Miha Volovšek — AI that is proven on your business

Three careers. One timeline.

Most people who sell “strategy and AI” added the AI half after 2022. Mine started in a research lab in the 2000s, on top of a transformation practice that started in the 1990s, on top of an IT career that started in the 1980s.

Information technology

40 years

since 1986 Business transformation

30 years

since 1996 BI, DWH, ML, data mining, applied AI

20 years

since 2006

Strategy, business-model and process redesign for 30 years. Applied machine-learning research at the Jožef Stefan Institute AI Laboratory for 15 of them. Production AI systems, in the cloud and on sovereign on-prem stacks, today — and, on the other side of the table, training and evaluating frontier models for the world’s AI data platforms. The combination is not a slide. It is the reason I can hold the whole chain — from the board decision to the model score — in one pair of hands.

What you are probably dealing with

Five sentences I hear in boardrooms in Ljubljana, Vienna and Zürich. If one of them is yours, we should talk.

We have Copilot licences, three pilots and a vendor proposal. Nobody can tell me what any of it is worth.
You have AI activity without an AI portfolio. The fix is a scored list, a baseline and a funding sequence — not another pilot.
The pilot worked. Then it met the real process and fell over.
The model was pointed at a process that was never designed for it. Redesign the process, define what the model may decide, then deploy.
The answers are generic. It doesn’t know how we actually do things.
Your AI has never been taught your business — your documents, rules, exceptions and vocabulary. Grounding it is engineering work with a measurable result.
Half the company is pasting customer data into public chatbots and I have an AI Act deadline.
Shadow AI is unmet demand. Governance that names owners and gives people a sanctioned, compliant tool ends it. A policy PDF does not.
The strategy firm wants €400k for slides. The integrator wants a two-year programme. I want a result this fiscal year.
One accountable senior adviser who does the strategy, the redesign and the build — with part of the fee released only when the score is in.

The book

Book cover: Your AI Doesn’t Know Your Business by Miha Volovšek

Everyone can buy AI. Few companies know how to train it.

Your AI Doesn’t Know Your Business is about the phase nobody budgets for: after the licence, after the pilot, when the model meets the real process and answers like a stranger. It sets out how a company teaches a model its own rules, exceptions and vocabulary, who has to do that work, and how to score whether it worked — the PROOF method in full.

Written for the executive who signs the invoice and the person who has to make the system perform. Publication 2027.

How the work runs

PROOF is my performance-management method for enterprise AI. On one screen it comes down to three things, and every engagement below is a slice of them.

Diagnose

Map the business model and the processes that carry the margin. Score every AI candidate on value, data readiness and risk. Set the baseline everything will be measured against. Decide what to stop.

Redesign

Rebuild the process around what a model can reliably do. Decide which decisions it may take and where people stay in the loop. Ground it in your own documents, rules and exceptions. Deploy in the cloud or on a sovereign on-prem stack.

Prove

Measure accuracy, consistency, coverage and containment against the baseline. Release funding in tranches at four gates. Leave behind a trained internal owner and an evaluation set your company keeps.

Why it works: every step produces a number the next step is judged by. That is what turns AI from a project into a managed function — and what lets me put my own fee on the result.

Engagements

A ladder, not a menu. Most clients start at the first rung; companies in a strategy cycle start at the fourth.

AI Value Scan

In five weeks you know where AI belongs in your business, what it is worth, and what to stop.

For
CEOs and owners with several AI initiatives and no basis for choosing between them.

Result
A ranked, costed AI portfolio, a board-ready decision, and the baseline every later result is measured against.

Format
Five to six weeks. Fixed fee. Executive workshops plus desk analysis.
Redesign & Prove

One process, redesigned around AI, in production, with a published score.

For
COOs, CFOs and unit heads who own a process with visible cost or cycle-time pain — received invoices, tendering, order handling, technical support, sales operations.

Result
A production system scored on accuracy, consistency, coverage and containment; a trained internal owner; an evaluation set you keep.

Format
Three to six months. Fee in tranches at gates; the final tranche is tied to the measured score.
Fractional Chief AI Officer

Board-level ownership of AI without a board-level hire.

For
Companies of 200 to 2,000 people where AI is nobody’s job and therefore everybody’s side project.

Result
A governed, measured AI function — portfolio, scorecard, AI Act and data compliance, vendor decisions — and a named internal successor within twelve months.

Format
Two to four days a month, six to twelve months, monthly retainer. Several mandates delivered and running.
Strategy with AI inside

A 2030 strategy in which AI changes the business model, not just the cost base.

For
Management boards and owners of mid-sized industrial and energy companies facing a strategy cycle.

Result
Strategic analysis, value proposition, business model, scenarios and process architecture — with AI as a structural variable, and a measurable programme behind it.

Format
Three to five months, three phases, workshop-led.

For boards and owners: AI due diligence on acquisitions and carve-outs, 100-day AI plans for portfolio companies, second opinions on vendor proposals and AI Act exposure. Retainer or per case.

Which one fits — a 30-minute call decides

Evidence

Outcomes first. Then the programmes and the roles that produced them.

+20%

Value added per employee in year one, leisure-equipment manufacturer, after delivery-process redesign.

≈10×

Return on the consulting fee, food-industry sales-process consolidation.

In production

AI agent handling received invoices for an energy-sector operator, recurring annual cost reduction.

Fractional CAIO

Several mandates, governed and measured AI functions built.

Programmes

Corporate strategy 2026–2030 for a national energy research and engineering institute — three phases, four scenarios.
Received-invoice process optimisation for an energy company.
AI strategy and use-case programme for a plastics manufacturer in an Austrian industrial group.
Job architecture and compensation systems for Slovenian manufacturers.
Strategy and process work for clients in industrial goods, energy and consumer goods.

AI, hands on

Cloud LLM production deployments; LLMs with a custom harness; small-language-model deployments.
AI evaluation and RLHF — trainer and evaluator for frontier-model programmes across micro1, Mercor, Turing, Outlier, Toloka, Ethos and OpenTrain AI; rubric design, graded evaluation, domain: management consulting and enterprise AI.
EU AI Act, GDPR, NIS2, Data Governance advisory for regulated and industrial clients.
PROOF — Enterprise AI Performance Management: PROOF Score, eight-link chain, four gates.

Background

Founder and CEO, aggregata (Ljubljana, since 2003) — 150+ projects for 50+ clients across Central Europe.
Research Fellow, AI Laboratory, Jožef Stefan Institute (2002–2017) — applied ML and knowledge discovery, EU projects.
Visiting Lecturer, University of Ljubljana (2000–2015) — IS design, BI, data governance, IT strategy, process redesign.
Built a consulting unit inside a BI/DWH software vendor from zero to 20 management consultants.
Dr.techn., Dipl.-Ing. Electrical Engineering with Business Studies, TU Graz, Austria.

About

I started in IT forty years ago because I wanted to understand how information actually moves through a company. The question never left. It took me through process redesign, strategy work for industrial and consumer-goods clients, and in 2003 into founding aggregata, an implementation-led transformation firm in Ljubljana that has since delivered more than 150 projects for more than 50 clients across Central Europe.

In parallel I spent fifteen years as a Research Fellow at the AI Laboratory of the Jožef Stefan Institute, working on applied machine learning in European research projects, and fifteen years teaching information-systems design, business intelligence and process redesign at the University of Ljubljana. I learned, from both sides, what a model can be taught about a business and what it cannot.

That is why the current wave of enterprise AI looks familiar to me — and why its failures do too. Companies buy capability and expect it to know their business. It doesn’t. Someone has to redesign the process, decide what the model may decide, ground it in the company’s own rules, and measure whether it performs. I do that work myself: today with an AI agent in production for an energy-sector client, with sovereign on-prem stacks where regulation requires it, and with the evaluation and RLHF work I do for the world’s AI training platforms, where my domain is management consulting itself. My book on this, Your AI Doesn’t Know Your Business, is out in 2027.

I work in Slovenian, German, English and French, and I put my fee where my method is.

AI that is proven on your business. Not promised.