Company: YoungCapital - Leading (temping) recruitment agency in The Netherlands
Case 1: Swapping a Costly Robot Call for a Simple Web Flow
Context
When I first joined YoungCapital, I was assigned the goal of improving the overall efficiency and reduce costs associated to recruitment. They thought AI could help as they were busy integrating a pricey vendor solution (€300K per year + extra).
I didn't want to find an excuse to use a technology. I wanted to bring value.
I stepped back to ask: what’s the problem we are facing, how do we test a solution fast, at scale, without adding friction for candidates or teams?
Problem
Recruiters took at least ~12min per candidate to ask simple and repetitive knockout questions. This meant that for high volume vacancies (thousands of applicants) it took hundreds of hours. Recruiters were also not storing the collected answers anywhere but their notes, so all these info would get lost.
Solution
I led my team to replace the complex AI vendor approach with a lightweight web form at the end of the application flow.
- Recruiters would select the questions only once when creating their vacancy.
- Candidates answered a short set of knockout questions in one go.
- Answers were stored in a structured way, so recruiters could make instant pass/fail decisions and route candidates faster.
- The same data also improved matching to better-fitting jobs when needed, using AI were it was actually adding value.
Outcome
~€300K p/y saved
Projected costs in using AI for robot screening which demonstrated to be bad fit for the task.
150× faster screening
Repetitive hours of work were freed up to be used in much more valuable tasks.
~89% completion rate
9/10 applicants would enrich their profiles with data which was used to successfully find a better match to 60% of the ones that didn't fit.
Company: Underlined - B2B SaaS company for AI-powered Customer Experience Analytics
Case 2: From Algorithm to Product
Context
As a part-time Data Scientist at Underlined, I spoke with a major airline that wanted a product to make sense of millions of customer feedbacks per month. They agreed to fund a research thesis and shared their data so we could build something that fit their use case.
The initial plan (and why it didn't fit)
Underlined's first idea was emotion mining plus a taxonomy-based topic classifier. I saw two blockers:
- Taxonomy maintenance: every new client/industry needed a new taxonomy and constant updates → not scalable.
- Emotion mining quality: peer-reviewed results (even from big players) were near random, so it wouldn't reliably flag real issues fast.
Reframing the problem
The airline didn't need "emotions." They needed to analyze all feedback at scale, know what each message is about, and spot negative spikes quickly so teams could act.
Solution
I fine-tuned state-of-the-art AI models to:
- Classify topics without a predefined taxonomy (works across clients and industries).
- Score sentiment (positive/neutral/negative) and focus on negatives to highlight what needs fixing now. The model reached ~85% accuracy and proved practical on the airline's real data.
Outcome
Underlined invited me to turn the algorithm into a product with a clear UI/UX so business users could run analyses and get reports in minutes. For the airline, analysis time dropped from weeks to minutes (≈95% faster), moving from manual samples to full coverage and faster, targeted fixes.
~95% faster analysis
Feedback analysis time dropped from weeks to minutes, moving from manual samples to full coverage and faster, targeted fixes.
Positive impact on CX metrics
Fast identification of spikes of negative feedback turned into actionable insights that led instant to resolution and positive impact on CSAT/NPS.
Takeaway: Start with the real job-to-be-done, then pick the simplest scalable approach. By replacing emotion mining and taxonomies with robust topic + sentiment models, we delivered a solution the client could trust and use at scale. This product was then ready for wider adoption by businesses around Europe and prove impact across industries.
Company: YoungCapital - Leading (temping) recruitment agency in The Netherlands
Case 3: Leading Enterprise AI Adoption During ERP Migration
Context
In Spring 2025, I stepped in as AI strategy lead. The company was busy migrating its ERP (including the Applicant Tracking System I worked on with my team) to a new platform, which meant that adding new AI features on software that was moving was not wise. However, we still wanted to use our time effectively, so we focused on something safer and more valuable for the specific situation we were in: making everyone in the company more efficient with AI.
Strategy
I applied enterprise AI transformation methods to YoungCapital:
- Capability maturity mapping: I assessed where we stood, where the biggest gaps were, and which moves would create the most impact fast.
- Company-wide priorities: upskilling at scale, safe, standardized tools, and clear guardrails aligned with EU regulations (including the upcoming AI rules).
Upskilling the workforce
Practical training and guidance so people knew what AI can/can't do, and how to use it in daily work.
Safe, consistent tooling
Selecting and rolling out approved AI tools with the right defaults, access, and documentation.
Policies & governance
Lightweight rules, examples, and checks so teams could move quickly without risking data, compliance, or brand trust.
Impact
- Drastical impact on time-to-delivery: reduced repetitive debugging time by 80% by integrating AI-assisted coding at scale.
- Lower cost of work: efficiency gains reduced time and effort per task.
- Efficient Workforce: ~700 upskilled employees experienced a boost in productivity thanks to e-learning and workshops on AI.
- Regulatory readiness: clear guardrails in preparation of enforcement EU policies.
Selected Case Studies