1+3 Buyer Proof Kit · Snowflake · dbt · Cortex
Before scaling AI analytics, get a Metric Trust Map, validation findings, and a Go/No-Go recommendation in 5 delivery days.
When Snowflake Cortex or dashboards return conflicting numbers, a plain dbt build stays green. I audit 1–2 priority KPIs (read-only on prod), wire CI parity checks, and deliver a 3-page Readiness Report.
- Fixed scope · 50/50
- €4,500 HT
- 1–2 priority KPIs
- 5 delivery days
- Read-only on prod
- Go / No-Go
Entry step: the Readiness Check, 2 days, €1,200 HT, credited against the Sprint within 30 days. See the entry step ↓
€1,200 HT · 2 days · Paid upfront
AI Analytics Readiness Check
2-day independent review of the questions your AI analytics must answer. You receive a readiness report.
The €1,200 is credited in full if you book the Trust Sprint within 30 days.
Independent judgment and question-set design, not a Cortex Analyst audit or a Snowflake run.
AI Analytics Trust Sprint
Read-only on prod · Sandbox Semantic View · 50% deposit, 50% on delivery
Problem
Snowflake Cortex and BI dashboards return conflicting KPIs when dbt metric definitions drift or lack semantic-layer parity tests.
Intervention
5-day fixed-scope sprint on 1–2 KPIs: read-only prod audit, Metric Trust Map, sandbox Snowflake Semantic View, CI harness, and 3-page Readiness Report.
Evidence
3-page Sample Readiness Report (/sample-report & PDF on dbt-revenue-variance-demo commit 243d6f0: 4 silent dbt faults caught) + Auditor replay (176 offline tests).
Buyer decision
Receive a Metric Trust Map, validation findings, and a Go / Conditional Go / No-Go recommendation within 5 delivery days, so you can decide what to address before scaling AI analytics.
Compare Sprint (€4,500 HT) vs. Audit (€3,000 HT) & Day-5 terms
Flagship Engagement (AI + Metric Readiness)
AI Analytics Trust Sprint
€4,500 HT · 5 delivery days · Fixed scope · 50% deposit, 50% on delivery
Best when you want to expose 1–2 critical KPIs to Snowflake Cortex or text-to-SQL and need a Metric Trust Map, a governed sandbox Semantic View, repeatable validation checks, and an objective Go/No-Go decision first.
Standalone Entry Point (dbt & Lineage Only)
dbt Project Health Audit
€3,000 HT · Fixed scope · 50% deposit, 50% on delivery · Read-only Snowflake + manifest.json
Best when you are not deploying an AI assistant yet, but need a prioritized lineage blast-radius report and CI test suite to stop silent metric drift and untested semantic changes in your dbt project.
Next Action (20 minutes, no pitch deck)
On the scoping call, we review 1–2 target metrics and your Snowflake/dbt setup, confirm whether the 5-day Trust Sprint or the dbt Health Audit is the right fit, and agree on read-only access boundaries.
3 Supporting Proofs — Audit Toolchain, SQL Gate & Enterprise Track Record
Honesty boundary and primary proof link stay visible up front; expand any case for the Problem → Intervention → Evidence → Buyer decision detail.
dbt Project Health Audit
Standalone diagnostic mapping structural dbt defects and lineage blast radius when plain dbt build passes (ERROR=0) while revenue or KPIs silently drift.
Expand Case 1 detail (Problem → Intervention → Evidence → Decision)
Engineering Proof: Recorded Replay (jaffle-shop + 2 seeded defects) & Fictional Variance Repo · Static manifest.json analysis + read-only Snowflake metadata + lineage blast-radius report & CI tests
Problem
Unvalidated dbt model changes, missing refund/date-coverage guards, and duplicated metric logic pass a plain dbt build (ERROR=0) while silently shifting revenue or KPI figures in downstream dashboards and AI tools.
Intervention
A static and metadata audit joining your dbt manifest.json with a read-only Snowflake role to map every structural finding and its downstream blast radius onto the lineage graph, paired with known-answer and dashboard-vs-metric parity tests wired into CI.
Evidence
1) Snowflake & dbt Auditor live replay (dbt-model-auditor.simonsangla.com: 176 offline tests, lineage blast-radius view on public jaffle-shop with 2 seeded defects). 2) dbt-revenue-variance-demo (dbt-revenue-variance-demo.simonsangla.com & GitHub repo: 4 silent mutations move revenue from 53,546.30 EUR to 9,550.90 EUR, 35,950.20 EUR, or 61,606.60 EUR while plain dbt build stays green, then fail deterministically once the 4 semantic and data tests are added).
Buyer decision
Decide which structural dbt defects, unguarded metric definitions, and high-blast-radius lineage nodes must be fixed first—either as a standalone €3,000 HT remediation roadmap or as the foundation for the AI Analytics Trust Sprint.
MetricPilot — Governed AI Analytics Gate
Three-brick reference pipeline enforcing a semantic contract, fail-closed sqlglot read-only SQL gating (blocking DROP/DELETE), and row-level figure grounding.
Expand Case 2 detail (Problem → Intervention → Evidence → Decision)
Architecture Walkthrough: Prebuilt Scenarios on Synthetic Sample Data · Semantic contract · sqlglot read-only SQL gate · Row-level answer verification
Problem
Business and risk leaders cannot approve natural-language SQL assistants when an agent can execute destructive statements or narrate hallucinated numbers that do not match the returned query rows.
Intervention
An inspectable three-brick verification pipeline: a semantic contract that restricts metrics to approved definitions, a fail-closed read-only sqlglot SQL gate that parses and blocks any write/DDL statement before database execution, and row-level grounding where charts render directly from result rows and unverified figures are withheld.
Evidence
Interactive three-scenario walkthrough (/demo/metricpilot.html) and technical case study (/work/metricpilot): Scenario 1 shows governed metric retrieval with full SQL and source-table lineage; Scenario 2 proves fail-closed refusal on a destructive DROP/DELETE prompt (Blocked · 0 rows touched); Scenario 3 shows how ambiguous metric grain is resolved against the semantic contract.
Buyer decision
Evaluate whether fail-closed read-only SQL gating and row-level figure traceability satisfy your security and data governance bar before exposing Snowflake tables to natural-language querying.
TripAdvisor — Governed dbt Standards & Internal Text-to-SQL
Real enterprise production tenure (2021–2025): built a dbt-backed analytics template with model standards, tests, and self-service rules, worked on an internal text-to-SQL assistant, with KPI root-cause investigations cut from days to hours.
Expand Case 3 detail (Problem → Intervention → Evidence → Decision)
Enterprise Outcome: Published Operator Track Record (Distinct from Synthetic Demos) · Real production analytics engineering, self-service governance, and executive KPI triage
Problem
KPI root-cause investigations took days before they were cut to hours.
Intervention
Over four years at TripAdvisor (Lead Analyst → Senior Lead Analyst, 2021–2025), built a dbt-backed analytics template with model standards, tests, and self-service rules, and worked on an internal text-to-SQL assistant for non-technical teams.
Evidence
Published qualitative enterprise outcome from Simon Sangla’s 2021–2025 tenure at TripAdvisor: "KPI root-cause investigations cut from days to hours." Primary source: published operator track record on the main site (/#about); secondary corroboration: public career history on LinkedIn (linkedin.com/in/simon-sangla).
Buyer decision
Does Simon's track record—KPI root-cause investigations cut from days to hours, a dbt-backed analytics template with model standards, tests and self-service rules, and work on an internal text-to-SQL assistant—match your needs?
Why hire Simon vs. buying another tool
The demos on this page are diagnostic instrumentation—you hire me for engineering judgment and an evidence-based Go/No-Go recommendation.
Expand consulting methodology & 3-tier evidence classification rules
Software alone does not resolve conflicting KPI definitions between Finance and Ops. With 14+ years in SQL analytics and Snowflake/dbt in production (including TripAdvisor, Lead Analyst → Senior Lead Analyst, 2021–2025), I use the inspection harness and SQL gate shown here to audit your actual dbt lineage, write the missing known-answer and parity tests into your CI, build a governed Semantic View in a non-production sandbox or clone, and hand you an executive-ready Readiness Report. If a metric is not yet safe for AI rollout, the report states No-Go or Conditional Go with the exact remediation steps—protecting executive trust before you scale.
Evidence Classification & Honesty Boundary
Every claim on this page is tied to an explicit source category so you never have to guess what is a synthetic engineering demo versus a real enterprise outcome:
- Synthetic / Seeded Reference Repo: 100% fictional data (seed 42, commit 243d6f0) used to demonstrate the exact 3-page Readiness Report deliverable and 4 silent dbt failure modes without exposing client data.
- Recorded Inspection Replay & Walkthrough: Supporting diagnostic tools (Snowflake & dbt Auditor on public jaffle-shop with 2 seeded defects; MetricPilot SQL gate on synthetic sample data) used inside engagements—not sold as standalone SaaS.
- Enterprise Track Record (TripAdvisor, 2021–2025): Real production leadership experience (Lead Analyst → Senior Lead Analyst), strictly separated from synthetic demos and limited to ratified qualitative outcomes.
Ready to verify your own metrics?
Start with a 20-minute Root-Cause Scoping Call.
Bring 1–2 KPIs where dashboards disagree or where a Snowflake Cortex / text-to-SQL pilot is stalling. We will review your dbt and Snowflake boundary, decide whether the 5-day AI Analytics Trust Sprint (€4,500 HT) or the standalone dbt Project Health Audit (€3,000 HT) is the right instrument, and outline the exact Day-5 deliverables.
04 / The operator
14+ years in SQL analytics; Snowflake and dbt in production.
Independent analytics engineer in Lisbon. Fixed-scope work, no hourly billing.
Verify the track record on LinkedIn (opens in new tab)At TripAdvisor, 2021–2025
- KPI root-cause investigations cut from days to hours.
- Worked on an internal text-to-SQL assistant for non-technical teams.
- Built a dbt-backed analytics template with model standards, tests, and self-service rules.
Book the scoping call.
20 minutes to see whether the sprint fits your Cortex or text-to-SQL setup.
Fixed scope, capped budget.
- 20 minutes.
- You describe the use case; I say whether the sprint fits and what access it needs.
- You leave with a yes at the fixed price, or a no.
Response time: < 24h. I sign your NDA before the call. Reference on request.