Productside Stories

AI ROI in Product Management. Velocity Isn’t Value.

Featured Guest:

Katherine Man | Group Product Manager at HubSpot
09/22/2026

Summary

In this episode of Productside Stories, Rina Alexin sits down with Katherine Man, Group Product Manager at HubSpot, for what Katherine herself describes as a moderate take on AI: honest about what it has changed, clear-eyed about what it hasn’t, and direct about the problems the industry is building toward whether it notices or not.

Katherine has spent over a decade building platform products for go-to-market teams and is currently leading a full-scale redesign at HubSpot. She makes the case that AI democratized the simplest 10% of every role, exposed how valuable the remaining 90% actually is, and quietly created a talent pipeline problem the industry will feel in five years. Along the way she shares the moment her own designers proved her wrong, HubSpot’s real customer outcome numbers, and why the billboard she saw on the way in from SFO captures everything that’s broken about AI marketing right now.

Key Takeaways

AI took the bottom 10% of your job.

  • The simplest part of the role got democratized. Which leaves the other 90%, the part that makes someone actually good, more exposed than it has ever been. Katherine’s view: the roles did not merge, the conversations got better.

Excuse culture is dead.

  • No PM on the team? AI is your PM. No designer? You have prototyping tools. Katherine has watched a tech lead write product requirement docs for six months and designers run their own research and scoping. Fully resourced teams are now the exception, and she treats that as a planning assumption instead of a complaint.

Output metrics are the easy answer.

  • Velocity is up. Quality systems have not caught up. Katherine points to customer-facing numbers instead: meetings booked through the prospecting agent, resolution rate through the customer agent. That’s the scoreboard that counts.

The grunt work was the training.

  • Writing bad PRDs by hand is how a generation of PMs learned to write good ones. Hiring for recent graduates is thinning while current employees stay put. Katherine is blunt that the industry has a five-year problem it hasn’t priced in.

The overconfidence trap is real, and it has a name.

  • Katherine walked into a design review backed by Claude, certain she was right. She was wrong. Her designers ran the comparison, showed her the breakdown, and she is now in edit mode. She tells the story on herself, and it’s the most useful thing in the episode.

Chapters

  • 00:00:00 – Introduction to Katherine Man and Her Background
  • 00:01:17 – How Katherine Entered Product Management and Her Career Journey
  • 00:03:45 – The Emergence of New Roles and AI’s Influence on Job Titles
  • 00:05:09 – The Blurring of Lines Between Product, Design, and Engineering Roles
  • 00:06:34 – Overstepping with AI and Learning from Team Pushback
  • 00:08:56 – The Importance of Skill and Experience in the Age of AI
  • 00:09:42 – Talent Pipeline Challenges and Opportunities with AI
  • 00:12:07 – AI’s Role in Entrepreneurship and Education
  • 00:14:41 – AI’s Impact on Team Relationships and Collaboration
  • 00:17:12 – Business Outcomes and Measuring AI’s Impact
  • 00:18:24 – Customer Trust, Data Ownership, and Ethical Considerations
  • 00:22:34 – ROI, Costs, and the Economic Impact of AI
  • 00:27:14 – Is AI Hype or Real Value? A Balanced Perspective
  • 00:30:58 – The Future of AI in Product Management and Advice for Leaders

Why Listen to This Episode?

In this episode, you’ll get:

  • A working test for AI ROI, built on customer outcomes rather than adoption dashboards and token leaderboards
  • A realistic model for covering unfilled roles, including where AI stretches and where it stops (prioritization, trade-offs, decisions)
  • The overconfidence trap, told through Katherine’s own story of walking into a design review certain, and wrong
  • A plan for building judgment in junior PMs when the tasks that used to build it are gone

Plus the billboard on the way in from SFO that explains everything wrong with AI marketing right now.

Introduction

Rina Alexin | 00:00:00 – 00:01:17

Hi everyone, and welcome to Productside Stories, the podcast where we dig into the very real and raw lessons learned from product leaders and thinkers all over the world. I’m your host, Rina Alexin, CEO of Productside. And today I’m talking with Katherine Man, a Group Product Manager at HubSpot, who spent over a decade building platforms that power how millions of go-to-market teams work every day, including ours, Katherine. So thank you.

Katherine’s path started in health tech at Athena Health, then data science and pricing at TripAdvisor, before she found a home at HubSpot for the last six years. She’s been featured as a speaker at Inbound for the last three years, recognized as a top 75 product management mentor three years in a row, and recently won Build’s Business Coach of the Year for her volunteering efforts. She’s also on the board for Build.

As you’re about to hear, she has a great take on how AI is and isn’t changing product management. Today we’re talking about where the lines between product, design, and engineering are blurring, what AI has revealed about the value of each function, and where AI is actually moving the needle for businesses. Welcome, Katherine.

Katherine Man | 00:01:17

Thank you so much for having me. Excited to be here.

How Katherine Entered Product Management and Her Career Journey

Rina Alexin | 00:01:21 – 00:01:37

I’m so excited to hear your story. You’ve done a lot of great work for HubSpot and for all of the other organizations you’re involved with. I always like to ask first: how did you end up in product management?

Katherine Man | 00:01:37 – 00:03:45

I like to joke that no one grows up really thinking, “When I grow up, I want to be a product manager.” And I’ve heard really cool stories of how people stumble into it. I have to give a huge thank you to Athena Health, a health tech company in the Boston area, where I got my start.

The story goes that I was doing a market research internship. I would look out the window and see other interns going on duck boats, clearly having more fun. I asked them what they were doing and they said engineering, user experience, and product management. I’ll be honest: my entry into product was partly because it looked like they were having a lot of fun and they definitely had more budget.

A director of product sold me on it when he asked: “Do you know what you want to do when you graduate? Do you like wearing a lot of hats and not really knowing what every day will look like?” And I was like, yes, also yes.

I thought I’d do it for a couple of years to “find what I really wanted to do.” Then I did it for that first year, fell in love with it, and ten-plus years later here I am. Still doing it. Still wearing a lot of hats.

Rina Alexin | 02:59

And product managers wear the most hats of any department.

Katherine Man | 03:06

For sure. Every company does it slightly differently, but the one constant is that every day is completely different. That’s a big reason I love it.

The Emergence of New Roles and AI’s Influence on Job Titles

Rina Alexin | 00:03:16 – 00:03:45

Product management requires a very wide skill set. That’s why organizations trying to figure out how AI changes things often look to product managers to lead the way. But I’m starting to hear new titles: product engineer, technical designer. When you hear those, what’s your first reaction?

Katherine Man | 00:03:59 – 00:05:09

Listeners can’t see the face I’m making right now.

For me personally, those titles are more about following the hype than reflecting real value. And a theme you’ll hear throughout this conversation: focus on the value at the end of the day. Now that we’re about two years into this, what I notice is that it wasn’t about new roles emerging. It’s about expanding the roles we already had.

I think back to “Prompt Engineer” when that first came out. I don’t see that role at many companies anymore. It was really a reflection of a new skill set people had to learn, not a dedicated new function. Same principle applies here.

The Blurring of Lines Between Product, Design, and Engineering

Rina Alexin | 00:04:41 – 00:05:09

I am hearing companies say that now that you can vibe code a solution, a product manager could take an idea all the way through to production-level code. So there is some blurring of lines. Where do you see those overlaps happening most?

Katherine Man | 00:05:11 – 00:06:34

There’s definitely blurring of lines, and I think it’s in a good way. The way I’ve seen it is that it’s really helped me and what we call our triad partners: product management, user experience, and engineering. It’s helped us have really good conversations because we now have more confidence to weigh in on each other’s areas.

For example:

  • For the first time, I’ve had genuine opinions on user experience and the language to articulate specific feedback, rather than just saying “I don’t like it”
  • On the engineering side, when an engineer gives an estimate, I can have better technical conversations. If you’re in Claude Code, you can read the code and ask questions about it

But at the end of the day, what I’ve noticed is that the roles are still there. It’s kind of democratized maybe the bottom 10% of the most simple parts of our roles, but it’s actually made me appreciate the rest of the 90% that makes you really good at something.

I like to think of that Spider-Man meme where everyone’s pointing at each other asking, “Is your role going away? Is your role going away?” And if anything, AI has made me appreciate the other roles more, not less.

“Roles are still there, but it’s democratized.”

Overstepping with AI and Learning from Team Pushback

Rina Alexin | 00:06:30 – 00:06:34

Has there been a moment where somebody stepped out of their usual lane and it didn’t quite land?

Katherine Man | 00:06:37 – 00:08:56

Yes. I’ll tell a story about myself.

Access to new language and the ability to explain things can feel like a superpower. For user feedback, I can take a screenshot, prompt it, share what I don’t like, and get back the specific language I think a designer will really respond to. The problem is that LLMs will make you feel like the best expert in the world, and that confidence can be dangerous.

We’re currently going through a massive redesign at HubSpot. We have a huge focus on ease of use. One principle I took a little too far to heart: for complex editing, I started exploring how to make things as in-place as possible. I went so far as to write up design principles, ran them by my design manager, and then brought them to the designer and the team.

They gave me early, rightful pushback. Yes, in-place editing could work somewhere, but it wouldn’t make sense for advanced customization. You still need an edit mode. And I realized: I was overly confident. I had AI backing me up and I was saying, “Well, according to Claude, there’s a way we should be able to do this.”

The team went and did the comparison work. They came back with real examples showing that in-place editing gets very confusing for more advanced customizations: you can’t distinguish between user level, admin level, and edit mode needs. We are now doing edit mode. And I take it as a learning.

The key takeaway: I had more trust in an AI response than in my own specialists. That was the mistake. I’m really glad the team pushed back.

“Overconfidence can be a risk with AI.”

The Importance of Skill and Experience in the Age of AI

Rina Alexin | 00:08:42 – 00:09:42

What you’re describing is that what makes a role or somebody in that role genuinely skilled is actually even more important now, precisely because AI gives everyone a little too much confidence sometimes. But AI does also help solve for the basic 10% of the work. The flip side of that is what happens to junior designers and junior product managers, whose role is more heavily concentrated in that 10%.

Katherine Man | 00:09:09 – 00:09:42

I see your point. That’s actually what makes the senior specialist so valuable right now. When 90% of the hard work is the thing AI cannot fully replicate, the people who have developed that expertise over years are suddenly more differentiated than they have ever been. The danger is assuming you’ve closed that gap just because AI gave you the language to sound like you have.

Talent Pipeline Challenges and Opportunities with AI

Rina Alexin | 00:09:35 – 00:09:42

I’ve heard this from many product leaders: how do you even build a pipeline of talent when some of those junior roles don’t hold up anymore?

Katherine Man | 00:09:45 – 00:12:07

Yeah. This is something I’m especially concerned about, because the data shows it’s not that companies are laying off current employees. The problem is a growing and accelerating gap in hiring for recent college graduates. And I think that’s going to become a real problem.

Think about how many of us got to where we are in our careers: it came from doing what I’d call the grunt work. I start saying things like, “Back in my day, I had to write my product requirement docs by hand.” And for any engineers I’ve worked with, they know I hate documentation. Now I can generate one with a few keystrokes. But I don’t think I would have learned how to write a good one if I hadn’t gone through all that manual effort.

The key risks the industry is facing:

  • Recent graduate hiring is declining while current employees stay in place
  • The grunt work that built PM judgment is increasingly automated away
  • The talent gap will likely show up in two to three years, but will probably hit hardest in around five

The irony: this generation is going to be the most AI-native and fluent candidates. And the data does show a silver lining: Gen Z is now outfounding Baby Boomers at 9% to 5%. If they can’t get traditional jobs, they’re starting businesses. But entrepreneurship is not a path for everyone.

AI’s Role in Entrepreneurship and Education

Rina Alexin | 00:12:14 – 00:14:41

I do believe in entrepreneurship, and I’m hopeful that organizations like the one you’re involved with will help solve what is really a social problem: making sure we still have opportunities to build experience and develop the judgment to know when AI is making you overconfident. But when roles on teams go unfilled, can AI stretch to cover the gap?

Katherine Man | 00:12:48 – 00:14:41

Yes, for sure, and that’s one of the biggest benefits I’ve seen. Going back to what I said earlier, I don’t think AI can completely replace the role, but you can stretch it. And I like to say it killed the excuse culture.

What I tell my teams is:

  • If you don’t have a PM, Claude is your PM
  • If you don’t have a designer, you have Lovable or Claude Code
  • If you don’t have an engineer, that one’s still tough, but you can stretch with Claude Code and ship code yourself for minor bugs or fixes

It’s actually very lucky to have a fully resourced team. Given the hiring environment in the industry right now, we have to set the expectation that not every role will get backfilled. And AI has been genuinely helpful there. One of my teams went without a PM for almost six months. The tech lead stepped up and started writing product requirements docs, running them by me but really owning it herself. Designers stepped up to lead user research and scope definition. The places where I still needed to weigh in were complex decision-making, prioritization, and trade-offs. But AI can get you pretty far even there, at least for a while.

AI’s Impact on Team Relationships and Collaboration

Rina Alexin | 00:14:39 – 00:17:12

How has AI use changed relationships on your team across different functions?

Katherine Man | 00:14:47 – 00:17:12

It’s changed for the better, though it followed an arc. At the very start, we were all dabbling in each other’s areas. Designers were contributing to PRDs. Engineers were weighing in on product decisions. It was a period of confusion about what AI actually changes. What we’ve more realized is:

  • The roles have stayed the same, but everyone is doing their own role faster
  • Designers have faster turnaround time
  • Product managers can write multiple PRDs simultaneously
  • Engineers at HubSpot have seen roughly a 60% increase in velocity

What’s really changed is the quality of the conversations. We can challenge each other more substantively. But the roles haven’t fundamentally merged. It’s accelerated everything and made previously slow parts of the job easier.

A concrete example: HubSpot is very customer-driven. We really believe in talking to customers. But customer interviews used to mean ten calls, no central place to store scripts, and then two to three weeks of painstaking work pulling together the research report. Now it’s a one-week turnaround. I upload all my transcripts from Fellow, prompt it for trends, and I have my document. The output is faster and quality of life for the work that used to grind is significantly better.

Business Outcomes and Measuring AI’s Impact

Rina Alexin | 00:16:26 – 00:18:24

I’m hearing a lot of output-related measurements. Our research found that most organizations measuring AI impact restrict it to adoption rate or productivity gains. Very few, less than 25%, measure anything outside that, like actual business outcomes. Velocity is improving, outputs are faster, time is being saved. But what about quality changes since leaning more into AI?

Katherine Man | 00:17:20 – 00:18:24

Velocity has absolutely increased. I’ll be honest and say that quality is still finding its balance. If you increase your output two to three times and your systems aren’t in place to catch the issues that creates, that’s something we’re currently investing in.

What I’m grateful for: HubSpot was never one of those companies with token-maxing leaderboards. That kind of metric incentivizes people in entirely the wrong direction.

What we focus on instead is early positive indicators of customer business outcomes:

1. Prospecting agent: sales reps using AI to find and reach new prospects are seeing an 80% increase in meetings booked
2. Customer agent: on the support side, the resolution rate has increased 2.3 times

“Velocity has increased, but quality is still catching up.”

Those customer-facing numbers are the scoreboard that counts.

Customer Trust, Data Ownership, and Ethical Considerations

Rina Alexin | 00:18:27 – 00:22:34

Do you have any concerns about shipping features with AI in terms of how customers perceive it? Especially given that HubSpot’s customers treat their data as theirs, not HubSpot’s. There’s a perception that once you apply AI to something, you lose ownership of that data.

Katherine Man | 00:18:32 – 00:22:34

There’s still a lot of trust to build. Working at tech companies, we tend to be the early adopters. But HubSpot serves small and medium-sized businesses. These are extremely busy people who don’t have time to invest in learning new technologies, and we have to be really careful with the quality and suggestions we’re putting in front of them.

What we’ve learned about building trust with AI suggestions:

  • Cite sources and explain where a suggestion came from
  • Explain the logic for why something is being recommended
  • Allow customers to customize even the prompts so outputs are relevant to their specific business

The data ownership question is a genuinely tough problem. Everyone wants the context that makes answers better, but they don’t want to offer the data that creates the context. I’ll even admit that on Claude and ChatGPT, I turn off training on my own personal data, even though I appreciate the quality that comes from users who volunteer theirs. There is a real tension there.

Where I think this is heading: a more closed-loop system where customers can be told, “Anything you put in stays in your system.” That’s probably where we need to catch up. But at the end of the day, if you want really good context for your AI to act on, you have to be willing to give up some amount of data. You absolutely would not want that data to be available to a competitor. It’s a genuine balance, and the industry hasn’t fully resolved it yet.

ROI, Costs, and the Economic Impact of AI

Rina Alexin | 00:21:40 – 00:27:14

Product leadership and company leadership tend to be very sensitive to costs. There’s pressure to use AI and build it into products, but the cost can be quite high. How do you think about ROI when the investment is significant even if you’re seeing business outcomes?

Katherine Man | 00:22:44 – 00:27:14

It’s tough. ROI is hard to make tangible internally right now.

Let me put it this way: if someone came to me today and said we’re removing access to all AI tools, I would feel like I can’t do my job. I’m completely dependent on Claude Chat, Cowork, and Claude Code at this point. That dependency is itself a form of ROI, even if it’s hard to put a number on.

On the flip side, if you try to measure that in terms of output and outcome: velocity is up, and we’re also seeing that the bar for quality has risen. Before, we might have shipped more mediocre product. Now it’s higher-quality product. But those two things are hard to quantify against each other.

A few practical observations on managing costs:

– We run internal training courses on how to use AI more efficiently
– Simple habits like using Projects to avoid re-sending massive prompts every time can meaningfully reduce token costs
– There are ways to cut down spending without reducing capability

The clearest ROI proof points I can point to are the migration and redesign timelines. Complex migrations that would have taken a typical engineering estimate now complete roughly four times faster. And the massive redesign we’re currently executing would never have been considered feasible within a single year. Here we are doing it in a year, possibly in six months.

The best measurement I can point to is the business outcomes: those leading indicators are positive. The lagging financial indicators will take another one to two years to fully catch up. But I’m confident we’ll see it.

 “I can’t see my life without it now.”

On the cost-of-not-doing-the-wrong-thing ROI: one of the hardest things to measure is how much value you preserve by discovering faster that something won’t work. Previously, shipping a prototype and getting customer reactions could take weeks. Now that you can prototype quickly and get reactions fast, you’re avoiding an entire category of expensive, late-stage pivots. That’s real ROI even if it doesn’t show up on a balance sheet.

Is AI Hype or Real Value? A Balanced Perspective

Rina Alexin | 00:27:01 – 00:30:58

A lot of people are quite nervous we’re in the middle of a hype cycle. Less opportunity than we think, more cost than we’re predicting, and therefore it’s all going to fall. Where do you think it actually is?

Katherine Man | 00:27:27 – 00:30:58

That’s where I see myself as moderate. I think we are absolutely on an S-curve in terms of technology, comparable to the shift from mobile to web and everything that unlocked. AI has democratized access to things that simply weren’t possible before. The clearest example: I know non-technical founders who have built their entire businesses on Lovable. That wasn’t possible two years ago.

The moderate take, though, acknowledges the real dangers:

  • Environmental impact and the energy cost of running large models
  • Widening socioeconomic gaps if access isn’t equalized
  • The hype is absolutely real, it’s just focused on the wrong thing

The part I disagree with is that the hype is about the technology rather than the customer problem you’re solving for. Story: I was in San Francisco for HubSpot’s annual Inbound conference last year. On the way in from the airport, I was seeing increasingly nonsensical AI billboards. It culminated in one that said: “Do you even AI?”

What does that even mean? We’ve completely lost the thread.

It has absolutely proven to provide real value. What I look forward to is when it just becomes table stakes. As a consumer, I now expect you to be using AI. I’m more concerned if you’re not. You don’t have to shout it out. I’d rather see what got built faster. We also forget that AI is a very broad term that has existed for a long time. We are specifically talking about LLMs. Machine learning and data science have been here for decades.

The Future of AI in Product Management and Advice for Leaders

Rina Alexin | 00:31:11 – 00:30:58

Based on your experience and moderate perspective: what is one piece of advice you would leave product managers and product leaders who are really trying to figure out where AI fits on their team?

Katherine Man | 00:31:34 – 00:35:09

Nothing can replace actually using it and trying it. My best advice: use it. By doing so you’re going to discover what it’s good at and what it’s not good at.

A concrete example of where this matters: there’s a big debate right now about whether the future of CRM will be completely AI-generated or not. My moderate take is it’s going to be a balance. LLMs are really good for generating things that are more creative, content-heavy, or don’t need a lot of structure. That’s why they’re excellent at mockups and writing docs. But for a CRM, there’s still a certain amount of structure and layout that users need. If I told sales reps that every time they loaded a record page, the layout might look completely different, that would be a nightmare. Muscle memory matters.

So the balance looks like:

  • Layout and structure for muscle-memory workflows: more traditional UI
  • Extensions, customizations, and integrations: absolutely, LLM generation makes sense

And the same thing I tell my teams: you are not going to build good products if you aren’t using AI yourself and finding what it’s good for.

My personal example: I built myself a fitness app on Lovable to help myself work out more consistently. I’ll be down in the gym editing the app on mobile right there. Lovable’s mobile editing experience is actually excellent, which I didn’t expect. And that firsthand learning directly informs how I think about what’s possible when I’m building product.

Connecting with Katherine and Closing Remarks

Rina Alexin | 00:33:55 – 00:35:09

I want to say, on behalf of my sales team, that I’m very glad you’re not changing the way our CRM looks every time we open it. Katherine, wonderful discussion. Thank you so much for being here. How can our listeners follow you or connect with you?

Katherine Man | 00:34:14 – 00:35:09

LinkedIn is probably the easiest way to find me. Search for Katherine Man at HubSpot. In my personal life I’m a bit of a Luddite: not on X, on Instagram but it’s very private and mostly just puppies and tennis. But yes, please reach out on LinkedIn. I would love to hear from you, keep the conversation going, and I’m always happy to chat or give advice.

Rina Alexin | 00:34:40

If you’re advertising puppies, I’m pretty sure some people want to hand you an Instagram.

Katherine Man | 00:34:44

That’s most of what my feed is. That’s how I relax.

Rina Alexin | 00:34:49 – 00:38:19

Awesome. Well, thank you so much. And thank you for listening to this episode of Productside Stories. If you liked today’s conversation, please don’t keep it to yourself. Share it with a friend and make sure to subscribe so you don’t miss a future episode. I’m Rina Alexin, and from all of us at Productside: let’s do product better, together.

Katherine Man | 00:35:09

Thanks so much.