Metro Decisions

Los Angeles · Orange County

Right-size your data.

Grow when you need to. Cut when you don't.

I'm Michael Bowen. I help small and mid-sized businesses in LA and OC build the data platform they actually need — and rip out the one they don't.

Forty years in data engineering. Xerox, Pilot Software, Hyperion, Boeing, Qantas, Guess, Teva, OpenText. Now local, focused, and lazy on your behalf: the smallest working system, running where you can afford it, owned by you.

Book a 30-minute fit call → What I do

Upscale

You've outgrown spreadsheets.

You need a real pipeline, a real warehouse, and reporting that doesn't take a week to change.

  • Data platforms that fit the business. DuckDB, Postgres, S3, dbt — modern, columnar, cheap to run.
  • Pipelines you can trust. Producer/consumer with control tables so you know where every number came from.
  • AI where it earns its keep. Agentic workflows for specialist, repetitive work — not chatbots for their own sake.
  • Ontology-driven models (Kermadec) so your vocabulary lives in one place and drives everything downstream.

Downscale

You're paying for software you don't use.

Legacy Essbase, an over-scoped Snowflake bill, a BI tool with three active users, a SaaS stack that's out-of-hand.

  • Migration audits. What you have, what it costs, what it actually does.
  • Modern replacements. DuckDB + Polars + Postgres often turns a $50K/yr line item into a $500/yr one.
  • Local-first, code-first. Your data on your infrastructure, in files you can read, in SQL you can grep.
  • Retire dead vocabulary. If nobody's using the report, we archive it. If nobody opens the dashboard, we delete it.

Products & practice

The tools behind the work.

kermadec Framework

Ontology-driven data engineering.

Describe the domain once in a shared ontology; Kermadec emits the dbt project, the schema tests, the Arrow flight schemas, and the lineage database.

  • mdu emit --target dbt-project — model your business, get a running warehouse.
  • SHACL → dbt schema tests — validation rules that travel with the data.
  • Merged lineage in .kermadec.db — every column has a story.

The "left of go" work: get the definitions right, and the code writes itself.

metro-method Pattern

Producer/consumer pipelines.

The pattern I've been building since 2016. S3 as the permanent record, DuckDB as the query-in-place engine, control tables for provenance.

Every ETL step named, tracked, replayable: Capture, Ingest, Cleanse, Stage, Merge, Calc, Report, Distribute.

Modern stack, timeless pattern. Works for a five-person team; scales when you grow.

The modern SMB data stack

Storage:
S3 or local disk. Cheap, versioned, encrypted.
Warehouse:
DuckDB for analytics, Postgres for operations.
Transform:
SQL first, Polars where it earns it, dbt for orchestration.
Pipelines:
Go for producers, Python for consumers, dlt for connectors.
AI:
MCP servers, agentic workflows, LLM where domain reasoning pays.
Cost:
Usually 5–10× less than the enterprise alternative.

Whitepapers

Read before you buy.

Short, opinionated papers on the SMB data stack. Drop your email; I'll send the PDF and one follow-up. That's it.

wp-istd

ISTD vs Medallion.

Idempotent, Set-Transformational, Denormalized design — six named layers with unambiguous roles, versus Bronze/Silver/Gold's three-layer shorthand. Which one you want depends on where business logic lives and who is on-call at 3am.

wp-metro-method

The Metro Method.

Producer/consumer pipelines with control tables. The pattern I've been running since 2016 — how it works, why it lasts, what it costs.

Who I work with

Small and mid-sized businesses in LA and OC.

Retail, media & entertainment, real estate, hospitality, professional services, healthcare — that:

  • — Have outgrown spreadsheets but can't justify a full data team.
  • — Are paying enterprise SaaS prices for tools built for the Fortune 500.
  • — Want AI in their workflows without hiring a research lab.
  • — Need one experienced practitioner, not a five-consultant slide deck.

If you're a two-person startup or a fifty-person shop and your data situation is "somebody knows how it works," we should talk.

About

Forty years. One laptop.

I've been doing this for forty years — from Xerox in El Segundo and Pilot Software in Cambridge, to Hyperion, Rolta, Hackett, Solver, Full 360, OpenText, and now back to Metro Decisions, which I founded in 2001.

I've architected 42-node Vertica clusters at Guess, benchmarked AWS Vertica + Tableau at Teva, run near-real-time ETL across five time zones at Nissan, built the global engineering financial model for 17,000 Boeing projects, and integrated regulatory pilot-fatigue rules at Qantas.

That work taught me two things:

  1. Most companies own more software than they use. Complexity is the default; simplicity takes work.
  2. The best code is the code never written. The best system is the smallest one that solves the problem.

I'm currently converting decades of Ruby, Perl, and enterprise-stack work to Go, Python, DuckDB, Iceberg, Ducklake, Temporal, and MCP. Not because it's fashionable — because it's cheaper, faster, and fits on one laptop.

Based in Los Angeles. Available across LA and Orange County; remote for the rest.

1. Fit call

30 minutes, free. You describe what you have and what hurts. I tell you honestly whether I'm the right fit.

2. Assessment

1–2 weeks, fixed fee. Inventory your stack, cost it out, one-page recommendation: keep, cut, add.

3. Delivery

Fixed-scope projects or monthly retainer. No long contracts, no minimums beyond the assessment.

Book a fit call

Tell me what hurts.

Drop your email and the thing you're trying to fix. 30 minutes, no slides, no pitch. I'll tell you honestly whether I can help.

No newsletter. No share. One reply, from me, to schedule the call.