Insights
Field notes from the work
Field notes from real analytics-engineering work — migrations, validation, and the silent traps you only see when you rebuild a system from source.
Verdict 29 field notes from real rebuilds
Latest Building a Data Warehouse for a Restaurant Chain: What We Learned at Carl's Jr Mexico What a multi-location chain's data problem really is, what the warehouse architecture looks like, and what to demand from whoever builds yours — from a 100+ location build.
- dbt Cloud vs dbt Core: What Mid-Market Teams Actually Pay (2026) dbt Core has no license fee, but it is not free. Verified 2026 dbt Cloud pricing, the real cost of owning dbt yourself, and the third option most comparisons skip.
- dbt Constraints on Snowflake: Enforcing Primary Keys in a Database That Won't Snowflake declares primary keys but never enforces them. What dbt constraints, the dbt_constraints package, and tests each actually protect — and which you need.
- Fivetran vs Airbyte vs Custom Pipelines: The Mid-Market Ingestion Decision (2026) Verified 2026 pricing models for Fivetran and Airbyte, what self-hosting and hand-built pipelines actually cost, and the three questions that decide it for a mid-market team.
- dbt Migration Benchmark: What 377 Legacy BI Objects to 51 Models Actually Took Real numbers from one BI-to-dbt migration: scope, timeline, 15 defects found, 0.002% validated variance, 606% Year 1 ROI. A reference card, not a case study.
- Questions to Ask Before Hiring a Data Consultancy Six questions that turn a consulting promise into something you can verify — who does the work, whether scope and date are fixed, how they validate the numbers, and what you keep when they leave.
- What a Data Stack Diagnostic Actually Is — and When You Need One A data stack diagnostic is a fixed-scope, two-to-three-week assessment inside your real systems: full object inventory, lineage audit, AI-readiness, and a fixed-price quote for the rebuild.
- I tripled my Claude Code throughput — measured, with scripts Three changes brought my Claude Max quota from 'out by Thursday' to 'full week with headroom.' Audit, methodology, and open-source telemetry scripts.
- Why Mid-Market Companies Outgrow Their BI Tools (3 Signs) Between $50M and $150M, the analytics tools that worked smaller become bottlenecks. Three concrete signs you've outgrown them — and the mid-market fix.
- dbt Testing: 4 Categories That Catch Everything dbt testing across 4 categories — schema, data, unit, reconciliation — with real SQL and YAML from a financial migration.
- AI in Analytics Engineering: What Actually Worked AI cut analytics engineering time 60% on a financial data migration. What worked, what failed, and the guardrail framework for financial systems.
- 606% ROI on a BI Migration: The Full Numbers $840K return on a 5-month Snowflake + dbt migration. 90% faster dashboards, 86% fewer models, Finance self-service. Full cost-vs-return breakdown.
- Why Your Analytics Migration Is an Investment, Not a Cost Analytics migration delivers 606% Year 1 ROI when treated as strategic investment. Learn the business case framework for modernizing legacy BI.
- Finance Self-Service: Sigma Input Tables + dbt Seeds Finance was filing Jira tickets to change one pricing rate. We wired Sigma input tables to dbt seeds — now they update pricing with no engineering.
- Cortex Analyst on Real Financial Data: What Worked We ran Cortex Analyst against production-validated dbt revenue data — 51 models, 6 product lines. The semantic layer worked, with caveats worth knowing.
- 86 Charts to 22 Pages: Keep/Merge/Kill Framework Keep/merge/kill framework for dashboard consolidation: 86 charts with 60s loads cut to 22 fast pages. How to audit, decide what to kill, restructure.
- 377 Legacy Objects to 51 dbt Models: A Snowflake Migration Three-layer dbt architecture + domain isolation + a reduction framework that cut 70% of legacy BI objects without losing reporting coverage.
- Okta + Snowflake Auth: Security vs Developer Velocity EXTERNALBROWSER fails headless. OIE rate-limits automation. IP whitelists block VPNs. Real auth friction from a 5-month Snowflake engagement.
- Snowflake Native dbt Without dbt Cloud: What Breaks in Production We ran production dbt on Snowflake's native integration — no dbt Cloud, no local install. Four things broke silently. Here's the fix for each.
- When the Right Answer Is Deliberately Wrong Code A 15-month fiscal quarter bug found mid-migration. The correct move: replicate it, validate parity at 0.00%, then fix in a separate PR.
- The Hidden Tax of Snowflake RBAC (and How to Fix It) 17–26 hours lost to Snowflake permissions in 2 months. The fix: functional roles, future grants, and per-developer isolation patterns.
- Financial Migration Validation: 0.002% Accuracy Under $2,000 total variance on a multi-year revenue migration. Dual models, FULL OUTER JOIN at every grain, and parity-first discipline.
- 15 Silent Bugs in a Financial BI Migration 54% pricing discrepancy, a 15-month fiscal quarter error, and 13 more bugs found reconciling a financial BI system during migration to dbt + Snowflake.
- Scope Growth Done Right: 377 Objects to 161 Models How a dbt + Snowflake migration grew 5x without scope creep — domain isolation, validation gates, and a framework for saying yes vs. no.
- AI Guardrails That Outlast the Consultant A CLAUDE.md starter kit, pre-commit hooks, and domain context template — shipped on the last day so the team could use Claude Code safely without us.
- CLAUDE.md for dbt: What Goes in the File That Matters Most Three categories of CLAUDE.md content for financial dbt projects, plus OS-level hooks that prevent AI coding assistants from touching revenue data.
- Claude Code Breaks at 60%: The Five-File Fix 28 Claude Code sessions, 9 days, one dbt pricing rebuild. The context window degrades at 60% — here's the file-based memory system that prevents it.
- When Finance Has to Ask Engineering: Sigma + dbt Write-Back We replaced Jira tickets for pricing changes with Sigma input tables writing back to dbt seeds. Finance updates rates directly — architecture + audit trail.
- Why I Built Clarivant: The Mid-Market Analytics Gap After 15 years at P&G and eBay, I saw mid-market companies stuck on 60,000-row Excel files. Enterprise analytics shouldn't require enterprise budgets.
SIGNATURE PAGE · countersign this file
Bring us the data nobody trusts.
The strategy call is direct with the founder. We take the engagements we can lead end to end — which means we turn some down.
Book the call — and we'll defend these numbers on the record.
15 silent production bugs a migration surfaced
Book a 30-min strategy call
Direct with the founder. No pitch. Bring your messiest data question.
Not ready to book? Write to us: [email protected] A straight answer within one business day. Or read the questions buyers ask us →