06-reference

practicaldatamodeling april 2026 survey results

2026-05-11·reference·source: Practical Data Modeling·by Joe Reis

April 2026 PDC State of Data Modeling Survey Results (Joe Reis)

Why this is in the vault

This is the results drop for the April pulse survey we filed a teaser note on ([[2026-04-14-joe-reis-state-of-data-modeling-april-2026]]). 334 anonymous respondents, six single-select questions, one optional open-text field, fielded April 14-27 2026 through the Practical Data community. It is a self-selected sample, not a matched panel against the January State of Data Engineering Survey (n=1,101) or the March AI Tooling Pulse.

The post is short by Reis standards but data-dense. The headline finding is one of the cleanest "vendors are selling against reality" signals we have logged in the vault, and it directly underwrites RDCO consulting positioning. This is exactly the external falsification check we said we needed in the April 14 note ("If Stockholm findings say modeling decays because of org politics, not technique, MAC needs a complementary org-change narrative"). The results landed firmly on the org-politics side.

Top findings (verbatim percentages)

"What would most improve data modeling at your organization?" (single-select)

95.2% point at people, time, requirements, or ownership. 4.8% point at tools.

Who owns data modeling decisions?

Roughly half the field works in environments with no real data-modeling owner.

Why models break: 62.5% of the breakdown reasons reduce to time pressure (built under pressure, then requirements drifted, then nobody refactored). Echoes the January survey finding that 59.3% cite "pressure to move fast" as the top modeling pain.

Refactor cadence:

Standards adoption:

Cross-tab: teams with enforced standards are roughly 5x more likely to report their models hold up than teams with loose or no standards.

Open-text completion rate: 142 of 334 (42.5%) wrote an open-text response. Reis reads this as high engagement, which it is. The qualitative content split into two camps: AI accelerating modeling neglect vs AI finally forcing the modeling conversation. Both camps internally consistent.

Mapping against Ray Data Co

Mapping strength: STRONG. This is the cleanest market-positioning signal we have filed all quarter.

1. The 4.8% tooling number is a positioning gift. Every observability-and-platform vendor in the data space sells against this. Reis is now on record (and so are 334 practitioners) saying tooling is not the constraint. RDCO is a consulting + framework offer (MAC + BEAM workshops + standards work), not a platform sale. We are selling the 95.2% answer, not the 4.8% answer. This should be a Sanity Check pull-quote within a week.

2. The ownership-vacuum finding directly underwrites BEAM-style requirements workshops. 42.5% answered "whoever builds the pipeline" and 7.8% answered "nobody." That is half the field without a designated modeling owner. BEAM event-matrix workshops (from the Corr/Stagnitto ADWD synthesis) are literally tools for establishing ownership at modeling time. RDCO can package this as a fixed-scope engagement: "We run a 2-week BEAM workshop, you exit with a named owner, a documented conceptual model, and a standards doc." That is a productizable consulting offer with a clear price tag.

3. The standards cross-tab is the closest thing to MAC validation we have. Enforced standards = 5x more likely to say models hold up. MAC is a standards artifact dressed as a testing matrix. The quote Reis pulls ("clear modelling standards from the start... documentation side by side with the code... decent review process") is essentially a description of what MAC formalizes. We should be quoting this respondent in pitch decks.

4. Refactor data is a hidden upsell. 68.3% refactor occasionally / rarely / never. The "rarely, only when something breaks" 33.8% is the firefighting trap. MAC severity tiers (Stop/Pause/Go) are exactly the mechanism for converting break-driven refactor into scheduled refactor. There is a "MAC for legacy models" angle here we have not pushed: most teams cannot rebuild from scratch, so what does MAC look like applied to a model already in flight? Sanity Check angle.

5. The two-camp AI split is a content-calendar gift. Camp 1 ("AI is making modeling neglect worse") and Camp 2 ("AI is forcing the modeling conversation") are both right, as Reis notes. RDCO sits cleanly in Camp 2 as the answer. The semantic-layer-for-AI pitch ("we are pitching a data model to staff this week and our biggest selling point is getting a semantic model for AI tooling") is the wedge: AI agents need governed, modeled data to function. That is MAC's market entry. The CIO-hooked-Claude-to-Snowflake quote is the cautionary version of the same story.

6. Stockholm keynote watch. The April 14 note flagged Reis's Stockholm keynote (May 7) as the next external falsification point. This results post appears to BE the Stockholm material, packaged for Substack. The half-year survey he mentions ("kicking off soon") is the next data point to track.

Challenge to our thesis: None. April 14's hypothetical risk ("if findings emphasize org/political gaps, MAC needs a complementary org-change narrative") landed. We do need the org-change story. BEAM + ownership-workshop framing is that story, but it is currently buried in book notes. Promotion needed: turn BEAM ownership workshops into a productized RDCO offer in the consulting-pages section.

Action for the board:

Related