Podcast Summary: The Directions on Microsoft Briefing

Episode: Three Smart Guys: It's All About the Data

Date: March 4, 2026
Host: Barry Briggs
Guests: Peter O’Kelly, George Gilbert
Theme: How data estate modernization is now imperative for enterprises in the age of AI—what it means, the challenges, and Microsoft's prospects.

Episode Overview

This episode brings together long-time Directions on Microsoft analysts Barry Briggs (host), Peter O'Kelly, and George Gilbert to discuss why “it’s all about the data” in the enterprise—especially as AI, Microsoft technologies, and competitive pressures compel organizations to overhaul, manage, and derive more value from their data estates. The conversation moves from the classic problems of data hygiene and modeling to hot topics like semantic layers, context graphs, data maturity, and the high-stakes 'battle of the data titans' among Microsoft, AWS, Google, Snowflake, and Databricks.

Key Discussion Points and Insights

1. The Urgency of Data Cleanup for Enterprise AI

• AI has turned data cleanup from an ignored chore to an urgent strategic priority. Enterprises now realize “garbage in, garbage out” with AI is more consequential than ever.
• Historical context: Barry recalls Microsoft's own challenge with 2,500 apps/databases—highlighting how data sprawl and technical churn have accumulated over decades. (00:40)
• Peter O’Kelly’s take: Data modeling is not rocket science, but it's been neglected and is now returning—driven by AI’s need for quality, consistent data. (05:35)

  “It’s lining up to be the most consequential case of the adage of garbage in, garbage out.” — Peter O’Kelly [01:52]

• AI creates an “incentive to clean things up” because faulty data will lead not only to wrong insights but also new risks around inadvertent oversharing and weakened security. (02:30)

2. Data Modeling: A Lost Art (Re)Found

• Components of data modeling: Conceptual (shared understanding of real-world entities), logical (transforming concepts into actual database structures), and physical (performance optimization).
• Challenges: Organizational silos, tech churn, “anti-agile” perceptions, and database vendors downplaying modeling (“schema on read”).
• Renewed relevance: LLMs and tools like Microsoft Copilot and Google Gemini can collaborate to analyze data usage and support modeling efforts.

  “Data modeling has never been super obscure... now, LLMs such as Claude and Gemini are very productive data modeling collaborators.” — Peter O’Kelly [05:35 & 07:22]

• Resource mention: Joe Reese's "Practical Data Modeling" Substack/book is cited as a valuable new resource. (07:40)

3. The Modern Data Estate: Maturity Models and Their Limits

• Classic approach: Data warehousing and data lakes often led to “a big mess,” simply aggregating but not harmonizing data across silos. (09:57)
• Gilbert's Data Maturity Model:
  • Low maturity: Siloed, aggregated “cubes”; analytics are basic and disconnected.
  • High maturity: Real-time tracking, deterministic process modeling, and finally, enterprise-wide knowledge graphs linking causality, probability, and business outcome prediction.

  “Data warehouses are where data goes to die.” — (Citing a senior analyst, paraphrased by Barry Briggs) [10:00] “We need to model the processes so we understand why things happen... much, much harder than entity-centric stuff.” — George Gilbert [08:31]

• Future focus: Moving from just modeling entities to modeling entire business processes and decision logic—crucial for enterprise AI. (08:55)

4. Buzzwords Explained: Semantic Layers, Ontologies, Context Graphs, Knowledge Graphs

• Peter O’Kelly: These terms have become “semantic spackle”—often deployed to patch over deficiencies in core data modeling. (14:48)

  “If you have your data models in order, you just don’t need a lot of the semantic spackle.” — Peter O’Kelly [14:55]

• Three facets: (15:21)
  1. Creating logical data models across silos (master data management 2.0).
  2. Building comprehensive repositories of facts, dimensions, metrics.
  3. Incorporating new resource types (documents, emails, transcripts), once “unstructured.”
• Context Graph: Seen as the future—captures not just data but reasoning, processes, and decision-making traces.

5. Systems of Intelligence and the Big Leap Forward

• Referencing Geoffrey Moore’s “Crossing the Chasm”: The panel debates the move from “systems of record” and “engagement” to “systems of intelligence” (and eventual “autonomy”).
• Context graph as enabler: It could capture the hard-to-document tacit knowledge and reasoning paths while the semantic (deterministic) layer encodes established processes.

  “The context graph... captures the reasoning traces, the tacit knowledge of why people make decisions... the deterministic semantic layer and the context graph reinforce each other.” — George Gilbert [19:12]

• Digital twins: The emerging vision is a digital twin of the enterprise with a “deterministic twin” (semantic/process layer) and a “cognitive twin” (context graph/decision model). (20:48)

6. The Battle of the Data Titans—Who Will Win?

• Giants in the ring: Microsoft, Amazon, Google (the hyperscalers), with Snowflake and Databricks ascending fast.

  “In the western world, [it’s] down to five... hyperscalers, then Databricks and Snowflake.” — Peter O’Kelly [23:12]

• Competitive dynamics: Snowflake and Databricks converging on the same multi-format, multi-purpose platform; both now support OLTP with PostgreSQL; both pursuing AI/ML.
• Microsoft’s position: Still building out a convincing platform, leveraging Microsoft 365 Graph, Fabric IQ, Work IQ, and the Purview catalog for data governance.

  “At this point I think [Microsoft is] lagging the market leaders in AI and data in several respects… Is Microsoft’s ‘good enough’ moat sufficient?” — Peter O’Kelly [24:20]

• The role of catalogs: The next “choke point” and battleground isn’t just storing data, but governing it via rich, enterprise-wide catalogs and metadata (“war for the catalog”). Microsoft’s Purview vs. Databricks Unity Catalog is a live contest. (29:53)

  “There is a war for the catalog and Microsoft’s armor in this is Purview.” — Barry Briggs [29:53]

7. The Outlook for Disruptors and the Vendor Landscape

• Too early to call a winner: Maturity of semantic layers and context graphs is still in “early innings”; room exists for left-field entrants, e.g., Palantir or RelationalAI, to disrupt—especially if they master modeling rich digital twins or reasoning traces. (26:39)
• Platform convergence: At the base/data platform layer, strong convergence on extended relational approaches and PostgreSQL (“Postgres has won” at the OLTP layer).
• Strategic advice: Enterprises must pick a “center of data gravity” and align strategic platform/governance for that—clumsily bridging via spackle is less attractive when chaos (and AI risk) threatens. (30:25)

  “If you don’t do this [clean up and consolidate], and you start introducing AI tools, it’s going to get chaotic fast.” — Peter O’Kelly [30:25]

Notable Quotes & Memorable Moments

• On the impact of AI on data hygiene:

  “Getting senior leaders to address the problem of data hygiene... was almost a Sisyphean challenge... but we suddenly have somebody pulling the rock, which is AI.” — Barry Briggs [04:56]

• On data modeling’s ‘anti-agile’ perception:

  “It’s seen often as kind of anti-agile. So it’s more of a perpetual program than it is a readily pointable project.” — Peter O’Kelly [06:26]

• On data warehouse criticisms:

  “Data warehouses are where data goes to die.” — (Cited by Barry Briggs) [10:00]

• On system convergence:

  “At the data platform layer... the patterns are pretty clear... For OLTP, I think that battle is over and Postgres has won.” — Peter O’Kelly [28:16]

Key Timestamps

• 00:00 – Introductions; why data now matters more than ever for AI
• 01:47 – The rise of “garbage in, garbage out” in AI, data chaos, data technology churn
• 05:35 – Data modeling explained and regained relevance
• 08:21 – The misery and necessity of data cleanup; the stairway to AI value is one business outcome at a time
• 10:54 – George Gilbert’s data maturity model: From silos to knowledge graphs
• 13:15 – Most enterprises are still low on the maturity curve
• 14:48 – Semantic layers, ontologies, context graphs: Real value or distraction?
• 17:54 – Can context graphs finally deliver “systems of intelligence” and ‘digital twins’?
• 21:44 – The vendor playground: Microsoft, Snowflake, Databricks, Palantir, the “war for the catalog”
• 28:16 – Is there still room for startups/disruptors? Where does platform convergence leave us?
• 29:53 – The criticality of catalogs and metadata governance
• 30:25 – Final advice: Consolidate your data gravity—AI will punish a messy estate

Conclusion

The panel frames AI as an irresistible force finally driving enterprise leaders to take data hygiene, modeling, and governance seriously—after decades of neglect. They anticipate major changes to how businesses define, manage, and draw insights from their data, with technical leadership (and vendor supremacy) hinging on mastery of not just data storage, but model richness, process modeling, and context-capturing reasoning traces. Microsoft is a strong contender but faces agile, innovative competition as the “war for the catalog” unfolds. The call to action: invest in data estate modernization before AI multiplies your mess—and your risk.