From Dashboards to Agents: How Salesforce is Redefining Data Interaction w/ Irina Malkova
🎯 Summary
Podcast Summary: From Dashboards to Agents: How Salesforce is Redefining Data Interaction w/ Irina Malkova
This episode features Irina Malkova, who leads a product data team at Salesforce, discussing the evolution of data interaction within the enterprise, moving from traditional dashboards to sophisticated, conversational AI agents, and the critical role of product telemetry and knowledge graphs in this shift.
1. Focus Area: The primary focus is the transformation of enterprise data consumption and decision-making driven by Generative AI. Key themes include the shift from static Business Intelligence (BI) dashboards to conversational AI agents embedded directly into workflows (like Slack), the strategic importance of product telemetry (data emanating from product usage), and the necessary technical underpinning provided by Knowledge Graphs to overcome limitations in standard Retrieval-Augmented Generation (RAG) systems within dense corporate data environments.
2. Key Technical Insights:
- Product Telemetry Expansion: Telemetry has evolved from purely engineering observability (error logging) to a crucial data source for ML/AI, especially recommendation systems and internal productivity tools, as every company digitizes further.
- The Dashboard-to-Agent Transition: Agents offer superior data interaction by moving insights directly into the user’s flow of work (e.g., Slack), reducing the cognitive load associated with interpreting complex, multi-dimensional dashboards that often require specialized knowledge to decipher.
- Graph RAG for Enterprise Data: Standard RAG struggles with dense, entity-rich enterprise corpora (where core terms repeat frequently, obscuring nuance). Graph RAG emerges as a necessary improvement, leveraging the inherent graph structure of enterprise data relationships to provide more accurate, nuanced answers, particularly when dealing with complex data catalogs or documentation.
3. Business/Investment Angle:
- AI Adoption is Workflow Integration: True enterprise AI value is realized when agents automate decisions and provide actionable advice directly within the user’s existing workflow, rather than requiring users to switch to a separate BI tool.
- Data Activation Layer Necessity: Tools like Salesforce Data Cloud are becoming essential as a “data activation layer” that unifies disparate data stores, enabling agents to orchestrate actions across the entire data landscape.
- High Stakes in Enterprise AI: Enterprise adoption is slow and incremental because the stakes are high—decisions made via AI can impact careers. Trust and demonstrable, measurable productivity gains are paramount.
4. Notable Companies/People:
- Irina Malkova (Salesforce): Leads a product data team focused on building data infrastructure for product telemetry and enabling internal AI/ML experiences.
- Salesforce Agent Force & Data Cloud: Highlighted as the specific platform combination (orchestration system + data activation layer) enabling their internal agent development.
- Architect Community (Salesforce): Credited with helping reintroduce and champion the use of graph technologies within the organization.
5. Future Implications: The industry is moving toward an agentic enterprise where data interaction is conversational and automated. While full organizational restructuring might be slow, significant, often unseen, productivity gains are already occurring due to faster learning curves (Malkova learned more this year than in the previous decade) and better tooling focused on semantics and retrieval. The future requires solving fundamental challenges in semantics and retrieval to make agents truly reliable across complex enterprise knowledge bases.
6. Target Audience: This episode is highly valuable for Data Leaders, AI/ML Engineers, Product Managers overseeing internal tooling or customer-facing data products, and Enterprise Architects focused on modernizing data stacks and implementing GenAI solutions.
🏢 Companies Mentioned
💬 Key Insights
"Most enterprises do not know what those things are because it was just this nebulous bunch of humans making decisions and things just somehow magically working not every enterprise has a very detailed map of metrics that matter the world they want team that typically has to put some some concept on it is a data team because the data teams job is take the complexity of what is a living and breathing enterprise and make a dashboard out of it."
"I think that they the teams hold the anthology of enterprise so what I mean is that every enterprise now needs to figure out what their processes are what is important to them what are the concepts that met that what is their growth model what things adapt to having more revenue less revenue where the levers on it etc."
"moving consumption for insights from dashboards to flow for it is going to suddenly take us to a place where we can hard measure the impact just like product folks measure the adoption of theirs and they can maybe test via data people can do the same."
"there is not like a just for marketing one just for finance one just for tech you have this customer lens problems that span departments and I think that is not the only thing that it can do is to do that. And that is that immediately shines the light on how we do not have that common layer of common technology common layer of data that we can feed into that agent."
"I think we must solve that before we talk agents and I am really I am expecting a lot of movement in the vendors based on the pieces like that maybe so maybe not all the way to the agents running the world yet but a lot of movement in these underlying important pieces of time. I do not think agents are going to run the world until you fix these foundational things that did not going to happen."
"it is really it is the dimension there that separates things seems to be the density of entities within that corpus and that is really where the graphs come in and I think it is graph RAG being an improvement over standard RAG on the entity dense corpus was the first thing that enterprises went to for unstructured data"