Full-cycle analytics —
from raw data to impactful decisions
15+ years in B2B and B2C (EdTech, AdTech, Digital Creative Products). Combines an engineering background, classical statistics, ML, and AI to drive a lift in LTV/ROI/MRR. Analyzes user behavior and unit economics, identifies growth levers through the HADI cycle, and prioritizes hypotheses and backlog using RICE, MCDM, SCM, and causal business models. Builds and maintains cost-efficient analytics infrastructure — from data collection to data marts and a semantic layer, giving agents and people a trusted foundation for decision-making. Runs advanced experimentation and ships custom AI copilots.
Open to consulting and always up for a chat about your analytics challenges.
Experience
The core
of a modern data platform
The logic layer that canonically defines how metrics and dimensions are computed is the core of any mature analytics system. Too often this layer is smeared across visualisation tools or buried in analysts' expertise — which makes it barely governable and lets it accumulate contradictions.
A semantic layer lifted out of the BI system lets you govern metric logic uniformly, use any visualisation tool, and make self-service and AI agents an effective way to access data.
Unlike a Context Layer, which contains only descriptions, a semantic layer answers semantic queries deterministically. This radically improves reliability and verifiability — which in text-to-SQL approaches is fundamentally limited.
Truth by Design
Data trust has always been a critical aspect of BI and analytics — but with the rise of AI, it has become the defining one. Even if an AI gets 9 out of 10 answers right but the tenth is a silent lie, such a system is unfit for use. Every AI response must either be easily verifiable by the user or explicitly flag an error. With a pre-validated semantic layer in place, erroneous queries simply become inexpressible, and user-side validation is trivial: every chart is annotated in terms of the semantic model.
With AI, the value of software is dropping sharply across a number of segments: if you have a clear understanding of the concept and architecture, turning it into code is no longer the challenge. The value of many solutions now lies in the paradigm, not the implementation.
The good news is that the semantic layer and the AI scaffolding on top of it serve as integration glue — consisting almost entirely of methodologies and best practices that AI can tailor perfectly to each specific situation. Deploying off-the-shelf solutions is often still the more practical path, but with in-house expertise, AI-driven customization can win on multiple fronts: TCO, time-to-value, no vendor lock-in, flexibility, and control over data.
Below is a proof of concept for an AI agent that operates an OLAP cube loaded entirely in the browser. This approach radically reduces the load on the DWH and enables the agent to explore and analyze data quickly and autonomously — rather than simply rendering charts.
A map of the industry's likely breakthroughs —
where AI changes the rules of analytics
Business Intelligence is often reduced to nothing more than a visualization system, which fundamentally contradicts the original meaning and definition of BI as a Decision Support System (DSS).
In this context, the concept of Decision Science naturally emerges — a discipline that makes decision-making itself an independent subject of study and design. DS encompasses practices and methods that are critical to DSS but have never been part of BI, falling instead under the domain of business analytics rather than data analytics.
DS consists of the following broad classes of methodologies:
1. Strategic Planning & Foresight
- Macro-environment and trend analysis: PESTEL / STEEP, weak signals analysis, horizon scanning.
- Modeling the future: Scenario planning, backcasting (planning from a desired future back to the present), cone of uncertainty.
- Stress-testing decisions: Pre-mortem (analyzing causes of a hypothetical failure before launch), war-gaming (simulating competitor moves).
- Alignment: Strategic roadmapping.
2. MCDM
- Trade-off and balance analysis: Pareto optimality (trade-off analysis), weighted scoring models.
- Advanced evaluation algorithms: Analytic Hierarchy Process (AHP / ANP), matrix-based ranking methods (TOPSIS, PROMETHEE, ELECTRE).
- Sensitivity assessment of the choice: Weight sensitivity analysis (how the decision changes if the priority on price drops), utility theory / value function.
3. Prioritization Frameworks
- ROI-based scoring (investment/return): RICE, ICE, Impact/Effort matrix.
- Assessment through the lens of time and money: Cost of Delay, WSJF (Weighted Shortest Job First).
- Customer-value orientation: Kano model (identifying wow features), Opportunity scoring.
- Categorical and quick filtering: MoSCoW (must-have vs. nice-to-have), Eisenhower Matrix, separating Sequencing from Prioritization (what we do first vs. what matters most).
4. Quantitative Business Models & Simulations
- Financial and product logic: Unit economics, DCF (Discounted Cash Flow), What-If analysis.
- Business-system dynamics: System dynamics (Vensim), causal loop diagrams, stock-and-flow modeling (understanding how churn, acquisition, and LTV influence each other over time).
- Risk and probability assessment: Monte Carlo simulations (estimating the probability of hitting a sales plan), agent-based modeling (crowd/user behavior).
- Operational simulations: Digital twins, discrete-event simulation.
5. A/B Testing & Causal Inference
- Classical experiments (Controlled): A/B/n tests, multi-armed bandits, sequential testing, test quality control (peeking problem, SRM).
- Quasi-experiments (Causal Inference) — when an A/B test is impossible: DID (Difference-in-Differences), synthetic control, instrumental variables (IV), RDD, propensity score matching. Cannibalization assessment (spillover / interference).
- Effect personalization: Uplift modeling, CATE (identifying the segment most affected by a feature or offer).
6. Operational Optimization
- Allocating limited resources: Linear and integer programming (LP / MILP), constraint programming (e.g., optimally distributing an ad budget across 10 channels subject to constraints).
- Flow and queue optimization: Queueing theory (calculating load on a call center or SaaS support team).
- Route optimization: TSP, VRP (routing problems for field sales or delivery).
7. Decision Governance & Behavioral Science
- Accountability architecture (Who decides?): Decision-rights matrices (DACI, RAPID, RACI).
- Systematic thinking and reflection (How do we decide?): OODA loop (Observe–Orient–Decide–Act), decision journal (logging decisions to analyze mistakes in hindsight).
- Behavioral economics (Where do we go wrong?): Simon's bounded rationality, accounting for cognitive biases (anchoring effect, status-quo bias, sunk-cost fallacy), nudge architecture (designing choice environments to steer toward better decisions).