Last week, we hosted a panel discussion on scaling software businesses in the AI era, bringing together perspectives from investors, advisors, and operators: Hank Chen, Partner at Bain & Company; Isabela "Bela" Peralta, Principal at Insight Partners; Kunal Agarwal, CFO of Gorgias; and Katherine Zhang, CEO of OPEXEngine by Bain & Company.
The conversation centered on a fundamental question: Is AI changing what it means to be a “great” software business?
We explored how investors are rethinking moats and revenue durability, how AI is changing the economics of growth and gross margins, where companies are finding real operating leverage, and which metrics leaders should be watching as agentic products become more prevalent.
A few themes stood out: traditional measures like retention still matter, but leaders need to look more closely at the quality and durability of revenue; AI can expand addressable markets; and capturing AI’s full potential often requires rethinking workflows and organizational structures, rather than simply adding AI to existing ways of working.
You can watch the full panel recording, listen to the conversation as a podcast, or read the full transcript below.
Transcript
Katherine Zhang: Welcome, everyone, to this OPEXEngine panel and thank you for joining. I’ll start and tell you a little bit about why we brought this panel together today. I'm Katherine Zhang, the CEO of OPEXEngine by Bain & Company. As we were thinking about putting together a conversation that would be interesting to leaders in the software space, we thought about it in this way…
We all know that AI is changing a lot of the software products that we make and work on today, but we're in tech, and we're used to products changing, right? I think the more interesting question for a lot of us is: is AI fundamentally changing what it means to be a good software business?
Do the metrics that we've traditionally used to judge software companies – such as retention and Rule of 40 – still tell us the same things, or has the growth versus profitability equation changed? And how are investors and operators thinking differently about these questions?
That’s why we put together today's panel: to discuss some of those questions from multiple perspectives. Our hope is that you'll leave here with a clear understanding of how the economics of scaling a software company have changed, how they haven't changed, and where you can focus as a leader.
With that, I'm going to start with some introductions. Like I said, I'm Katherine. I'm the CEO of OPEXEngine by Bain & Company, where we combine operating data from hundreds of software companies to give leaders an independent external perspective on their business performance so they can make better strategic decisions. I'll turn it over to Hank to introduce himself.
Hank Chen: Good afternoon, everybody. I'm Hank Chen. I'm a Partner in Bain & Company's Boston office. I lead our tech insights practice, where we work with the world's leading investors and management teams to look at software companies and improve value. I’m excited to join this panel today and be with you here.
Isabela Peralta: Hi, everyone. I'm Bela, and I'm a Principal at Insight Partners. I've been with the fund for eight years now. I focus my time specifically on due diligence, growth, and strategy for our portfolio, with a strong concentration on healthcare and life sciences. I’m excited to be here today.
Kunal Agarwal: Hi, everyone. I'm Kunal. I'm the CFO of a company called Gorgias. We are a conversational commerce platform -- for Shopify in particular. So we’re selling agentic products, so this feels like a pretty relevant conversation. I'm excited to be here.
Katherine Zhang: Great. We're going to dive in at the very highest level of the questions today: what creates value in a software company today? I will go to Bela first to kick off this question.
Isabela Peralta: I think that to answer this question, it'd be silly of us not to also account for what the market has said drives value today, right?
I think that if you look at Q1, the market was very clear on the potential risks that accounted for in software. I think the North American software index fell around 24% in Q1. The S&P was down 4%. Morningstar rerated around 130 of the software companies when looking at it through an AI lens, given reduction in kind of perceived moats.
I think what's interesting is that if you look at the underlying economics of a lot of these companies that were rerated… their retention rates hadn't moved. ARR and bookings were still outpacing revenue. Margins were still climbing. So the cut didn't come from performance actually deteriorating.
I think what drove the cut was from the expected pricing power and defensibility of some of these companies. I think that everybody started to worry about AI lowering the barrier to entry, competition increasing, and questions around retention that today look bulletproof -- how would that all look in the coming years?
I think to answer your question around what really is driving value today in software companies, it is the answer to the question of: what can you do that your models can’t copy? Because we've seen the model layer starting to commoditize. So maybe to get a little bit more specific around how we think about this defensibility and creating value in software companies, I'd boil it down to four different characteristics:
One being access to proprietary data that compounds -- and I want to be very clear here that I'm not talking about data volume. I'm talking about the feedback loops that actually improve your product with each customer interaction.
Second, I think it's whether you are truly the system of record or system of action. How embedded are you into the workflows of the day-to-day customers? Third is access to distribution across your installed base, and fourth just trust and compliance.
I might be a little bit biased because I basically grew up in healthcare, but I wasn't surprised to see that Veeva was one of the few companies that got a moat upgrade when looking at it through an AI lens because their data is tied to regulatory requirements and mission-critical workflows. So I think it really comes down to: how are we redefining moat and revenue durability?
Katherine Zhang: Thanks, Bela. Hank or Kunal, anything you'd like to add to that? I think this idea of: “are moats still the same? Have they changed?” is kind of an interesting one to go after in this AI era.
Hank Chen: That’s 100% where investors' heads are at. I think when we're doing work for leading investors, it's probably the number one question in the diligence now, right? Where is the moat? Has the moat eroded?
And as Bela was saying, nothing changed in the performance of those companies, but if that moat has eroded, all of a sudden the terminal value of the model is no longer going to work anymore. I think that was an assumption that worked for a very long time for SaaS companies, and now there’s a serious question there around: will that drop off a cliff? Am I in danger with this particular [portfolio company] or this target we're looking at? That is definitely the number one question we're thinking about these days.
Katherine Zhang: And is it any different today than it was before? Because we've always looked at moats in looking at tech assets. That's a question for any of you.
Hank Chen: Absolutely. I think we've always looked at moats. In the past, maybe it was more from a competitive point of view or, you know, I'm looking at this company in this particular space. There are other incumbents. There are also new entrants, and what is my defensibility of this particular company relative to, perhaps, a newer entrant? But I think that's shifted now completely.
Isabela Peralta: Yeah, I do think the one thing that has changed when it comes to moats is our views around efficiency. I think that in the past we looked at potential tech teams and your ability to deploy features and be efficient with your resources as a differentiator. Now, I think that's table stakes because anyone can build lean. But I think that underlying your question, Katherine, is this question around: has our definition of a great software company changed?
And coming from Insight Partners where we're very bullish on the software landscape as a space, we get this question around: what is your view on software versus AI? That, to me, is kind of like saying, what is your view on apples versus apples?
Definitionally, AI is software. I think we're at a moment in time where [there’s] a distinction, but really without a difference. What makes a software company great is still what makes an AI company great. It's about: how embedded are you into the customers? Do you have compounding data? Do you have strong gross retention? I think that the game, actually, is still pretty much the same.
Kunal Agarwal: Yeah, I've got a couple of views on this. I agree that the system of record is there. I think the [way you evaluate the moats] – you know, as a company shipping and building agentic products, the moat is, one, the access. It's not only a system of record, but it's kind of like the transaction depth because we're a support company, and so it's, say, someone asking a merchant, "Where's my order?"
And that's what the agent does. Like, anybody can go and say, "Hey, we built an agent to go and answer 'Where's my order?'" but what differentiates us is now the ability to go issue the refund, edit the shipping address, reship it, and going into all those other places that sit within your execution layer. So it’s transaction depth plus your right access to the system of record. I think that's a differentiator in how you build your agentic product, right?
I think for the second piece – thinking about like, for us, what's actually creating value? In the old world of enterprise software, a lot of what you were doing was competing within the IT budget of, "Hey, I have a certain tool. What's your spend on this tool? Where else are you spending?"
I think for us right now, we're also displacing the human labor side of it. So, I think that's a fundamental change for us in this new age, right? To say, "Hey, the ROI of our product is that you don't need as many human support agents."
Our product can go and do a lot of that. So it's not just about: where are you spending in your IT budget? But where are you spending on the human labor, which I think changes some of the TAM dynamics for some of these companies.
Katherine Zhang: So thinking about some of this… Bela, you mentioned that efficiency is maybe a little bit different in the sense that there's more focus on that. It's more table stakes now. So has AI changed the economics of growing a software company? And Kunal, we'll start with you since you are [bringing the perspective of] a CFO.
Kunal Agarwal: Yeah. I think it certainly has changed a lot of the conversations. After spending twenty years in enterprise software, and now working at an agentic product, some of the fundamental trade-offs that are different.
So in an enterprise company, your gross margin was kind of like an inherited thing that you had, right? You were thinking about a gross margin, and then your incremental software spend was kind of on top of that built-in gross margin. And I think right now the difference is gross margin is almost a product decision.
What I mean by that is: fundamentally, each incremental transaction that we do from a product has real costs associated with that -- real LLM costs. So one is the instrumentation of: we have to be able to have that visibility into what's happening in real time and hold different teams accountable for that.
But essentially, you have this differentiation now of: how much do we optimize for quality versus cost? And that's on like a per transaction conversation. This becomes a fundamental reshaping now of finance and product and engineering working together to figure out how we optimize cost and which things are best served by which models at which time. So I think gross margins are fundamentally different, and it's now very much of a product and finance question versus something that just got inherited with a set of things.
The second thing that's different for us in the age of AI is: an agentic product doesn't come out of the box fully ready to go like enterprise software would work. With traditional enterprise software, you sell a product, it either works out of the gate or you pay for an integration, then it works, and then it's steady state. I think with an agentic product, where you're seeing the dynamics of the P&L change a little bit, there needs to be a little bit of customization.
There needs to be some training of the model -- of the tool -- to be able to be operational. So now what you're saying is, okay, how can I develop the right implementation -- the right onboarding of the product -- to enable that product to work out of the gate? And what we've had to do is fundamentally rethink the organization around this.
I think the way that we now operate our org is fundamentally different from how we did it three or four years ago. And I think the difference is, honestly, most companies are using AI to make a bad org chart run faster. And I think my advice is that you've got to think about rethinking your entire org chart in the age of AI versus just continuing to just try to go a little bit faster with kind of an old setup.
Katherine Zhang: This reminds me of one of my favorite – I think from the '80s possibly -- business school articles. I think it’s called, “Obliterate, Don't Automate.” Don't just automate the bad org chart or AI the bad org chart. You have to rethink it.
Hank Chen: And I think the ones that figure out how to do that maybe it's not Rule of 40 anymore. I'll plug our other podcast, Dry Powder, with Hugh MacArthur. In a conversation he had there with Robert Smith, Robert talks about how maybe it's Rule of 70 now -- once you figure out what this new structure looks like. You're getting higher returns from agents and maybe that changes economics significantly in your favor.
Isabela Peralta: Yeah, I mean, I think a lot of the conversation around efficiency in this AI era has focused around the bottom line and from a cost perspective, how are you scaling? I think it's equally important to think about your top line.
I think that we are moving to a world where best in class in terms of top line momentum used to be you went from one to three to 10. Now it's one to 10 to 50. I think that the question that a lot of investors and operators should be asking, though, is about the durability of those revenue streams.
I think that with the AI hype, there's definitely a lot of value that's being created, but there's a lot of what I sometimes call AI tourism -- a lot of piloting going on with folks actually utilizing pretty deep pockets. I mean, to Kunal's point, you're now able to tap into different P&L line items to trial some of these products.
Maybe you're thinking about insourcing, maybe you’re thinking about different solutions. So when we start to look at that path around, hey, one to 10, I think it's less about what's the time to one to 10, but more the time to a durable 10, right? You could reach $500M ARR with five reps and still have a broken business if you're serving it at 45% gross margins and churning a bunch. So I think that going that step deeper into the quality of the revenue when you think about efficient growth is equally important.
Kunal Agarwal: Yeah, I think durability is a really, really important point. I think Bela has made a really important point. We see a lot of customers trialing a product, and they don't get enough value. And your ability to deliver time to value is so important with an agentic product because if you don't do it right away, the customer is not going to get value, and they will switch at the earliest opportunity. So I think it puts so much more pressure on upfront (the first 30, 60, 90 days) making sure there's value, making sure you're sitting with the customer, and making sure they're deploying it in the right way. Which is why, again, rethinking your post-sales organization is really important.
But I think measuring, you know, your GRR at renewal, how many people are actually renewing the product, then how many people actually expand. We used to think about GRR and NRR in the enterprise software world as these “givens” of certain ranges that you would have. In an AI world, those ranges have a large amount of variation now. And I think that speaks to the quality of the product and the stickiness of the product.
Katherine Zhang: I'm curious, then, because you've all talked about redesigning the orgs and where you can really change things to get more leverage -- do you have a favorite example or case from your experiences on this is a specific area where a company was able to do this?
And I'm asking this because one of the things that we found in surveying a set of companies recently is, much like Kunal said, gross margins are down if you are more AI heavy. The other thing that we found is with a lot of companies, their costs are still there. And the third part of this, and this is why I'm asking about specific examples, is… the number one question that leaders had for their investors and boards was: could we get more playbooks on how to use AI? So I'm curious if any of you have some specific examples of things you've seen companies do that have really properly brought leverage from AI.
Isabela Peralta: Yeah, I can take it, right? I think that at the beginning, when we were looking at the whole of AI transformation cases across our portfolio, step one was just: how do we get these tools in the hands of people on the ground? How do we get engineering to talk to product to talk to sales?
Immediately, I think the conversation shifted to more: how do you drive efficiency from a sales perspective, from a marketing perspective, from an R&D perspective, whether you're using X coding automation tools or Y. I actually think that the most interesting use cases are around those companies that have thought about AI transformation and efficiency in terms of opening up new markets -- so new TAMs -- that previously they might not have been, A, able to compete in or, B, serve at an attractive margin.
One of the companies that I work with – they are a specialty-specific EHR, very much focused on practice management, patient engagement, payments, etc. All pockets of the market which notoriously have high gross margins. They had never quite been able to tap into revenue cycle management given the high service intensity of this market.
When we started thinking about AI transformation then AI efficiency, there was this question around: if you look at the different subsegments of RCM, what are some of the buckets where you could potentially automate these service, high labor kind of components to actually deliver a product to your customers at an attractive gross margin.
All to say, I think that was a very interesting example because I'm trying to think about opportunities for our companies more from a top-line lens. I think that the OpEx lens will follow, and I think it's extremely important to account for, but when it comes to really capturing value of the customers and expanding your addressable dollar or size of the price for you to go out and get, thinking about AI transformation from a product and capability perspective and not a feature, but truly a product is what's been most needle-moving for some of our companies.
Kunal Agarwal: Yeah, I'd say for us, we have driven and adopted -- so my team is responsible for driving the AI transformation across our company internally, and I think what has enabled us is two things.
One is that we just don't need as many layers of management. Where people's jobs before were just to manage other people -- we don't need that. And so, I think it's just going to be leaner organizations and empowering people with more tools to be more effective versus just less people. You make your people in-house more effective. I think you naturally self-select out the people who are going to embrace the new way of working versus the ones who are stuck in the old way.
An example of this is and how we've enabled us to be more effective within our go-to-market organization has been: we have now created individual cockpits for each one of our reps -- each one of our customer success managers -- who every day wakes up and sees, "Here's all of my active accounts. Here's what needs nurturing; here’s what needs help."
We have built out churn analysis that can highlight customers who may be at risk of certain things. And so it just helps you scale and handle more accounts, right? Whereas before you were kind of limited to say, like, "Hey, I only have a certain amount of bandwidth to handle X number of accounts."
Now it can be, like, X times ten because now you're able to automate a bunch of the outreach. You can know which accounts need help versus not. And so, there's an example of the efficiency that you can gain internally if tooled the right way.
My biggest thing is that everything needs to be connected across the company, and so we've built our own internal context layer, so everybody has the same definitions and the same measurements. Being able to play a multiplayer game with AI internally is really important so all the knowledge is compounding internally versus everybody having their own Claude subscriptions and doing their own set of things. The way that the best companies are going to take advantage of this is like playing an online multiplayer game.
Hank Chen: I love that example. I mean, in some of the work that we're doing on the R&D side, the discussions we're having are around changing the software development lifecycle, and it's early days, but what does that look like? In the past, a scrum team was six to eight people, four to five developers, a couple QA folks, and maybe a scrum master.
That team is going to look completely different in the agentic era, right? Perhaps you really only need an architect, one developer working with a bunch of agents. And then the product management side is going to change as well. Perhaps you don't need a very large product management team if you have agents writing epics and sprints and detailing product requirements. Maybe the product management team changes significantly. So you're talking, you know, a smaller labor base, a lot more tokens, and that's sort of where the direction of travel is going on the SDLC side.
Katherine Zhang: Great. We've talked through a lot of examples. So as an operator and investor, then, where should we focus? Bela, I'll throw that one to you first.
Isabela Peralta: Building off of my last example, I think that we need to be very clear on what is AI that expands TAM versus AI that is a feature. To Kunal's point, again, AI that truly matters captures a new budget and grows the market, usually by taking work off a labor line rather than adding a nice-to-have to an existing software spend. That's one.
Second, not to be repetitive, but it's how are we treating ARR versus what should just be called an experimental AI budget? I think that investors and even operators need to be very clear on separating durable revenue from, again, this tourist revenue. And underwriting to gross retention, not net.
And on this point around net and usage models, I think a third thing that some of the investors are potentially getting wrong is chasing ARR per employee as a vanity metric, right? We hear a lot about these startups scaling from one to 100 with a very lean labor force.
I think that if I tell you, "Hey, we're generating half a million dollars per head," that sounds great until you learn that it's usage revenue that can just evaporate and you're somehow equating that to they've built an efficient distribution engine. So I think we need to be careful about not confusing lean burn with distribution or a moat.
My last point, which I think I hit on at the beginning, was this idea around confusing data volume with a data moat. I probably hear the words data moat five times a day, and more likely than not, the use cases that I hear about are not really moats. It is just volume. Again, a moat when it comes to data is when you have this compounding feedback loop where every customer can measure the improvement of the product as one additional customer joins the platform, and I think that is pretty hard to replicate and few -- very few -- companies can actually show the flywheel coming into effect.
Kunal Agarwal: I'd say as an operator, one of the areas that is really important for us to have a lot of organizational-wide visibility into is your unit costs. That never used to be a thing in enterprise software. Unit costs were just not something we ever really cared about because your incremental cost was close to zero for most things.
Like I said, for us, with LLM costs, there's a very real discrete cost per transaction that happens. So, the key is to measure it and then also optimize what that is. And so, it’s a new way of operating that hasn't traditionally been there -- especially for engineering teams and operations and product teams. You’re thinking about, “Okay, what model serves what type of question in the most efficient way? How do we best optimize customer quality, but also do it in the most efficient way possible?”
It’s just, I think, a new discipline that has emerged in probably the last eight to nine months around how do we properly track, measure, and use this to measure our own discernible costs.
Now, the second piece of this is the thing that everybody's been faced with in the last three months: your own internal LLM costs. The tools and costs that you provide to your employees. And as OpenAI and Anthropic have shifted to more of a usage-based model, this becomes really important for us to measure actual ROI, right?
And so, for engineers in the good old token-matching days everyone had unlimited budgets; you could go do whatever you want. But those costs become real significant really quickly. And how best are you intermediating those costs, how best are you driving efficiency, and what are those costs being sent for?
This is all a very evolving thing that we're working through right now and trying to figure out how we best apply that. I don't think there's one path that solves all of it, but it's a very real thing that I think everybody's trying to figure out real-time.
Hank Chen: One framework that we use in our work that's been helpful is just thinking about a two-by-two. Kunal, you said earlier one of those axes is: is AI displacing humans? And on the other axis, it’s: is AI displacing software workflows? And thinking about the confluence of those in each quadrant. If you're, displacing both workflows and people, then that spend is going to go away and that business is going to go away.
If it's the opposite: if it's low workflow displacement and low people displacement, you have a real moat and possibility of adding to it with AI. If it's one of the other quadrants, AI is going to be predominant in those situations.
But that’s been one helpful way for us to kind of think through this when we look at different businesses, you know, what's got a more of a moat and less of a moat. And I really like, Bela, the way you described the data moat.
Katherine Zhang: All right. We've got one more question for the panelists, but for those of you listening in: the Q&A is open. We have a couple questions already in there, and we will get to as many as we can in the time we have.
To wrap up this portion: are a lot of these things here to stay? Do you have predictions for how some are going to change? Are any of these things, you know, going to be gone by the end of the year? How do you think about whether this is the new normal?
Hank Chen: I think there’s going to be continued change, right? We're very clearly in a dynamic period. There are meta questions that nobody knows the answers to, like “When are we going to hit AGI? What’s going to happen if quantum computing also comes into play?” Because these different technology disruptions build on one another, it's kind of hard to envision all those things together.
I think investors, from what I see, are really struggling with: “what do I have on my hands? What are my portfolio companies? Where are each of them at? What level of risk exists across my portfolio?”
And at the end of the day, the question is: “how much time do I have?” Like, how much time do I have if I have an issue with a company? How do I know that now and what do I do with it? Do I try to sell the company quickly before everything collapses? Or if I have a winner, how do I know that it's a winner? How can I make sure the moat is really there? And what do I do to extend value and differentiation so that I'm relevant in the new era?
I love history, so I look at the different tech disruption ages -- everything from the Industrial Revolution to railroads through computing, cloud, and now agentic revolution. What's happening is the technology diffusion cycle is happening faster. If you look at the delta in how long those cycles take shape, the agentic revolution -- by far -- is the fastest. There was an article in The Economist recently showing the capex spend relative to other revolutions.
But also the problem with that is the technology diffusion cycle of when AI becomes real, and what it's going to do to the incumbents is shrinking. And so, if you're owning one of those incumbents, you sort of don't know like, gosh, am I in trouble? And how much time do I have to go and do something about it right now?" And that's what everyone's scrambling to do.
Kunal Agarwal: I think from our side, what's real is: AI's real, and it's changing the way we work. There's no static state, so things will be different tomorrow than they will be today. And so I think what matters most for us is speed to execution -- cutting down the cycle time that the entire organization runs. And that's not just product and R&D, right? Building twelve-month roadmaps just doesn't make sense anymore. We’re now building quarterly roadmaps because who knows what's going to happen four months from now. So we have to move faster than ever.
The cycle time from customer feedback to product to execution to feature -- that needs to be cut down dramatically and move faster and more of like an AI software development lifecycle.
But also the rest of the teams around us. Our biggest constraint right now is not capital, it's speed. I think the markets are forming around us, like behaviors are catching up to market expectations. So you want to be best positioned to capture that market, because I think that's going to be the biggest thing in the next twelve to eighteen months. I think time will weed out a lot of people who are actually category winners versus followers, and those followers are going to have a tough time.
Isabela Peralta: Yeah, I agree with everything that has been said. I think it would be silly of anyone to think that change is going to stop in any way. I think change is the new normal, and that's why when we look at our portfolio companies -- and specifically talk to our management teams -- the constant feedback is this constant disruption from within.
How do you really change the mindset of the executive team and the CEO when it comes to your ability to move and respond to the market in a more agile manner? I think it's something that has changed, and it's something that people can no longer ignore -- this idea around not prioritizing customer and customer love above all. I think that AI has changed the bar around user satisfaction, and ultimately what can make an AI native beat an AI incumbent is if that user experience is 10 times better and allows for 10 times more efficiency.
I go back to healthcare. I remember when I looked at some of the first EHRs in the space, an NPS of zero was a good thing. People just had the expectation that your EHR inherently was bad, and I think that that's no longer the case and no longer something that customers are willing to accept. So it truly is about putting the customer and the user experience at the forefront. And then this constant disruption from within -- from these leadership teams -- really internalizing an AI mindset of how they move about the organization to be able to adopt and react to the market in the most efficient manner.
Hank Chen: I really like both of those points. Kunal, I was going to ask you – Bela, given your point about customer demands -- the customer has to be delighted in the agentic world. I think immediately. Even faster than they were in the SaaS world, right? Given everything and all the hype we've heard about AI, if you're not delighted in that first instant, you're toast.
So I’m really curious -- I think you were saying something like that before about how you were going to market with your products?
Kunal Agarwal: Yeah. It's a good point, Hank. I think that it has changed the way we think about time to value and delivering value. Like again, I think the hard and the nuanced part with AI agents is that it requires customer education. It’s not out of the box at 100% of capability. You have to integrate it to the workflows. You have to train it. You have to onboard it.
And so, what we've been trying to do is build agents for our agents to make the agent get onboarded with less user help and involvement. But it's not going to ever be 100%, right? You're always going to have some human intervention.
The key thing is: how do you compress down the time as much as possible to start delivering value? To start delivering those “aha” moments? And that’s the tough world in an agentic place. A lot of that comes to product architecture conversations of: how do you develop the product? How do you think about the product knowing that you need to be able to get people to see value as quickly as possible? Because I think we've all chatted about the tourist nature of some of these AI products, and you don't have a lot of time to get people to feel like, "Yeah, this product works. It does what I thought it was going to do." So I think it's changed the way that we think about resourcing and developing the product.
Katherine Zhang: Maybe a good segue from that to one of the questions that came up in the chat here: Kunal, specifically directed at -- you mentioned a new metric, this LLM cost per transaction, and you were talking about time to value here. Are there any other new metrics that companies need to be measuring in the next six to nine months?
Kunal Agarwal: I think there's a lot of metrics that exist, but you look at them in different ways. I don't think it's necessarily a creation of net new metrics, but I think looking at your cohort analysis is going to be really, really important.
To look at like, say, retention rates by certain cohorts. As you're releasing new products, are you seeing a change in the retention rates? If not, why? Being able to go under the covers to really understand quickly the time variance between what you're releasing and the impact that it has.
For us, churn is one of the most important things to look at, but churn as a broad metric is not super useful. You have to look at the details behind churn in different customer cohorts... different products that you're using. Because to me, ultimately, we'll be successful if we have the best product.
Go to market is a thing that becomes a lot easier when you have the best product in the market and you're delivering value. So the ability to get really granular as the CFO and chief accountability officer for the rest of the team on like: is our product delivering value? Are we doing it quick enough? Those are the things that are really important.
Katherine Zhang: Thanks for that. You brought up something related to another question that we see in the Q&A, which is for all panelists: have you seen examples of AI driving increased churn or increased organic recurring revenue in mission critical B2B software? So the specific question says: not data-only businesses, but software that manages workflows.
Hank Chen: I think we're not going to see it just yet. I talked a little bit about the technology diffusion cycle. So, there's going to be a little bit of a buffer time, and this is why I think it's really tricky. There's a little bit of time where it seems like everything's going to be okay, and then maybe it's not.
I think we're still in that early part where businesses that are likely going to be disrupted haven’t quite seen that gross churn in as stark of a way as we would think it would automatically fall off a cliff. I don't think it's going to do that. Think about some of the last disruptions, right?
Think about newspaper advertisements with Craigslist -- it probably took a while before that died off. It's going to be shorter here, but I think it's still going to take a little bit of time to see that come through.
Katherine Zhang: Okay, and then let's look at this next question -- again, related to disruption. How do you think about segments that were historically attractive for private markets, especially PE? So vertical markets like legal. And these are markets that are now squarely in the crosshairs of some of the frontier models -- they have domain expertise but are rapidly coming under fire. How should we think about software serving those types of vertical industries like medicine or specialty engineering?
Isabela Peralta: Yeah, I mean, I think legal tech in itself is a little bit specific because it got exposed to a lot of what was language work on top of documents, and that's exactly what the frontier models will do natively. So, I think that when you start looking at other verticals, it's more around what's left when the model can already do the specific language task, right?
Again, coming back to medicine or everything specialty engineering here is a good example. I’d look to underwrite it to: what is hard for the models to get access to? One could be proprietary or permission data that they can't access. Second is how to think about liability and regulatory accountability -- someone has to kind of own the outcome and a general model won't.
And I think the third one -- definitely not the least important -- is just how to think about that last mile integration into the system of record. Or where there'd be the physical or the clinical workflows.
So verticals where the value was in text – as in interpreting text -- I think are in trouble. But verticals where the value is around the regulated data, around accountability, around workflow embeddedness -- I think hold up. And that's probably healthcare, for example -- where I see it playing out and one of the reasons why I'm so bullish around the space.
Katherine Zhang: All right, and let’s end on a more high-level question. We’ll combine two of the high-level questions here. Where are companies most likely to overestimate the value of AI today? And what parts of building and scaling a company will still depend most on human judgment?
Kunal Agarwal: I'd say, in my view, I think people are wildly overestimating the efficiencies and number of heads they’re going to get from AI. Like, yes, there will be some efficiency here and there, but you still need people to go out and sell. I think you can make those reps more efficient, but it doesn't mean, for instance, I need half the sales reps I needed before. So I think we're kind of overestimating a little bit that every company can operate with half the headcount. I don't think that's completely reasonable for most companies.
To your second part of the question, I think the things that we will continue to see a lot of investment in is the ability and education for people internally on how to operate these things to get more and more efficient and ideate. And the judgment and trust -- it applies to people in every single function, right?
A lot of the time you spent on creating analysis or creating work -- that primary document creation stuff is now getting largely automated. But your ability to now change your lens to: what makes sense? How do I distill this? How do I tell the story more effectively internally?
How do I spend more time influencing? That trust, judgment, taste -- that is going to be highly, highly important.
Even the most junior folks in our company aren't spending time building models anymore. They're spending time simplifying it. How do I tell the story of what matters, what it means, and how do I communicate this effectively? So I think the skillsets are changing a little bit.
Isabela Peralta: From my end, in terms of what is being overvalued today, I think that there's still this question around when efficiency is generated in a specific space, to what extent does a company have the right to retain some of the savings and the value that's being created to the customer in terms of pricing power? And to what extent does price actually get commoditized and the savings ultimately accrue to the end customer?
Being a little bit more specific, for example, I've spent a lot of time looking at scribes -- like documentation scribes in healthcare. And there's this question around: look, today there's immense value that's being created because you're reducing the time it takes for a doctor to create a note from an hour to minutes -- single digit minutes, which actually translates into a lot of pricing power for the scribe at that moment in time because the ROI is so clear.
But as we start seeing the scribe space become much more competitive – intensely competitive -- how do we think about price compression? How do we think about the customer saying, "Wait, I need to capture more of the savings that's being generated," because the capabilities between some of these scribes might not be as different.
So I think that we are going to start seeing a little bit more price compression across some specific sub-segments that today is probably, quote unquote, "inflating," some of the margins for these companies.
The second one on what I think still depends most on human judgment is any part that is relational in nature. It's -- whether you have an investor or you're speaking to the customer -- that storytelling capability. Or when you're looking at judgment around what matters to a specific person in a specific moment for a specific market. Internally, as well, how you think about people management and building culture.
I think that building a great product is just the first step, but then being able to actually manage your company, your employees, and communicating to that customer. Again, going back to this idea around obsession of customer love, how you keep that relationship will be more important than ever.
Hank Chen: I'll add to the second question. On human judgment, I think a lot of this comes back to every management team needing to think hard about: what is our business exactly?
I'll just take a silly example. You know, the advent of cars and the buggy whip. Are we in the buggy whip business, or are we in the transportation accessories business? How you define that has great implications on whether your business is going to survive or not.
I wake up every day to LinkedIn messages saying the consulting industry is dead. That's because there's a misunderstanding of what our business is. Our business is not building PowerPoint slides and doing research memos. It is giving advice to executives and investors, and I think that's always going to exist. So I think it's really important to understand: what is the fundamental driver of our business? Where is the value? And really think hard about that.
Katherine Zhang: I think that is a wonderful note to wrap up on. So thank you everyone for attending. We'll be sending out a recording, and a transcript if there's anything in this discussion that you want to revisit. And of course, if today's discussion has gotten you interested in benchmarking your company's performance against peers, we at OPEXEngine would be glad to continue that conversation with you. So again, thank you so much for attending, and a big, thank you to our wonderful, insightful panelists.
