AMA: Barracuda Sr. Director, Product Management, Sheila Hara on AI Product Management

Effective teams communicate releases by focusing on problems solved—not features shipped.As Marty Cagan emphasizes, customers don’t care that something shipped; they care that a frustration went away. What works: Lead with the problem, not the release NOT-“We released a new phishing classifier.” BUT- “You’ll now see fewer false positives in non‑English phishing—so your team spends less time reviewing noise.” Anchor updates to known customer pain “Many customers told us setup took too long. We si ...Read More

Accuracy is table stakes. The AI features that create durable value are the ones that perform well across trust, usability, and outcomes in real‑world workflows. The AI features that win aren’t the ones with the highest benchmark scores — they’re the ones customers trust, adopt, and rely on daily. Accuracy matters, but outcomes, predictability, and usability are what turn AI into real product value. When we evaluate AI features, we consistently look at a broader set of criteria: 1. Time‑to‑Value ...Read More

When ROI is ambiguous—which is often the case with AI—we don’t force false precision. Instead, we use a set of decision frameworks that balance near‑term impact, strategic optionality, and downside risk. The goal isn’t to prove AI is better—it’s to decide where AI meaningfully changes the shape of the bet. Traditional bets are evaluated on execution certainty and market size. AI bets are evaluated on learning velocity, leverage, and strategic optionality. We don’t hold AI to a lower bar—but we d ...Read More

The common pattern across leaders Across incumbents and startups, the strongest agentic AI implementations share a few traits: Outcome‑first design – Agents are measured by work completed, not answers generated. Bounded autonomy – Clear limits, checkpoints, and escalation paths build trust. Deep workflow embedding – Agents succeed when they live inside real systems, not on top of them. Context and orchestration – Multi‑agent coordination and shared context matter more than raw model power. Expla ...Read More

At a high level, I believe: Adoption = Tooling × Behavior Change × Workforce Enablement 1. AI embedded inside existing workflows (not a destination tool) The single strongest adoption driver is in‑context AI—where AI shows up exactly where work already happens (IDE, PR review, Outlook, Jira, Teams), not as a separate app or “go try this” experience. Forbes and Prosci both highlight that asking users to switch tools or craft prompts creates friction and abandonment. [forbes.com], [prosci.com] Mic ...Read More

AI expands the solution space. But, PMs are still responsible for choosing the right solution, for the right problem, for the right customer, at the right time. Even in the age of AI, the PM remains essential in solutioning because AI can generate many possible solutions, but PMs decide which one actually solves the right customer problem in the right way. Here is an example:An AI system might propose multiple ways to reduce phishing false positives—auto‑blocking more aggressively, adding comple ...Read More

When everyone is talking about AI, differentiation doesn’t come from having AI — it comes from where AI sits in the workflow and what outcome it changes. Our lens is to work backwards from customer pain and measurable outcomes, not from AI capability. AI is the means, not the value. I know Im gonna ruffle some feathers by saying any of this. 1. We ignore AI claims and look for broken workflows Most competitors lead with statements like “AI‑powered,” “agentic,” or “copilot‑driven.” We look past t ...Read More

The most successful AI‑driven personalization efforts treat AI as a learning system, not a campaign engine. When teams focus on relevance, intent, and continuous feedback—backed by clean, behavior‑centric data—AI becomes a force multiplier rather than a novelty. How have you used AI personalize messaging and targeting for different customer segments? We’ve used AI to move personalization from static segments to behavior‑ and intent‑driven experiences, where messaging adapts based on what custome ...Read More

AI is fundamentally changing market research by shifting it from a slow, episodic activity into a continuous learning system. AI is most powerful in market research when it helps us ask better questions, listen at scale, and learn continuously. The organizations winning with AI aren’t the ones talking about models—they’re the ones designing research around human decisions, then using AI to reveal patterns humans alone can’t see. Traditionally, research relied on surveys, panels, and interviews t ...Read More

AI agents are most powerful in product discovery when they act as continuous research partners, not as idea generators. They help teams see patterns earlier, test assumptions faster, and reduce the cost of learning. We think about AI agents in discovery across four practical roles: 1. Synthesis Agents: turning scattered signals into insight AI agents can continuously ingest and synthesize signals that are usually fragmented across: Customer interviews and call transcripts Support tickets and esc ...Read More

The most effective operating model I’ve seen is a tight, outcome‑driven partnership where Product, Data Science, and ML work as a single discovery team—not as sequential handoffs. In practice, that means three things: First, we anchor discovery on a customer or business decision—not a model. Product frames the problem in terms of what decision needs to be improved or what friction needs to be removed, and DS/ML helps assess whether data and modeling can meaningfully move that outcome. This avoid ...Read More

Yes—AI ideas should be invalidated when data constraints make the promise risky.A strong product team does this by pressure‑testing the data before committing to a solution. A practical team‑level process: Start with the data, not the modelTeams should examine data coverage, bias, freshness, and labeling quality early—especially across customer segments, regions, and languages. This helps separate what’s theoretically possible from what’s viable. Translate data gaps into customer riskThe team sh ...Read More