Fractional Head of Analytics · Snowflake
I find why a key business metric changed. Fast.
MetricPilot checks every answer against the underlying data. If the evidence is missing, it refuses to guess.
01 / The proof
See how MetricPilot checks an answer
Choose a sample scenario. Follow an accepted answer back to the query and result rows—or see the system stop an unsafe request before it runs.
Interactive walkthrough of prebuilt scenarios using synthetic sample data. It is not connected to a live customer system, and no customer data appears here.
Open the interactive demoNothing leaves your VPC
The whole system runs inside your Snowflake account. No external API ever sees a row of your data.
Traces to a row
Every figure links back to the query and the result rows that produced it. Charts are drawn from results, not narration.
Read-only by construction
The SQL gate admits read statements only — and shows you why anything else is refused.
Withheld when unverified
Any number the model states that is not present in the result rows is withheld, not displayed.
1/2Ask it to delete data and the gate refuses before execution: the statement is shown verbatim, nothing runs, zero rows touched.
2/2Every figure in the takeaway is checked against the result rows directly beneath it — same numbers, same table, nothing in between.
Demo runs on seeded data. In an engagement this sits on your tables, inside your Snowflake — nothing leaves your VPC.
At TripAdvisor, 2021–2025
- KPI root-cause investigations cut from days to hours.
- Built and scaled an internal text-to-SQL assistant for non-technical teams.
- Built a dbt-backed analytics template with model standards, tests, and self-service rules.
02 / The engagement
How an engagement runs
Fixed scope, capped budget — you know the deliverable and the ceiling before we start.
01 SPRINT
Metric Root-Cause Readiness Sprint
Your North-Star metric moved. I connect to your warehouse and isolate the root cause deterministically — variance decomposition tree + executive WBR brief in days.
Runs on MetricPilot — the system shown above.
- Variance decomposition tree
- Executive WBR brief
- 30-day Snowflake alert template
from €5,000
up to 5 delivery days
Everything ships as artifacts in your warehouse — SQL, dbt models, playbook, audit trail. Nothing runs on my infrastructure; your team operates it without me.
If the scoping call shows the cause cannot be isolated from your data, I say so before you pay — not after.
03 / The operator
14+ years turning data into decisions
Independent Analytics & AI engineer — 14+ years in SQL, Snowflake, and data modeling, forged in enterprise B2C environments where an unexplained 2% drop in conversion costs millions. I don't build generic dashboards and I don't bill by the hour to explore your data: I build deterministic systems that locate the leak, then hand your team the keys.
Verify the track record on LinkedIn (opens in new tab)simonsangla.com Jan 2026 – Present
Independent Analytics & AI Engineer
Building MetricPilot — governed analytics on Snowflake Cortex where every number traces back to a row: a semantic contract on your tables, safety-gated read-only text-to-SQL, and a full audit trail. Plus AI-native builds: custom Claude Code skills, MCP servers, and Next.js apps.
Tripadvisor Jan 2021 – Dec 2025
Lead → Senior Lead Analyst
Built the Insight Bot (text-to-SQL on Cortex Analyst) — an internal natural-language assistant for non-technical teams — and the Business Metric Wizard, an ML/NLP engine that cut KPI root-cause investigations from days to hours. Shipped Cortex AI to production, led the Eppo experimentation rollout and the Mixpanel→Snowflake event layer, and built a dbt model template with model standards, tests, and self-service rules.
Technical Skills
Snowflake (Cortex AI · Cortex Analyst · semantic layers) · dbt · SQL · Python · TypeScript · Next.js · MCP servers · root cause analysis · Eppo
Education
MSc Statistics (TIDE) & Econometrics · Université Paris 1 Panthéon-Sorbonne
Also shipped on Snowflake
Snowflake Native App
Skantle Scan
Finds wasted Cortex AI spend per actor — and refuses to show any figure it cannot re-derive.
Snowflake Native App · Pure SQL engine · Streamlit
Case studyConversational BI
Secure BI Chat
Plain-English questions, SQL-backed answers with automatic charts — behind a fail-closed read-only guard.
Streamlit · Cortex Analyst · sqlglot guard
Case studyStop guessing. Start fixing.
If your North-Star metric is bleeding and your current team cannot pinpoint the exact technical or behavioral root cause, we need to talk.
Read-only, single privilege, nothing leaves your VPC — a security pack built to clear InfoSec review in one pass.
Fixed scope, capped budget — no open-ended contracts.
- 30 minutes.
- You describe the metric move; I tell you whether it is auditable and what access I would need.
- You leave with a fixed quote or a no.
Response time: < 24h. NDA-ready. Reference available on request.