Library

The reading path

Everything the lessons and labs point back to, in the order it's worth reading. Four parts, about an hour end to end. Dip in from a lesson, or read it straight through.

Part 1 · Start here

The role and the product

Before the frameworks, two definitions that everything else hangs on: what a Data Product Manager actually owns, and what makes a data asset a product rather than a dataset, a dashboard, or a pipeline. Then the four ways this job inverts ordinary product management, and a ladder for judging how mature a data organisation is.

  1. 1What is a Data Product Manager?A DPM sits at the intersection of data, technology, and business, managing data warehouses, platforms, pipelines, and ML models the way a PM manages a product.2 min read
  2. 2What a Data Product Is (and Isn't)A curated, reliable, reusable data asset designed for ongoing use, with an owner, service expectations, and embedded quality. Not a dataset, not a dashboard, not a pipeline.3 min read
  3. 3The Four Key Shifts (Data PM vs. General PM)Risk profile, success metrics, customer journey, and team dynamic all invert when you move from general PM to data PM.1 min read
  4. 4Data Product Journey: Four Maturity StagesMost orgs sit in Ad-hoc or Reactive. The goal is Purpose-Driven: initiatives measured by clear North Star, functional, and granular metrics.1 min read

Part 2 · The Playbook

Six weeks, one framework a week

The 6-Week Data Products Playbook in reading order. Week 1 finds the problem and the metric tree. Week 2 turns definitions into a model. Week 3 activates it with ports and SLOs (and the self-service layer that makes that affordable). Week 4 launches where users already are. Weeks 5 and 6 prove it with usage, feedback loops, and root cause. Meridian's story follows the same arc.

  1. 1The Bullseye: Data Product-Market Fit (Week 1)Before building anything: retrieve requirements from real end users, map their existing journey, and identify North Star/functional/granular metrics.1 min read
  2. 2North Star, Functional, and Granular MetricsNorth Star = the domain's leading business metric. Functional = the leading metric for a sub-area. Granular = an indicator of effectiveness.1 min read
  3. 3The Canvas: Data Product Design (Week 2)Build a Metric Dependency Tree and the logical data model (entities, dimensions, measures, relationships, SLOs, semantics) that powers it.1 min read
  4. 4Activation: Input Ports, Transforms, Output Ports, SLOs (Week 3)A data product activates once four pieces exist: input ports, transformation steps, output ports, and SLOs — ideally declared, not hand-built.1 min read
  5. 5Self-Service Infrastructure (SSI)Ready-to-use resources (policy, monitor, workflow, contract, service, secret, compute) that let engineers declare intent instead of building plumbing from scratch.1 min read
  6. 6The Launch: Go-to-Market for a Data Product (Week 4)A data product with no adoption strategy fails like any other product. Position it in the user's language, and make usage effortless.1 min read
  7. 7The Proof: Product Performance (Weeks 5–6)Post-launch is where you spend the most effort: usage metrics, A/B tests, and feedback loops that evolve the SLO model and expand use cases.1 min read

Part 3 · Hands on

Working with the data

The formulas and the habits. Two cheat sheets for SQL and pandas, the short list of SQL a Data PM actually writes, how to think about data quality as promises you can check, how to read a funnel three different ways, and where a metric physically lives in a medallion architecture. Every Lab exercise links back to one of these.

  1. 1SQL 101: the formulasThe clause order, filters, aggregates, GROUP BY/HAVING, and joins that every exercise in this app is built from.2 min read
  2. 2Python & pandas 101: the formulasLists, dicts, and the DataFrame moves that mirror SQL: select, filter, group, sort, merge.1 min read
  3. 3The SQL a Data PM Actually UsesCTEs for readable steps, window functions for per-row context, date bucketing for trends, and the four bugs that silently corrupt metrics.2 min read
  4. 4Data Quality Dimensions (and how to write an SLO)Six dimensions — completeness, uniqueness, timeliness, consistency, validity, accuracy — each becomes an SLO once you name the measurement, threshold, and check time.1 min read
  5. 5Funnel Analysis: Overall vs. Stage-to-Stage, Cohorts, VelocityThree different funnel questions need three different queries: overall conversion (by cohort), stage-to-stage conversion (from the history log), and time in stage (velocity).1 min read
  6. 6Medallion Architecture: Bronze, Silver, GoldA three-layer pattern for structuring a data lakehouse: raw ingestion (Bronze), cleaned/augmented (Silver), business-level aggregates (Gold).1 min read

Part 4 · Context

The world in 2026

Why this role matters more now. Agents act on business terms without a human to reconcile meaning, so ontologies became infrastructure. Monitoring tells you an agent ran, not that it was right. Cost per correct answer replaced raw capability as the number a board reads. And one manufacturing case study that shows the whole pattern in a different industry.

  1. 1Ontology as Operational Infrastructure for AI AgentsAs LLM agents act autonomously, the ambiguity humans used to resolve silently (what is 'revenue'? what is a 'customer'?) becomes a hard failure point.1 min read
  2. 2Agent Observability ≠ Agent CorrectnessMost teams running AI agents have monitoring (did it run?) but not evaluation (was it right?) — and standard software testing assumptions don't hold for agents.1 min read
  3. 3Lean AI: Matching Model Size to Task DifficultySmall language models can be ~100x cheaper than large ones per conversation; mature teams route by difficulty rather than defaulting to the biggest model.1 min read
  4. 4Case Study: Multi-Factory OEE & KPI MonitoringOEE = Availability × Efficiency × Quality. Legacy/manual OEE computation breaks at multi-factory scale; a medallion pipeline fixes latency and drill-down.1 min read

Part 5 · The market

Tools of the trade

The names you'll hear in every interview and vendor pitch, organised the way the market is: a map of the stack first, then the warehouses and lakehouses where SQL runs (Snowflake, Databricks, BigQuery, Redshift, Fabric), the tools that move and shape data (Fivetran, dbt, Airflow, Kafka, Informatica, Talend, Ab Initio), the BI and semantic layers people actually look at (Tableau, Power BI, Looker, Qlik, Redash), the quality and governance tools that turn SLOs into monitors (Monte Carlo, Great Expectations, Soda, Collibra, Alation, Atlan, DataOS), and the ML platforms. Each entry says what the tool does, how you'd use it in your first hour, and what a Data PM asks about it.

  1. 1The Modern Data Stack: a MapWhere each category of tool sits between a source system and a decision, which vendors lead each category, and which pieces of Meridian's stack they correspond to.2 min read
  2. 2Warehouses and Lakehouses: Snowflake, Databricks, BigQuery, Redshift, FabricThe platforms where data lands and SQL runs. How each one thinks, what it charges you for, and the questions a Data PM asks before a workload lands on it.3 min read
  3. 3Moving and Shaping Data: Fivetran, Airbyte, dbt, Airflow, Kafka, Informatica, Talend, Ab InitioThe tools that get data from sources into the warehouse and turn it into modelled tables: managed connectors, SQL transformation, orchestration, streaming, and the enterprise ETL suites.3 min read
  4. 4Seeing the Data: Tableau, Power BI, Looker, Qlik, Metabase, RedashBI tools are the consumption layer. The important difference between them is where the metric definition lives: in each dashboard, or in a governed semantic layer everyone reads.3 min read
  5. 5Trusting and Governing: Monte Carlo, Great Expectations, Soda, Collibra, Alation, Atlan, Unity Catalog, DataOSThe tools that turn week 3's SLOs into monitors, make data findable, and encode who may see what. Also where data product platforms like DataOS sit.3 min read
  6. 6Data Science and ML Platforms: MLflow, SageMaker, Vertex AI, Dataiku, NotebooksWhere models are built, tracked, and served, and what a Data PM needs from them: reproducibility, a registry, and evaluation before anything reaches a decision.2 min read