06-reference

seattle data guy solving the right problems

2026-10-15·reference·source: SeattleDataGuy·by Ben Rogojan
data-engineeringproject-scopingsource-of-truthdata-leadershiprequirements-engineering

"Are Data Teams Solving the Right Problems?" — SeattleDataGuy (Ben Rogojan)

Why this is in the vault

A six-question pre-project checklist for catching the specific failure mode where a team solves an easy, adjacent problem (a dashboard, a platform migration) instead of the hard business problem it was actually asked to solve.

The core argument

Six questions Rogojan says he asks himself and other data leaders before taking on a project, framed as timeless despite how much the tooling (LLMs, agents) has changed in the last two years:

  1. Are we substituting an easy problem for a hard one? Attribute substitution — swapping "reduce churn" (hard: requires cross-team behavior change) for "build a churn dashboard" (easy: ships, looks like progress, doesn't move the metric). Same pattern at the platform level: a 12-month migration from Platform A to B gets celebrated as a win, then months later the team realizes the platform was never the real constraint — politics, policy, and process were.
  2. What question or decision is this data supposed to support? Applies to dashboards, pipelines, and now agents. In a world where producing POCs is nearly free, failing to tie the build to a specific decision means it ends up "on the shelf" regardless of how fast it shipped.
  3. What is the real source of truth for this problem? His worked example: Salesforce says a contract is $120k, the billing system was quietly updated to $100k, and now Sales and Finance each have a defensible-sounding but wrong answer for "what's the source of truth" depending on which one they're asked.
  4. What knowledge exists only in someone's head or spreadsheet? Teams build isolated, un-managed workflows; the "real" number often lives with a single IC ("Bob's head"), not in any system.
  5. What business process are we modeling? Jumping straight to "pull in all the Salesforce data" produces siloed datasets that can't actually be joined across the CRM/ERP/warehouse systems that each hold a different slice of the same business process (the opportunity, the order, the fulfillment).
  6. What happens if we don't take on this project? Worth asking leadership directly — answers range from "multi-million-dollar loss" to "would be nice to have," and the latter should lose priority against a stacked backlog even without an explicit cut decision.

⚠️ Sponsorship

Self-consulting CTA for Codestrap, Rogojan's own data/software engineering consultancy, in the top-of-issue preamble before the article starts — "That is the gap our team at Codestrap helps companies close... Feel free to reach out." This is the same self-promotional pattern documented in prior SDG backfills (e.g. [[2026-07-18-seattle-data-guy-data-warehouse-integration]], [[2026-08-26-seattle-data-guy-ai-software-factory-maturity]]), not a disclosed Estuary adviser relationship this issue — no Estuary mention appears in this one. Worth noting: the CTA pitches "connect technical solutions to real business outcomes," which is the exact thesis of the article that follows — the six questions double as a soft sales argument for why a company needs outside help (Codestrap) to ask them rigorously.

Curation section

"Articles Worth Reading" carries two items, both self-cross-promo (no third-party item this issue — a deviation from SDG's usual ~50/50 split):

Also namechecked but not linkable: a "Video Of The Week" ("What It Actually Takes to Build a Data Pipeline System From Scratch - And Why You Probably Shouldn't") — title only, no URL in the plaintext body, consistent with the pattern seen in other SDG issues.

Mapping against Ray Data Co

The sharpest connection is question 3 (real source of truth), and it isn't hypothetical for RDCO — [[feedback_verify_blockers_against_source_not_notes]] documents the exact same failure mode happening twice: a stale note got treated as a live 🔴 blocker instead of being re-checked against the actual source (Jira, ticket, auth file), because two "systems" (the note and the ticket) disagreed and the wrong one got trusted by default. Rogojan's Salesforce-vs-billing-system example is the same shape — two records of the same fact diverge, and whichever one you ask first becomes the "truth" unless you deliberately check.

Question 6 ("what happens if we don't take on this project?") also lands directly on [[project_deep_research_backlog_inflation]] — a queue that self-inflates ~2.9x because items get added without anyone asking what's lost by not doing them, and on /check-board's practice of auto-archiving stale Ray-owned tasks rather than letting "nice to have" silently hold a priority slot forever.

Question 1 (easy-problem substitution) is a useful check against the dashboard-as-deliverable trap specifically because RDCO's own build discipline leans the other way by design — [[feedback_targeting_system_prioritization_filter]]'s four-layer filter ("can Ray do X?") exists precisely to catch "yes, technically" answers that don't actually address the underlying constraint before committing build time.

Related