CPO and AI-native founder shares how to intelligently manage a roadmap with AI

Tara Sakhuja is the CPO of Luv or Pop, and the founder of an AI-native business called Data Dumpling. Before these roles, she worked in product at companies like Meta and Bumble.

We sat down with Tara to learn how she's changing her PM workflows with AI. Along the way, she shared one of the biggest traps for people building AI products.

Intelligent products, built intelligently

I'm Tara Sakhuja. I operate at the intersection of AI, product strategy, and human behavior.

I’m the Founder and CEO of Data Dumpling, an AI platform helping companies understand who their customers are, what signals matter, and where their teams should focus. I’m also Chief Product Officer at Luv or Pop, where we’re rethinking dating and online communities through one of the most viral shows in the world, Pop the Balloon or Find Love with Arlette Amuli.

Before this, I spent a decade building products at Meta and later at Bumble, working across ads, commerce, business messaging, integrity, trust, and machine learning systems.

A single question connects all these experiences: How do you build systems that help people make better decisions?

At Meta, I built systems operating at the scale of billions of users. At Bumble, I focused on safety, trust, and how AI impacts deeply human experiences. Now, as a founder, I build AI-native products from first principles.

We’re entering one of the biggest shifts in product development since mobile. The biggest change isn’t that products are becoming intelligent. It’s that the way we build products has to become intelligent too.

The next generation of products

Data Dumpling is AI-native. It connects company intelligence, buying committees, relationship signals, and real-time intent, helping teams identify their highest-value opportunities and understand what action to take next. We work with companies across enterprise technology, consulting, and industrial sectors, helping revenue teams move from static lists and manual research to AI-driven decision systems.

At Luv or Pop, I lead product strategy for a consumer social platform reinventing dating with trust, identity, and community. We’ve scaled to tens of thousands of users across the US, UK, Canada, and parts of Africa, building systems for verification, matching, and authentic human connection.

While the products seem very different, the underlying product philosophy is the same: The next generation of products won’t just help people do more work. They’ll help people make better decisions.

Whether helping a GTM team find their next customer or helping someone find a meaningful relationship, the challenge is designing technology that compounds human judgment.

How AI is allowing smaller teams to achieve product rigor

One of the biggest shifts for me is how AI has increased the operating leverage of small product teams.

When I was at Meta, we had significant infrastructure for decision-making. Dedicated teams, mature experimentation platforms, dashboards, data scientists, researchers, and analysts helped product teams understand user behavior and make better decisions.

Historically, only companies with hundreds or thousands of people could achieve that level of product rigor. AI is changing that.

Even with my smaller teams, we use AI heavily to understand user interactions and inform our next focus areas.

We connect our product data to Claude and use tools like Mixpanel to understand user behavior. Previously, a small team would spend significant time building dashboards, writing queries, and manually analyzing patterns before addressing the product question. Now, we can have a conversation with our data.

So, when we launch a new feature or run a campaign, we can quickly understand:

• Which user segments are responding?
• Where are people dropping off?
• Which behaviors correlate with retention?
• What should we test next?

That doesn’t mean AI makes the decision for us. It changes where we spend our time.

The human role moves away from pulling data and toward asking better questions.

A model can identify metric movement, segment the change, and suggest possible explanations, but at the end of the day, human product judgment can actually tell you which hypotheses are credible, whether the movement matters, and what to do next.

How rapid prototyping changes the conversation

AI reduces the distance between an idea and real customer feedback.

We use AI heavily during early product development. Instead of starting with a written concept document, we now quickly vibe code prototypes and interactive experiences that allow the team to feel the product before committing engineering resources.

That has changed the quality of conversations we have internally and externally.

A product idea can sound great in a document, but the moment someone interacts with it, you notice things you would otherwise miss: user friction, emotional reactions, confusing flows, or assumptions that don't hold up.

We rigorously stress-test prototypes, collect feedback, iterate, and only then move into full development.

And this is particularly important because AI also makes it easier to build impressive things that don’t solve meaningful problems. For example, at Data Dumpling, we experimented with synthetic customer personas because we had rich intelligence about companies, buyers, and decision-makers. We could simulate how different personas might respond to messaging or campaigns. Technically, it was one of the most interesting things we built. But our customers didn't need it.

Tara's Thoughts

We rigorously stress-test prototypes, collect feedback, iterate, and only then move into full development. And this is particularly important because AI also makes it easier to build impressive things that don’t solve meaningful problems.

How AI affects customer discovery and roadmaps

Historically, customer discovery was lossy. A PM would take notes, summarize conversations, identify themes, and translate those insights into roadmap decisions. Much depended on what someone remembered or documented.

I’ve rebuilt that workflow with AI. Read AI captures every customer and prospect conversation, and Claude synthesizes them. We’ve built custom skills that map those conversations directly against our roadmap, helping us understand which problems come up repeatedly, which customer segments care about them, and how those requests connect to revenue opportunities.

At any point, we can understand not just how many customers asked for a feature, but also how important the problem is, who is experiencing it, and whether solving it moves the business forward.

The result is that product decisions rely on a continuously updating customer intelligence system rather than a collection of static notes and one-off conversations.

Why PMs should be wary of building based on what foundational models can't do

If you're building AI products, you need to understand how quickly foundation models improve. Just last year, one of the first AI capabilities we built at Data Dumpling involved content generation. At the time, generating highly contextual outreach using company intelligence felt like a real competitive advantage.

By the end of the year, models had dramatically improved their ability to understand images, links, and long-form context and to generate high-quality content out of the box. Something that initially felt differentiated became table stakes remarkably quickly.

I think it's dangerous to build products whose primary value relies on what the model can't do today. Instead, we focus on building around customer workflows, proprietary data, and systems that become more valuable over time.

Why product leaders must experiment, but be decisive

My advice is simple: be curious, but be decisive.

New models, tools, and capabilities emerge almost every week. No one has all the answers, so teams that learn fastest are those willing to experiment.

Start with any slow, repetitive, or frustrating part of your product development process. Try different AI tools. See what genuinely improves your team's work. Measure the impact, keep what works, and don't be afraid to abandon what doesn't.

At the same time, there's a temptation right now to rebuild everything because AI exists. That's not the right approach.

If a process already works well, AI doesn't necessarily belong there. AI should solve real problems, not create new ones.

Also, not every decision should be automated. At Data Dumpling, while AI automates much of the lead discovery, buying committee mapping, and intent signal analysis, we build our entire workflow around a human in the loop. AI is exceptional at surfacing opportunities, connecting signals, and prioritizing where to focus attention. But a person still makes the final decision, whether that's pursuing an account, changing a strategy, or deciding how to act on a complex buying signal.

That's a deliberate design choice. The goal isn't to replace judgment. It's to give people better context so they can make better decisions.

Ultimately, success in the AI era doesn't come from using the most tools. It comes from knowing where AI creates leverage and where human judgment still matters.

Follow along

You can follow Tara Sakhuja's work on LinkedIn and Instagram. And check out Data Dumpling and Luv or Pop.

More expert interviews to come on The CPO Club!