01
Decision & Metric Audit
Trace the numbers that carry a decision, reconcile competing definitions, and identify the weakest evidence before it becomes a roadmap commitment.
Decision map · metric definitions · evidence audit · repair plan
Independent decision & analytics advisory
Make the decision clear before making the dashboard bigger.
Roz Intelligencia helps product and operational teams turn ambiguous questions, disputed metrics, and uncertain rollouts into credible evidence and a clear next move.
When the signal is unclear
Clarity, not more reporting.
Leadership metrics disagree across teams.
A rollout is moving faster than its measurement plan.
An ROI claim is persuasive but difficult to trace.
Important logic lives in notebooks, spreadsheets, or BI.
Services
Focused engagements for consequential decisions.
Each engagement starts with the decision and ends with an explicit action, boundary, and owner—not a folder of unexplained analysis.
02
Experiment & Impact Evaluation
Design a credible test before launch—or evaluate a rollout with the strongest counterfactual the operating environment can support.
Success criteria · guardrails · power check · decision-ready readout
03
Operational Diagnostic Deep Dive
Turn a moving metric and competing explanations into a structured diagnosis, evidence-backed priorities, and a focused next test.
Hypothesis tree · decomposition · evidence gaps · action options
04
Metric-to-Model Productionization
Move business-critical logic out of one-off analysis and into governed data models and semantic definitions teams can reuse.
Grain and contract · dbt model · semantic metric · handover
From the journal
Two useful notes for better decisions.
Practical thinking on experimentation, evidence integrity, and sustainable measurement.
Experimentation 6 min
Why a Negative Experiment Can Be a Successful Decision
A negative result can protect a team from scaling the wrong change—if the experiment was designed around a real decision.
Read Fraud Detection 4 min
Identifying Fake “New User” via Promotion to achieve Sustainable Business Growth
A practical way to identify abusive activity from promotional users before it becomes a sustainable-business problem.
ReadWhy Roz Intelligencia
Evidence that can survive the meeting.
Decision before dashboard
The question, owner, and available actions are defined before the method or visual.
Measurement before rollout
Primary metrics, guardrails, power, and decision rules are agreed before results arrive.
Useful negative results
A well-designed pause or no-change decision can protect more value than a flattering readout.
Definitions built to last
Stable logic moves upstream so the answer survives beyond one analyst or dashboard.
The engagement method
From an ambiguous question to an operating decision.
- 01
Clarify
Name the decision, owner, deadline, affected population, and available actions.
- 02
Audit
Trace definitions and evidence. Make weak inputs and missing data explicit.
- 03
Design
Choose the analytical, experimental, or data-model approach before seeing results.
- 04
Build & evaluate
Create reproducible analysis or governed measurement with limitations attached.
- 05
Decide & operationalize
Turn the evidence into a recommendation, ownership, monitoring, and learning loop.
Selected proof themes
Work that changed the next move.
Examples are deliberately sanitized to protect employer and operational confidentiality.
- 01 Stopping an AI rollout when a controlled canary showed meaningful regressions.
- 02 Finding value hidden by a flat raw trend through a credible counterfactual.
- 03 Challenging an attractive ROI case when its load-bearing input could not be reproduced.
- 04 Repairing a metric definition before it shaped production decisions.
- 05 Moving repeated reporting logic into governed, reusable measurement assets.
About
Product analytics with an engineering spine.
Roz Intelligencia is an independent decision and analytics advisory. The work combines product analytics, experimentation, operational diagnosis, metric governance, and analytics engineering.
The recurring pattern is simple: clarify the question, audit the evidence, choose a credible design, build what must be reusable, and leave the client with a decision they can defend.
No ROI claim should be stronger than its weakest load-bearing input.
Start with the decision
Bring the question as it actually is. We will work out what evidence it deserves.