and says clearly which use cases are not a fit for AI. It isn't useful when the use case is already clear; in that case start directly with a PoC sprint.
How is a PoC Sprint different from a tech demo?
A tech demo shows that something works — usually on the happy path, with curated examples, no eval. A PoC sprint builds the use case end-to-end against your data, with an evaluation setup and honest error analysis. The deliverable is not just a running prototype but a written verdict: does the use case work in production with the current data? Where do the real limitations sit? What would a production implementation cost? The sprint is explicitly designed so that a clear "no, not worth it" is a valid result — and even then your money is well spent.
Can you also advise without delivering implementation?
Yes. AI Readiness Assessment, Architecture Review, and Vendor/Hosting/Governance Strategy are pure consulting engagements — the deliverable is always a written document, not code. You can book these offers even when you plan to implement internally or with another partner. That is explicitly part of the model: no vendor lock-in, no hidden pull towards "and now we also have to build it". If you want to do a PoC with me after the consulting, gladly — if not, equally gladly.
What does an Architecture Review cost?
An architecture review is calculated project-based and typically sits between CHF 5,000 and CHF 15,000, depending on scope and complexity: a single RAG setup or a planned design lands at the lower end, a running multi-agent system with several data sources and integrations at the upper end. The deliverable is a written review with prioritised findings, tradeoff discussion, and concrete recommendations — not a generic best-practice PDF. Precise quotes after a non-binding discovery call in which we scope the review together.
How do you measure the ROI of an AI project?
Honest answer: ROI on AI projects can almost never be calculated seriously before the PoC — anyone showing a spreadsheet with five decimal places is selling storytelling. What can be measured: time saved per case (stopwatch, before/after), error rate against the human baseline, throughput time, volume per employee. My approach: before the PoC, fix the two or three metrics that matter; during the PoC, measure the real numbers; afterwards, calculate honestly. If the ROI doesn't hold, that is itself a result — and a better one than keeping an AI project alive for two years because nobody is measuring.
Concrete Applications
Sales / Support Triage Agent
Classify inbound, draft replies, route to CRM — with approval queue.
Offer Generator Agent
Offers from CRM + catalog + tone-of-voice profile as draft — sales rep does the final polish.
Meeting → Action Agent
Transcribe meetings, extract decisions and action items, sync to project tools.
An honest assessment instead of a pitch deck?
Book a non-binding 30-minute discovery call. We figure out whether a readiness assessment, a PoC sprint, an architecture review, or a vendor strategy is the right entry point — and what effort it realistically involves.
Book a discovery call