I’m talking to Bhaskar Sunkara, CEO of bicycle.AI, which provides an AI analyst product designed to monitor revenue-critical KPIs, investigate the business and technical drivers behind KPI changes, and take a “governed next step.” Bhaskar explains why analytics products often fail when they overwhelm users with telemetry instead of focusing on the signals that matter. Drawing from his experience as founding CTO of AppDynamics, he shares how his team moved from low-level technical monitoring to business transactions like logins, checkouts, and bookings. The key lesson? Start with the right metric at the right level of granularity, then use deeper technical analysis to explain why something changed.

Bhaskar also breaks down how bicycle.AI serves multiple audiences inside an enterprise. Business leaders want measurable outcomes, KPI owners need answers about what changed and what to do next, and data teams require trust, governance, and traceability. He explains how, in order to support these different users, Bicycle separates product experience into four core surfaces: pull features like dashboards and chat, and push features like alerts and data stories. Alerts further help operational users respond quickly to KPI changes and data stories provide executives with strategic narratives around trends, causes, and business impact. During our chat, Bhaskar also draws a line most AI products blur: be explicit about which findings are deterministic and which are only a theory. He connects this directly to my CED framework, separating the conclusion from the evidence from the underlying data, and argues that how much you automate should be governed by one question: how costly is being wrong?

I also probed Bhaskar about their moat. He’s learned that enterprise adoption requires winning over both executives who care about revenue impact and analytics teams that need confidence in the system’s recommendations. Bhaskar also explains why their long-term advantage comes from the DEAL framework: Detect, Explain, Act, and Learn. By continuously incorporating validated decisions, business context, and customer-specific knowledge, the platform becomes more useful over time. We finish up with his advice for fellow analytical AI product founders, including why AI makes user experience more important, not less: it is the connection between agents, decisions, humans, and accountability.

Highlights / Skip to:

Links

Today, I'm talking to Rana Gujral, CEO of Behavioral Signals, which provides AI that interprets human behavioral cues in speech to help route call center conversations more effectively, improve customer service performance, and detect voice-based fraud. Their moat is a decade of voice data tied to real business outcomes, not the model itself, as Rana explains.

During our conversation, Rana shares his practical framework for making the value of their AI obvious to the various humans in the loop that the product needs to “touch,” and he argues that a one size [UI] doesn’t fit all. In Rana’s product, they discovered that customer service reps need ambient assistance, supervisors need aggregate patterns, compliance teams need audit trails, and executives need outcome metrics tied to business results.

He also explains why having measurable ROI isn't enough. Early renewals for Behavioral Signals suffered because the people signing the checks couldn't actually see the product's impact. Rana's solution? “Ship the meter” alongside the intelligence. If your AI works quietly in the background, you still need reporting UIs that clearly communicate the product’s value.

For founders struggling with stalled POCs, Rana breaks down the three-stage evaluation journey his team developed after repeatedly seeing deals fail at predictable moments. By designing the customer experience around those milestones, his team transformed how buyers gained confidence throughout the evaluation process.

Finally, we explored why great B2B AI products don't succeed by becoming another dashboard. Rather, they succeed by closing the loop between decisions, outcomes, and learning. Rana also fills me in on his upcoming book, The AI Instinct, which focuses on how AI changes human judgment rather than simply advancing model capabilities. And his parting advice? Listen to find out!

Highlights / Skip to:

Links

I've been seeing teams heavily invest in copilots, agents, semantic layers, governance frameworks, and increasingly sophisticated models, yet many still hear the same feedback from sales prospects: “We may just build this ourselves?" Or they don’t hear it, but suspect the customer is doing just that. 

Whether they actually can DIY the solution is the wrong question. The bigger question is *why they believe they can.* Your product may have a genuine competitive advantage, but your real challenge is that this advantage isn't obvious to buyers. The moat exists, but it is invisible.

What makes this relevant is that many capabilities once considered differentiators are rapidly becoming normalized. AI copilots, agentic analytics, governed data, semantic layers, and broad integrations now appear across nearly every platform in the category. As AI accelerates development, sophisticated engineering alone becomes harder to defend as a lasting advantage.

So what actually creates a durable moat if the engineering and product seems easy to copy? I explore four areas: proprietary data, trusted relationships, and products that accumulate institutional knowledge remain difficult to replicate. And finally, user experience itself as a strategy. As users increasingly access your intelligence through AI agents rather than dashboards, their experience may become the moat that competitors can't copy.

I've been seeing a recurring pattern with companies selling APIs, MCPs, data feeds, and other developer-focused AI products. While the technology is often sound if not impressive, sales momentum sometimes slows when prospects have to imagine how the product will create value in their own environment. My perspective on this is that the flexibility that makes these tools powerful can also make them harder to evaluate.

Flexibility can adversely increase the Invisible Intelligence Gap, and I think certain types of AI-based solutions (LLM) may actually increase this because the boundaries of the product are often so much wider than ever before (if not invisible to the buyer). So, how to close this gap? Well, one way is to build a visual UI that showcases what’s possible with your API/feed/data solution. You take the buyer out of the conceptual space and make things concrete. So today, that’s what we dig into: when to consider adding a UI, how far you need to go with it, how you can use Copilot/AI agents to help customize these example implementations, and the benefits you might see. 

It’s a common pattern for teams building B2B analytics and AI products: the proof-of-concept goes well, the buyers sound excited, and everyone assumes the deal is about to close—until it quietly stalls out. The assumption is usually that sales needs to follow up harder or marketing needs more enablement material. But often, the real issue is that the product itself cannot communicate its value without humans in the room explaining it.

I call this the Invisible Intelligence Gap. Buyers may understand the promise during a guided demo, but once the sales engineers leave, customers are left trying to figure out workflows, use cases, trust concerns, integrations, and organizational fit on their own. This gets even harder with broad, general-purpose AI tools and chat-based interfaces that sometimes assume users already know what to ask.

The solution isn’t simply shipping more features or training content. It’s designing products that clearly reveal their value, reduce customer effort, and continue selling themselves after the POC ends, and getting that design right starts with the right product strategy. 

If you’re hoping that adding AI to your analytics product or capabilities is going to unlock new revenue, sales, and greater user adoption, but you’re not sure what’s involved in this transformation, this episode is for you!

Today, I’m talking with Juan Sequeda today, an expert in knowledge graphs and ontologies who most recently was Head of the AI lab at data.world, which was recently acquired by ServiceNow. Juan and I met while speaking at CDOIQ a few years ago, and after being on his former podcast “Catalogs and Cocktails.” (With a name like that, I naturally had him out to my local tiki bar while visiting Cambridge!)

Talk-to-your-data products – effectively next-gen business intelligence applications – are a hot topic right now, and this has made much of Juan’s PhD work in semantics highly relevant right now as companies try to make analytics more user-friendly via natural language. 

Juan is clear that the starting point for this transformation isn’t the model or the UI, but actually the customer’s workflow—and that was like music to my ears! Analytics only matters when it drives action, so the real challenge is not answering more questions, but enabling better decisions and outcomes.

A key theme is semantics, which, in product design language, I think of as making users’ mental models of their business or domain map logically to system and data models so that AI produces the right answers in the right context. Juan outlines a practical path to getting started with this: strong data modeling, a well-defined semantic layer, buy-vs-build considerations, and throughout, a constant focus on what the customer’s workflow and problem is.

Speed is often confused with good product thinking. The idea is that if teams can ship prototypes, dashboards, and models faster, they will automatically learn faster. But execution speed alone doesn’t ensure a clearer understanding of what’s actually worth building.

Instead, teams often fall into a loop driven by demo feedback. They present working prototypes, and users respond to what they can see in the form of interface design, visualizations, or surface-level data behavior. While this feedback feels positive, it’s often misleading. Teams can end up reacting to presentation (UI) feedback only to find it does not change propensity to buy or increase user adoption.

The key idea today is that prototypes can either be used to clarify the problem space and user needs or to validate the solution presented. Where I see most teams fail is that every artifact or prototype is seen as a solution to validate, and they can miss the forest for the trees.

Another approach borrows from blue ocean thinking, which focuses on creating value by looking for overlooked opportunities in the empty space—beyond the known “problem space” your customer knowingly lives in now. 

Because AI lets us move so fast with prototyping, I think there is an exciting possibility to explore the blue-ocean spaces where your product could evolve to produce value. 

As always, we seek to go beyond building “technically right, effectively wrong”—which doesn’t make people buy, use, or refer your product.  Today, we look at what AI can help us to do to see even farther beyond the immediate problem space. 

I’ve seen this challenge again and again with teams building analytics and AI products: nobody can define what quality to the end user means or how to measure. The answer? “Adoption.”  The problem is that “amount of usage” tells you nothing useful about your customer’s experience with your product beyond “it’s not zero.” So what should you be measuring instead so your buyers don’t quickly abandon once the end users get their hands on the keyboard (or agent!)?

The answer is to understand through qualitative measures what users’ experiences are like now, so you have an objective baseline from which to compare future product investment. When you can define their current experience’s quality, it’s much easier to imagine their better future, and you also now have a change you can measure. Measurable outcomes are the foundation of high-value, sticky B2B analytics and intelligence products—and when your end users’ lives are improved, the sales close, and the renewals aren’t questioned. So today, I jump into “how do you measure UX?” so you aren’t surprised when the sale doesn’t close or that renewal doesn’t come through unexpectedly. 

I'm talking with Steve Ancheta, CEO of Zig, a platform designed to free sales teams from repetitive, non-revenue-generating tasks. CRM and logistical tasks can consume up to 72% of the week of a sales team, but Zig’s AI agents handle them so reps can focus on closing deals. Unlike tools built for managers, Zig follows a rep-first design—simple, intuitive, and aligned with the motivation to sell more—while also creating an intelligence layer that preserves institutional knowledge and accelerates onboarding for new hires. 

I wanted to chat with Steve about how he built a product that is both used—and worth paying for—with AI under the hood. Rather than relying on chat prompts, Zig surfaces prioritized tasks in panels and cards, integrates with CRMs and Slack, and builds confidence scores from user interactions. 

Because Steve comes from the world of sales—and that’s the domain his product sits in—I wanted to explore his “problem clarity” and share that with you, since I often find data and technical founders to be more solution-oriented and lacking in this area.  Steve was an open book with me, and I’m hoping other founders trying to turn analytical complexity into commercial clarity can see how Steve is using AI and agents to make data work for end users—and worth paying for.

Finally, I also challenge Steve to answer whether Zig.ai is a software company or a services company with a product behind the scenes—a question you might also ask yourself depending on your GTM model.

I’ve seen this pattern repeatedly with teams building analytics and AI products: the issue usually isn’t the quality of the models or the sophistication of the data. The technology often works just fine. The real breakdown happens earlier—when teams begin with the data they already have and try to figure out what to build, instead of starting with the decisions their customers need to make.

That approach often produces polished dashboards and compelling features that generate interest, but fail to drive real action. The missing piece is context. Decisions in the real world depend on incentives, habits, risk tolerance, and uncertainty—not just clean data. If your product doesn’t reflect that reality, it won’t meaningfully change behavior.

Another common trap is assuming all available data is *evidence* worth surfacing. This “more is better” mindset leads to cluttered analytics tools that offload interpretation onto users. Even conversational AI interfaces can fall into this, encouraging open-ended exploration without helping users reach decisions.

The analytics and AI products that succeed take a different approach. They’re designed around decision-making to reduce uncertainty, fit into real workflows, and guide users toward clear actions. In doing so, they bridge the gap between analytical capability and real-world value, making the product’s intelligence tangible, usable, and worth paying for.