Susan Chase

As data moves into Microsoft #PowerBI, access controls from platforms like Snowflake and Databricks often break down—exposing sensitive data and creating compliance gaps. TrustLogix extends enterprise access policies into Power BI, so analysts can work with governed data across both direct queries and imported datasets. The result: consistent controls, fewer risks, and faster, trusted insights. 🔗 Learn how it works: https://hubs.li/Q03r84n40 #DSPM #DataAccessGovernance #Snowflake #Databricks

10

1 Comment

A lot of the current conversation around agents focuses on risk. We should probably be talking more about architecture. Agents Are Not the Problem. Architecture Is. There is also a lot of concern about data governance. People worry about what an LLM might see, what it might leak, or what actions it might take. Those concerns are valid. But a lot of the conversation jumps straight to the risks without looking at the architecture. We already have the building blocks to do this safely. Start with identity. Agents should not be treated as some special class of actor. They should be identities just like users or services. That means they fall under the same IAM controls for authentication, scope, delegation, and audit. Next is governance around the model. The LLM should not be the trust boundary. It is just the reasoning engine in the middle. What matters are the controls around what goes into the model and what comes out of it. Ingress controls decide what context the model can see. That might include filtering data, masking sensitive fields, or limiting what can be retrieved. Egress controls decide what the model is allowed to do. That might include tool permissions, approval workflows, and output filtering. Finally, systems of record should stay outside the AI layer. Access to systems like Salesforce, ServiceNow, Jira, or GitHub should go through a governed integration layer such as MCP. That creates a controlled path for agents or models to interact with real systems. The flow ends up looking something like this. Users and agents → IAM identity and access controls → governance around the LLM → MCP for tool access → systems of record Agents are not the real risk. Architecture determines whether these systems are safe or not. The model is not the security boundary. Governance around it is.

3

2 Comments

Data catalogs and MDM platforms are on a collision course. Why? Success in data governance - and management - requires the ability to: 1) Discover and classify your data 2) Define data governance policies for your data 3) Implement and enforce those governance policies across all enterprise use cases, both operational and analytical. The biggest limitation of data catalogs (aka 'metadata mgt') tools has always been that they are essentially silos, largely disconnected from every other mission critical software platform in the organization. You could discover your data and define governance policies for it, but you wouldn't have the ability to enforce those policies in either the analytical or operational systems consuming it. Or put another way, you can manage your metadata in a catalog, but your ability to activate that metadata it is dependent on some serious effort to integrate your catalog to downstream systems. This is where MDM software comes into the picture. As a discipline, MDM has always been about enforcing governance policies in both operational *and* analytical systems for the data that matters the most to your organization. Rules are defined to allow you to identify errors or inconsistencies in your mission critical metadata (customers, products, assets, etc.), and MDM platforms take the steps needed to rectify them in accordance with your policies. And when the system can't (or shouldn't) automatically enforce those rules, data stewardship workflows are triggered to put a 'human in the loop'. This is all enabled through integrations MDM hubs create with downstream consumers of data across the enterprise. This means that MDM systems play a critical role to allow for metadata being managed in a data catalog to be both governed, 𝐚𝐧𝐝 𝐚𝐜𝐭𝐢𝐯𝐚𝐭𝐞𝐝, at scale. This is my way of saying I think it's somewhat inevitable that the barrier between MDM platforms and Data Catalogs will slowly dissolve over time. This will happen through partnerships (like the one Profisee has with Microsoft for their integration with Purview) or it will happen through catalog providers integrating MDM capabilities directly into their platforms. The convergence of data catalogs and MDM isn’t just about tighter integration - it’s about enabling a new capability in the modern data stack. Some (like yours truly) would call this a next-generation semantic layer. I've heard others refer to this as a context layer. Whatever the label, the idea is the same: we need systems that don’t just document data, but ensure that its meaning and quality are consistently applied wherever it flows. It's a powerful combination. What do you think? Do you agree, or disagree? If so, why? #datacatalogs #masterdata #masterdatamanagement

129

64 Comments

Just published: "The Silent Drop: My Journey with Snowflake OpenFlow CDC Pipelines" 🚀 If you've ever stared at a green NiFi processor that refuses to output a single FlowFile, this one is for you. I walk through the specific "gotchas" of setting up CDC in a Bring-Your-Own-Cloud (BYOC) environment—from handling "Zombie" Postgres connections to fixing silent schema issues. Read the full story here: 

6

In her latest tutorial, Senior Manager of Information Design, Alyssa Huff, shares 10 heuristics, or rules of thumb, for improving dashboard design. Read now: https://pfdata.co/4ayeUZe Learn more from Alyssa: https://pfdata.co/4ppGKvp #informationdesign #visualanalytics #datavisualization

5

One thing I keep seeing in enterprise AI rollouts: the highest-risk data usually isn’t in tables it’s in the messy stuff. Tickets, call transcripts, notes, emails… the place where PII shows up casually. That’s why I liked seeing Snowflake move in two practical directions recently: A clearer path to structuring unstructured data with Cortex AI Functions, so teams can extract consistent signals without building a one-off pipeline every time. AI_REDACT reaching GA automated PII detection + redaction for unstructured text, built into the platform. From an AI & Security Platform lens, this is the kind of “boring guardrail” that actually enables speed: you reduce privacy risk upfront, and you make cross-team alignment (engineering / data / security) much less painful. If you’re working with unstructured data today, where do you enforce redaction at ingestion, at query-time, or right before model access?

1

This week brought a lot of back-and-forth on dbt modeling layers and best practices. It’s easy to frame it as a technical debate, "staging vs. intermediate vs. marts", "persistence vs. ephemerals". But what stood out to me is this: patterns make the difference. Patterns aren’t about rigid rules. They’re about giving teams a common language to reason about trade-offs, whether that’s cost, performance, or clarity. Some of the questions that surfaced: 🔹 Where should business logic live so it’s reusable but still readable? 🔹 When does persisting an intermediate table create value vs. unnecessary cost? 🔹 How do we balance speed with guardrails for governance? There isn’t one universal answer. But the conversations flow better and the alignment sticks, when patterns lead the way. I’ve always been an advocate for this, and weeks like this reinforce why. It’s not about perfect layers. It’s about patterns that teams can trust, evolve, and scale with. #DataModeling #DataEngineering #PlatformThinking #ArchitecturePatterns #ArchitectMindset

25