Why great AI products are all about the data | Shaun Clowes (CPO Confluent, ex-Salesforce, Atlassian)

Episode Summary

Executive Summary: Sean Kahl argues that product management is still immature because too many PMs operate inside the building instead of outside it, relying on activity over insight. He says AI’s biggest impact on PM will be data management and context, not models alone, and that durable SaaS value comes from business rules and workflows, not just UI or database forms. He also shares how growth, PLG, and career breadth create leverage.

Main Topics: Why product management remains underdeveloped (Priority: 5/5): Sean argues PM is still a relatively random discipline because many PMs focus on internal execution, politics, and delivery rather than customer, market, and competitor value creation. Great PMs synthesize external signals into differentiated bets. How PMs become truly effective (Priority: 5/5): He recommends spending most time thinking outside the building, using structured customer/market/competitor analysis, seeking counterfactuals, and pairing qualitative input with data-informed reasoning rather than activity for activity’s sake. AI’s real impact on product work: data and context (Priority: 5/5): Sean says LLMs are synthesis machines, but only as good as the quality, recency, and structure of data fed into them. For AI products, the winning layer is context and data management, not just models or prompts. Why SaaS apps are harder to clone than they look (Priority: 5/5): He rejects the idea that AI will easily wipe out incumbents like Salesforce, Jira, or Workday, arguing the real moat is years of embedded business rules, workflows, and configuration—not merely UI or data models. B2B growth and product-led growth (PLG) (Priority: 4/5): Sean explains the evolution of B2B growth teams from proving value, to scaling repeatable wins, to integrating with product, sales, and marketing. He argues PLG is valuable when balanced with sales rather than treated as a religion. Career development through breadth and ‘bingo card’ learning (Priority: 4/5): He describes choosing roles to fill missing boxes—consumer, enterprise, sales, growth, data, platform—so he becomes more versatile and dangerous in the best sense, able to spot patterns across functions and contexts. Failure, calibration, and decision-making (Priority: 4/5): Sean shares a past product failure where a sustainability product was kept alive too long despite weak fit. He emphasizes using data to challenge intuition, but not waiting for perfect certainty before deciding.

Key Arguments: Most PMs are not great because the discipline still rewards internal activity over external value creation; the job is to find reliable differentiated value in the market. A small number of PMs can create disproportionate leverage; a 10x PM can generate 100x-ish returns because PM multiplies the work of others. The best PMs operate from the outside-in: customer, market, and competitor perspective first; internal considerations second. Data is a compass, not a GPS: it should disprove weak ideas and sharpen intuition, not replace judgment entirely. LLMs are only as useful as the data and context they receive; AI success depends more on data management than model novelty. In SaaS, the real moat is embedded workflows and business rules accumulated over years, not the surface UI or basic data model. AI may strengthen incumbents more than it disrupts them, because dominant systems already own the workflows, rules, and distribution. PLG works best as one motion in a multi-motion company; the strongest businesses combine PLG and sales rather than choosing one exclusively. Career growth comes from deliberately seeking adjacent, different experiences that expand pattern recognition and functional judgment. Good decisions require enough data to avoid guesswork, but not so much that action is delayed past the point of value.

Data Points: Years into product management as a discipline: 15 to 20 years - Sean says PM is still surprisingly undeveloped despite being around this long. Qualitative interview saturation: 7 to 14 people - He cites the Nielsen rule of thumb for when interviews stop yielding new insights. Internal focus recommendation: 80% outside the building - He references the idea that PMs should spend most of their time thinking about external market/customer dynamics. PM exclusion rate: 90% of requests said no to - He says PMs are often tasked with rejecting most incoming ideas, which makes the role unpopular. Customer count at Atlassian: 300,000 customers - Lenny corrects Sean while discussing how large customer bases make incumbents hard to attack. Global holdout group: 10% of all people - Sean describes having a permanent experiment holdout group for long-term impact comparisons. Long-term experiment reversals: 40% of the time neutral - Lenny mentions Shopify’s practice of checking cohorts years later and finding many positive short-term experiments wash out. Product usage time frame: 2 years - The failed sustainability product remained in market for two years before being killed. Listening / sign-up scale: 23,000 product teams - Referenced in the BuildBetter ad as current usage scale. Subscription retention: 93% - Referenced in the BuildBetter ad as a retention metric.

Pivotal Quotes: "“Always talk from the customer’s perspective, from the market’s perspective, from the competitor’s perspective.”" — Sean Klaus: Sean’s central advice on how PMs should frame documents and decisions. "“LLMs can only be as good as the data they are given and how recent that data is.”" — Sean Klaus: His core thesis on AI’s impact on product management and product workflows. "“It’s the years and years and years of evolution of the underlying workflows of the product to support the customers.”" — Sean Klaus: Explaining why SaaS incumbents are harder to clone than they appear.

Implications: PMs should build external awareness, use data rigorously, and avoid becoming delivery coordinators. For AI, focus on context/data plumbing, not just models. For startups, durable advantage comes from workflow depth, distribution, and multi-motion growth.