How to transform data into €140 million, build data competence on a large scale – and know when it's better not to listen – with Filip Vitek
A conversation between Maximilian Hahnenkamp (Scavenger AI) and Filip Vitek (EVP AI & Data, CommentSold)
Since studying, a common thread runs through Filip Vitek's career: data as a competitive advantage – from individual contributions to team leadership all the way to the C-suites. In the podcast with Maximilian Hahnenkamp, Vitek discusses how Data/AI creates verifiable cash, why many AI implementations fail, why decision-making ability is the new data literacy, and which industry disruptions are realistic in the next three to five years.
Two examples that can turn a company's course
Monetary Impact (TeamViewer): Behavior recognition between free-tier and licensing needs. Precise triggers → Conversion tool → €140 million additional revenue in four years.
Unexpected Signals (Allianz): Open feature search for vehicle pricing – even for exotic variables. Result: zodiac signs as the second strongest predictor for claims – replicated in several countries. The point: Data opens perspectives that no one comes up with "off the top of their head".
Where AI is today essential – and where it is still hype
Mature & „no-brainer“
Audio transcription & meeting minutes
Document summaries (RAG/„closed book“)
Product recommendations in commerce
Active ingredient/molecular research in pharma
Exaggerated expectations
Humanoid robotics in everyday life
„One-person companies“ as the norm (teams are getting smaller, but not that small)
„Talking to animals“ – nice sci-fi, currently not reliable
Why so many AI projects fail despite the tools
„The big labs deliver components, not finished solutions – and take no responsibility for your use case.”
Vitek's analogy: Early electrification brought individual devices (microwave, washing machine) that did not communicate with each other. The same applies today: great models, but isolated. Those who let them run rampant in the “open” will fail due to unpredictable user behavior, variance, emotions. The next leap occurs when companies take orchestration, guardrails, and integration seriously – instead of just clicking tools together.
Data Literacy 2025: Unlearning SQL, learning to decide
Vitek breaks down data competence into three components:
Extraction (formerly SQL)
Interpretation
Decision & action
Forecast: 1 & 3 will be strongly automated by AI (query, visualization). What differentiates is whether teams
can validate AI conclusions (plausibility, bias, causality), and
can derive good decisions and consistent actions.
Will data still be an advantage in 2025?
Yes – micro: Own customer, usage, process data remain moat.
Probably not – macro: Open web/syndicated data, synthetic populations, etc., level informational advantages. Uniqueness arises from proprietary business & customer data.
Three highly underestimated use cases
Systematically scraping competitive data: Prices, assortments, launches, newsletters – legal, fast, effective.
Video content at scale: Converts better than static; production is now cost-effectively automatable.
Automating the hiring pipeline: Parsing, pre-screening, skills assessments, interview summaries – enormous time and quality levers.
Leading in Data/AI: Self-disruption as a requirement
Technology cycles are short; stagnation = regression.
Discipline: Those who demand data must decide based on data themselves – including transparency.
Allowed deviation: Data shows the optimal – document & evaluate conscious deviations (when did intuition win?).
Why many companies despite their commitment are not data-driven: Impostor syndrome (“say but not do”), power distances (disagreement is uncomfortable), and lack of measurement: What is not measured cannot be managed or improved.
Vitek's 3-point plan for executives
Store everything (orderly): Digital storage of every meaningful data point; retention policies instead of forgoing.
Auto-crunch instead of pull: Reports, alerts, dashboards push – data works for the decision-makers.
„Which data point supports this?” – as a default question in meetings; allow deviations, but mark them and evaluate retroactively.
Hype cycle: Peak? Yes. Beginning? Also.
Expectations are partly above the results – with the current approach. If the orchestration changes, a 10- to 100-fold benefit is possible. Without a change in direction, disillusionment threatens.
3-5 year outlook: realistic, not magical
AI movies on demand: Fully AI-generated movies – with selectable actors per user.
Universities reimagined: Digital professors, adaptive curricula, certification beyond traditional pathways.
Autonomous driving scales: Robot taxis/trucks come into mass operation (hardware turnover needed, but momentum is there).
Career advice in the end: Become top-tier
„AI makes the best a bit better – but the average suddenly a lot better. Those who stay in the middle risk being replaced.”
Upskill now: Domain knowledge, decision quality, systems thinking, AI orchestration. Then AI works with you, not against you.
Conclusion:
The next leap in productivity will not come from “more models” but from better choreography: utilize proprietary data, lead use cases closely, make consistent decisions – and turn data into the servant of the business.