Lenny's PodcastShaun Clowes: Why AI products win on data, not on the model
What if great AI product work is really data management work; Confluent's leader grounds PMs outside the building, where customer insight differentiates.
CHAPTERS
- 0:00 – 5:32
Why product management still feels “undeveloped” (and why most PMs aren’t great)
Shaun lays out his core concern: product management outcomes and performance still feel random for a discipline that’s decades old. He frames PM as a leverage role where a truly great PM can multiply team impact dramatically—yet the profession doesn’t reliably produce “10x PMs.”
- •PM as an underdeveloped discipline despite 15–20 years of maturity
- •Randomness in behaviors and outcomes vs. more established professions
- •Why PM leverage can create outsized (100x+) returns
- •Why the profession gets criticized more than engineering/design
- 5:32 – 11:30
The fastest way to level up: live outside the building
Shaun’s prescription is simple but demanding: anchor everything in the customer/market/competitor perspective, not internal process. He argues PMs lose effectiveness when they get pulled into politics, Scrum management, and delivery mechanics rather than differentiated value creation.
- •Spend ~80% of thinking on what’s outside the company
- •Start every document from customer/market/competitor POV
- •Avoid getting dragged into internal politics and delivery-only work
- •PM’s job: find reliable, differentiated value in the market
- 11:30 – 13:24
Avoiding “activity without outcomes”: structure your discovery and challenge yourself
Many PMs believe they’re customer-centric, but Shaun says they often recycle the same sources and fail to synthesize. The real advantage comes from deliberately seeking disconfirming evidence, analyzing competitor signals, and comparing what users say vs. what they do.
- •Don’t only talk to the usual customers—seek unfamiliar perspectives
- •Actively fight availability/confirmation bias
- •Look for counterfactuals and evidence you’re wrong
- •Synthesis matters more than collecting more raw inputs
- 13:24 – 17:25
Practical AI for PMs: using LLMs to stress-test strategy and competitor fit
Shaun describes how generic LLMs can accelerate qualitative synthesis if you ask the right questions. The key is prompting for mismatches, edge cases, and contradictions—using AI to uncover what you don’t want to hear.
- •Right-size qualitative research (e.g., Nielsen-style interview ranges)
- •Avoid leading questions that contaminate learnings
- •Prompt LLMs to find where strategy doesn’t fit customer interviews
- •Use LLMs to infer competitor strategy from public positioning
- 17:25 – 19:21
“Feedback rivers”: building an always-on stream of customer demand signals
Building on the ‘feedback river’ concept, Shaun explains how teams can continuously ingest and organize feedback at scale. He shares how Confluent uses LLMs to semantically cluster inbound requests and track what’s rising or fading over time.
- •Great PMs intentionally swim in a constant feedback river
- •Aggregate inputs: interviews, NPS, field requests, competitor info
- •Use LLMs to summarize, deduplicate, and cluster requests semantically
- •Trend detection across hundreds/thousands of feedback items
- 19:21 – 24:29
The real AI moat is data management (not the model): context, quality, and recency
Shaun argues LLMs are only as good as the data you can feed them, and most information decays quickly. For both internal workflows and AI-powered products, the hardest work is sourcing, structuring, and delivering timely context—not swapping models or polishing prompts.
- •Models are powerful but ‘dumb’ without fresh, relevant context
- •Information decays fast; stale data undermines decisions
- •AI product success hinges on data access, quality, structure, and latency
- •Most engineering effort goes into moving and managing data for the LLM
- 24:29 – 31:18
Will AI commoditize SaaS? Why “forms on databases” is misleading
Shaun challenges the idea that AI will make it easy to clone major SaaS apps and topple incumbents. The durable lock-in isn’t just UI or schema—it’s years of accumulated business rules and configurable workflows that become unique to each company.
- •SaaS looks like ‘forms on databases’ but is hard to replicate
- •Real differentiation: business rules and workflow evolution over time
- •Customer-specific configuration becomes a proprietary black box
- •Even in an agentic future, agents still need those system rules
- 31:18 – 35:41
Distribution in an AI world: noisier channels, “meet the bar and be different,” and data-first UX
They explore how distribution remains the hardest problem, and AI may worsen signal-to-noise via spammy outreach. Shaun points to startups winning by embedding analytics and outcomes directly into workflows (example: Ashby) and using data as a first-class product feature.
- •Distribution advantage is always critical; consideration sets are hard to enter
- •AI may reduce effectiveness of cold outreach by increasing noise
- •Disruptive wedge: meet baseline utility and differentiate meaningfully
- •Example: Ashby’s workflow-integrated analytics as a UX differentiator
- 35:41 – 39:35
Data-driven decision-making, reformed: data is a compass, not a GPS
Shaun reframes data as a tool for direction-setting and falsification, not automatic answers. He warns that “data obsession” can slow teams and be misused as a weapon, and argues for right-sizing data to the decision at hand.
- •Right data, right time, right place matters more than ‘more data’
- •Data helps disprove bad ideas and calibrate intuition
- •Over-indexing on data can make you wrong or slow (or both)
- •Avoid using data as authority without solid analysis
- 39:35 – 45:50
How to sanity-check analysis: upstream/downstream checks and avoiding fool’s gold
Shaun shares a practical method for validating surprising results: trust intuition first, then investigate rigorously. He recommends examining what happened before/after a metric shift and zooming out to ensure the effect matters for the business (scale, retention, revenue).
- •If data contradicts intuition, assume it may be wrong—then validate
- •Check upstream context and downstream consequences
- •Verify representativeness (sample bias, selection effects, small segments)
- •Tie local metric wins back to business outcomes (e.g., ASP/MRR)
- 45:50 – 50:18
Building B2B growth teams: from ‘gold rush’ experimentation to scalable systems
Shaun walks through growth team phases: proving value, finding repeatable wins, scaling, and integrating with the broader org. He explains why growth teams often fail—either by random walks of experiments, inability to operationalize, or friction with adjacent functions.
- •Phase 1: prove it (exploration and quick wins)
- •Phase 2: systematize and demonstrate repeatability
- •Phase 3: scale across the funnel (activation to upsell)
- •Common failure: poor org fit and ‘random’ experimentation without a model
- 50:18 – 56:02
The evolution of PLG: incentives, balance with sales, and why ‘zero PLG’ is dangerous
Responding to skepticism that PLG ‘doesn’t work,’ Shaun argues the real risk is neglecting end-user experience because B2B incentives naturally prioritize buyers and RFPs. The best companies balance PLG with enterprise sales, making the motions feed each other to create resilience.
- •Without PLG, end-user success becomes a hobby, not a system
- •Incentives drive outcomes; someone must own user success metrics
- •PLG and sales aren’t mutually exclusive—hybrid motions can outperform
- •Resilient businesses combine many customers with significant revenue
- 56:02 – 1:07:46
Career strategy as a ‘bingo card’: building versatile superpowers across domains
Shaun explains how he chose roles to fill experience gaps—PLG, consumer, enterprise distribution, infrastructure—making him more adaptable and pattern-aware. He shares a pivotal story about Atlassian’s general counsel demonstrating deep product strategy insight, inspiring Shaun’s goal of being ‘dangerous’ across functions.
- •Choose roles that stretch you while staying adjacent to your strengths
- •Variety builds pattern-matching and versatility across problems
- •Salesforce as a ‘PhD in marketing’ and distribution excellence
- •Aim to contribute broadly without meddling—be ‘dangerous’ in many areas
- 1:07:46 – 1:14:08
Failure corner + final operating advice: kill zombies early, don’t let your calendar run you
Shaun recounts launching a “zombie” sustainability product that should have been killed after six months but lingered for two years—highlighting the cost of not forcing hard decisions. He closes with guidance on protecting time for outside-the-building thinking and deciding with the right amount of data (not too little, not too late).
- •Lesson: create earlier forcing functions to end failing bets
- •Cost of resource drain from zombie products and unclear right-to-win
- •Protect thinking time; calendar-driven PMs lose strategic leverage
- •Decision timing: avoid <30% info and >70% info paralysis
- 1:14:08 – 1:21:34
Lightning round: books, tools, motto, and Sydney travel tips
In rapid-fire format, Shaun shares influential books, favorite lightweight TV, and a standout product (Glean) for org-wide knowledge search. He ends with a motto about trust and caring, plus a Sydney recommendation: canyoning and rappelling in the Blue Mountains.
- •Book recs: The Lean Startup; Inspired (Marty Cagan)
- •Product pick: Glean for enterprise knowledge search + expertise discovery
- •Motto: people don’t care what you know until they know you care
- •Sydney tip: Blue Mountains canyoning and waterfall abseiling/rappelling
