The Science of Scaling in Agent-Native Markets

Julia Nimchinski:

And we’re about to transition to our next fireside chat. in about a minute, with Mark Roberge. Live TV… Hang on for a minute.

Mark Roberge:

Hey, Julia.

Julia Nimchinski:

Welcome back!

Mark Roberge:

Yeah, thank you. Are we live, or are we recording? We’re live.

Julia Nimchinski:

Oh my god.

Mark Roberge:

Gosh, cool.

Julia Nimchinski:

Welcome back, Mark Roberge.

Mark Roberge:

Thank you.

Julia Nimchinski:

No introduction. Amazing seeing you. For everyone watching, Mark is the co-founder of Stage 2 Capital, senior lecturer at Harvard Business School, former CRO at HubSpot, and one of the most influential visionaries in modern GTM. And Mark, fantastic having you back. Congratulations with your new book. We will be sending it to all and partners.

Yeah, what’s the latest.

Mark Roberge:

Thank you, Julia. Well, you know… Yeah, it’s been great out there talking about it and integrating it with this AI era. I do think it’s an important time, as some of your questions have alluded, to retract back to first principles, so we’ll be talking a little bit about that today and where we are at. And again, thank you for the support on it.

It 100% goes to, support mental health, which we can speak about a little bit at the end on why. And I think it also is, an important piece to our AI journey as a society.

Julia Nimchinski:

Super excited to dive in. So, last time we spoke, you introduced, your four phases of how AI will reshape GTM. Yeah. And it was about a year ago, and you mentioned that we barely entered as a community in B2B Phase 1. I’m curious, a year later, where are we in those four.

Mark Roberge:

Yes. Making a little bit of progress, but not as much as I would have hoped. I think we’re still in phase one on the go-to-market front, and Phase 1 is really… It’s really about maximizing… productivity through selling time. I think today’s AI is very capable of exponentially increasing selling time.

And so let me, let me kind of walk you, Julie, and the audience through Some of this… narrative… points back to, like, a lot of board meetings where the teams are claiming that they’re super AI-native, and I’m always like. how do you know in sales?

And so, I think we have a massive advantage over our engineering peers, where productivity and engineering is more difficult to measure, but productivity and sales is easier. So I feel like the end state, you know, long-term measurement is Some combination of productivity per rep. And perhaps the percentage of your team that’s making goal.

You can do both very easily. the… If we just focus on productivity per rep for a second. algebraically extracting that back one layer, I find the revenue velocity formula to be very, useful.

And revenue velocity formulas are an awesome first principles formula that isn’t talked about as much in In the revenue teams, and it’s essentially… The number of active opportunities that a rep or the entire team is working at any given time. times the ACV, Times the close rate, divided by the sales cycle.

Alright, so if your rep is working 20 opportunities at any given time. Has a 10% close rate, so 2 customers. with 100,000 ACV, so 200,000 total revenue. and a two-quarter sales cycle divided by 2, that means they’re producing $100,000 a quarter. That’s the… that’s the formula.

And so, when you look at it from that lens, you have to ask yourself, well, which of those four variables is easy to exponentially increase, to exponentially increase PPR? And, I think the number of opportunities is. And just let’s walk that through. What if we tried to double ACV? I’d be concerned about that.

Because that essentially means you’re doubling the price for the market, and that could move you into a totally different market than you intend to be in, okay? What if we try to double… The close rate and the sales cycle.

Well, those two things are very dependent on buyer action as well, and that’s something we’re talking about here in the agent-to-agent. Where, like, buyers just… they haven’t adapted their buying process with AI as fast as sales has adapted their selling process, so it’s a little bit of a bottleneck for many teams. So those are harder to 2X.

But the number of opportunities that a rep works at any given time with the same adequacy and quality is highly in the control. It’s almost 100% in the control of the vendor.

So that’s our opportunity, and if we extract that back one more level, a great way to do that is selling time, the percentage of time that our sellers are with buyers or prospects. Historically, best in class has been, like, 30%. with today’s AI, it could be 60 or 70%. So, that’s kind of where we are right now.

I think we’ve got a pretty deep understanding of the opportunity, and if I translate that whole narrative to actions that you could do this afternoon, number one, start measuring your selling time. You can do that with AI.

And then number two, just like any good product manager, come up with all the AI use cases that you can put in place prep for meetings, set meetings, post-mortem, sales room auto-creation, serum updating, whatever. There’s a long list.

and start tackling those with the intent to increase selling time, with the intent to increase the number of opportunities, and with the intent to increase the productivity per rep. That’s where we are. We haven’t seen a lot of movement yet on it.

Julia Nimchinski:

Love it, still. moving from this practicality, Mark, and, into your strong conviction, you mentioned that despite all the possibilities, and obviously here we’re going a little futuristic, A to A, B2B, we had Brian Solis keynoting all topics about it today. Yes.

Mark Roberge:

Awesome.

Julia Nimchinski:

But I’m curious, you mentioned that it could be this era, this two years, that we’ll see the highest failure rates. startup of, like, the AI-native land. Can we look at it through the science of scaling methodology, and is that connected at all to LAR?

Mark Roberge:

Totally. So if we just look at the basic, the highest level framework of the science of scaling, to help you understand when you’re ready to scale and how fast. Step 1 is find product-market fit, step two is find go-to-market fit, and step three is exploit growth and melt.

And I would say, like, The two areas that… the current crop of AI-native startups are very vulnerable to is, first and foremost, is a durable moat. I think there’s just a lot of question marks about how they will defend against Claude, how they will defend against a copy… a lower-priced copycat attacker.

I think a lot of founders say that their moat is their speed of execution, and… I don’t know. I just don’t… Really get why they think that they can… be the fastest company in the world. Especially when, like, 25 of them tell me the same thing, that they’re the fastest, so… I don’t know. Like, that’s just… that’s just tough.

So I’m worried about moat, and then the second one, to your point, is I’m worried about, retention in the long term and the LIR, which are related. When you don’t have a moat, you don’t have LTV.

And we’ve already seen, like… I mean, just like in the highest level examples, like… everyone was on Cursor, and then they diversified a lot to Cloud Code, and then Codex came out, and there’s a bunch of jumping over there. I mean, these are very fragile deployments, and that’s just in the coding arena.

So, yeah, Moat is the biggest risk, and then, LIR is the second.

Julia Nimchinski:

bit of execution connects well with the product market as a feeling, to your point. And, we’re seeing a lot of proliferation, obviously, of all sorts of products and platforms, and lots of founders I know are building, shipping, like, crazy, 50 products, platforms, and obviously distribution. is the problem.

But I’m curious your lens on the resource allocation. Do you think that the traditional model, 50% engineering product. 30% sales, marketing could be the problem and the limiting factor to profitability.

Mark Roberge:

Definitely. You know, and I wrote about that, like, a couple weeks ago. The… you know, getting the book out in February has been exhausting, but, you know, obviously worthwhile, and… I was on a massive roadshow, across every major city.

I had just written a couple weeks ago about having been in Silicon Valley for 3 days and had shook hands with 179 founders and executives, everything from one-on-one coffees to intimate 12-person demo, dinners with, like, some really cool C-suite individuals.

And all the way down to, like, seed-funded founders in a room talking about the science of scaling. And that was definitely one of the patterns that was extremely apparent is… we’ve never seen this velocity of product development, so certainly a lot of the CROs were struggling on the enablement side.

And… We are starting to see some velocity on sales. But, like, Buying patterns and buying processes are nowhere close. And so, that is the clear bottleneck to the growth aspirations of these vendors, in terms of, like, them… implementing the software and helping the customer see value? Is the customer is the holdup?

And when you start thinking about that. Where everything else is 3 or 4X, but the customer adoption has not. Should we be rethinking our capital and resource deployment strategy? Whereas, like, generically speaking, you know, make it up, we’re like, you know, 40% in R&D, You know, like, 25% in sales, 10% in marketing.

10% in customer success and adoption. Is that, like, wrong? Should customer success and adoption be 30%? Because, like, that’s the bottleneck. Should we have way more FDEs? Should we have more engineers working on frictionless adoption? Should we be building agents to support that? I don’t know. A lot of the evidence points in that direction.

Julia Nimchinski:

In your methodology, the three fundamental constructs, and obviously there are more, but, ICP seems critical, LIR, and PMF. I’m curious, where do you see the place for agent-to-agent, and do you see it at all?

Mark Roberge:

Oh, it’s enormous for agent to agent. These are first principles, right? That’s why… partially why I wrote the book is… We’re… this is such a massive shift in… how we build product and sell product and run an organization.

And unfortunately, in the first phase of doing so, we have somewhat iterative results because we are looking at how it was done in the last 5 years and replicate it, which leads to an iterative result, as opposed to, like.

asking ourselves why we do it that way, and stripping it down to its studs, and then rethinking it within the capabilities of the technology. You know, a perfect example is in the web, like, if you were around in the early years of the internet, everybody thought the extent of the internet was putting your company brochure online.

And that was, like, the absolute tipping, like… Tip of the opportunity. And that’s kind of where we are now, is like. you know, we’re, like, running these ICP agents with, like, looking for, like, intense signals in social media, which was, like.

Just an iterative… like, build off of, like, how it was done in 2019, as opposed to ripping it down to its studs. And so… This can get complicated fast, but… Let’s just try to talk about ICP in a first principles manner.

And first off, just, like, the rooting of… If I was to develop a deterministic model to assess whether a particular customer or customer segment is in ICP or not. One of the first principles, mistakes that a lot of companies do is they try to build that model to minimize CAC. By maximizing close rate, And minimizing sales cycle.

And unfortunately, they leave LTV and retention out of the equation, and that’s very dangerous, because… I’ve never seen, or rarely seen, a company Where the customer segment that has… the highest close rate and the lowest CAC?

had the highest LTV, And so if you… if you run your ICP model in this way, you’re gonna end up selling a lot of, like, low LTV, high-churn customers. And so, like, to, like, build your model correctly, you have to account for both LTV and retention. As well as low CAC.

Like, let’s find good segments where we can generate a lot of lifetime value profits, but also serve them and sell them in a way so the profit equation makes sense. Okay, so that… that’s some rooting that I think a lot of the industry is messing up right now.

And historically, we have done that by hypothesizing on certain revenue sizes, certain industry sizes, certain geographies that will help us frame an ICP. So we sell to mid-market companies based in the U.S, that are in healthcare. That is so limiting in the world of AI. Like, you don’t… you don’t have to define it with any number, 100 dimensions.

Let AI do what it does well. Which is… There’s, you know, whatever, 50 billion companies out there in the world? I just wanna… and I have sales capacity to sell… 15 new companies to start a sales process with 15 new companies this week.

tell me which 15 I should go after, knowing that at the end of the day, I’m trying to maximize lifetime value and retention at a reasonable CAC And, you know, to make the business work.

Julia Nimchinski:

And…

Mark Roberge:

And agents can just do that. Yeah. So, so, like, that’s, like… so, and then, like, if we imagine this agent-to-agent world that you’re, like, pushing the, you know, frontier on, Julia. Yeah, I mean, that’s what it would do, is an agent would just go out and evaluate 17,000 possible people to call. And figure out the next one that I should go after.

And in, like, a… In an even more Star Trek version. You could imagine a point where… the… The same human develops the product and distributes the product.

And in a way, ICP connects those, because one of the intentions of ICP is At some point, our product managers interviewed a bunch of potential customers and used those interviews to build a product for them. And we better make sure that we’re selling to the same people that they interviewed. to make sure that’s connected.

And in a Star Trek version of this AI era. You could imagine that AI is so good that the same person builds and distributes.

Julia Nimchinski:

So, you see an agent becoming an ICP, correct me if I’m wrong.

Mark Roberge:

Oh, yeah, and then in that world, like… Yeah, I mean, when you’re talking about selling to agents, and I had kind of, like, a follow-up to you on that. Like, let’s just imagine an agent selling to… an agent seller selling to an agent buyer, and do some of these principles apply? Yeah, they all do.

Like, the whole premise is… as a company, product-market fit, you need to continually assess whether your product is generating the value For your end customer that you promised them. And that’s measured in the long term by retention, and the short term by the lead indicator attention. If you have a buyer agent.

Buying from a seller agent, that doesn’t change it. You still need to make sure that it… whatever the seller agent told to the buyer agent, that that value is going to take place. You still need product market fit. And then in step two, go-to-market fit.

Go-to-market fit means that You can… Acquire and acquire those customers and deliver the service profitably, with good union economics. If we’re operating in a world where Seller agents, sell to buyer agents. That still needs to make sense. The concept of profitability doesn’t go away in that world.

Like, the entire input to the formula changes, like, it’s more compute-driven, but that still needs to make sense. It’s probably largely going to be driven by your price and the cost of compute.

Right, so, like… And then just, like, making sure that, like, if you’re… if you have enough compute to go after a thousand new accounts this week through… with your seller agents. you have to pick those thousand accounts based on the strength of your ICP, which an agent is calculating. So, yeah, it’s hard for me to see.

I mean, please push back, Julia, because you’ve been with all these different folks today, like, do you see it a different way? Do you see how these principles may no longer be applicable?

Julia Nimchinski:

No, I’m 100% aligned, and yeah. Okay, cool. I’m…

Mark Roberge:

Because no offense, if you don’t, you know, I just, obviously, we’re learning together, and I love to learn from you and the thoughts you see, but cool. Good to hear you on that one.

Julia Nimchinski:

I’m also curious to hear your thoughts on the… Finally, we’re moving from, you know, this insane era of sales specialization, all of the micro-functions in every, every functional segment. To the CROs listening, VPs of sales, and leadership, how would that affect them? We just had a panel, which you really embody.

back, you know, when HubSpot essentially been the CRA and architecting the whole model, but it’s rare that that would be the responsibility of a CRO. Do you see it more common now?

Mark Roberge:

Probably. I think we’re in the early phases, so just to, like. give a little more substance to Julia’s question here. Obviously in the last 20 years, we have been on a massive specialization journey. In go-to-market and in R&D, but let’s just talk about go-to-market.

We went from one role, account executive, that picks up the yellow pages, books a meeting, sells the meeting, and services the customer. to a bunch of roles, SDR to AE to AM to CSM, etc. And then the introduction and growth of RevOps. And… and the introduction of this Chief Revenue Officer role.

Which, like, I suppose I was one of the individuals that… represented a much more data-driven, systems-thinking approach to that role that was largely just, like, managing people. And… when I started writing about that, like, 10, 15 years ago, the ecosystem went too far.

I saw a lot of people hire what was really a great RevOps person and put them in charge of sales and CRO, and that didn’t work. Because at that point, 10 or 15 years ago. the hiring and management of people was still the most important thing. If I had to choose between a team that had A-plus human management. And enablement skills.

and C++ RevOps systems data skills. Versus vice versa, I would prefer the human-driven. And that’s likely to change in the next 5 years or so, if not already.

Where if I had to choose, I’d probably choose the A-plus systems and data and process skill with a C-plus human enablement skill, because the model will be so good that it can make C-plus people perform well. and outperform EA-plus people that are not tech-enabled.

And so, like, That’s the direction we’re going in, and there’s a lot of arguments that we will… Go away from the specialization, because specialization comes with a cost, the advantage of having SDR, AEAM Is you can take a human and deploy them on the role that makes most sense for their abilities.

And also, if you found arguably the hardest talent to find is great discovery and champion development and closing, I don’t want to waste that on appointment setting, or… Or, like, renewals. That was the hypothesis of specialization. But it comes with a cost. Starting with, like, a really bad buyer behavior, experience, where they’re handed off.

And also, like, a lot of management… requirements, because people tend to solve for their function. Like, SDR sent meetings, regardless of quality. reps close customers regardless of quality, and CSMs have to clean up the mess and try to retain them.

But if you set your own meeting and close your own customer that you then have to service, you’re not gonna set bad meetings and you’re not gonna close bad customers. And so that whole thing, that whole specialization. really caused the need for the CRO role. to, like, manage all those boundaries.

But with, like, AI, we could go back to a full-cycle AE role, and I think we should, and that completely changes the CRO role. Where it’s much more like the architect that is really thoughtful and deep in expertise at enabling that person with exceptional Process, science, tech, and data.

Julia Nimchinski:

And speaking of the future, Mark, do you see it more full-cycle function, full-cycle marketing for full-cycle sales, metaphorically speaking, or would it be more cross-functional? You mentioned that finance could be merged, sales, or product design.

Mark Roberge:

Yeah, I think it could get there. you know, I think these are sequential.

I think the first phase that we’re starting to see some signs are… are… is, is just, like, the functional side of go-to-market, the functional side of R&D, the functional side of finance, which still has these major departments With the, the boundaries, which is inefficient, because, like, finance always tries to get closer to sales, product always tries to get closer to support.

And vice versa, and we’ve built these org structures because of limitations of humans. Like, we don’t have people study finance and then go code, and vice versa. And I think that the next step could be this… GM role that builds and distributes and counts the beans financially as one role, if they’re completely AI-enabled.

I mean, that’s very Star Trek-y, we don’t have the tech yet, but you pointed out some companies, I think, like Block, who are already doing some of this, and… I think that’s where we’re headed.

Julia Nimchinski:

Very cool.

Mark Roberge:

So it’s, like, basically, just, like, to, you know, to pinpoint that, Julia, like. You’re more aligned with a customer type than a function. So, like, you might be in charge of, like, the dermatology segment. And you’re there to tweak the product and distribute and count that, you know, as a GM, is where we would possibly head.

Julia Nimchinski:

And speaking of that, Mark, you mentioned that one of the biggest innovators still was for incumbents, and obviously the opportunity for startups would be the organizational design itself. Yeah. Curious, speaking of Stage 2, What are you seeing in your portfolio companies and the companies you’re advising?

there’s block with their, you know, world model, and the DRIs, and something you also evangelized a little bit back in the HubSpot days, but Here is your lens on this.

Mark Roberge:

Yeah, I mean, I think that’s a big opportunity, is… The… it’s… I think a lot of people, including me, have a lot of conviction that The org… the optimal org design 5 years from now will look very different than what it looked like 5 years ago.

And part of that will be… A different hiring criteria and skill requirement for the different folks involved, whether they’re engineers or salespeople. It’s gonna be a lot harder for a 10,000 employee business that was started in 2005 to move to that. Than a 10-person business that was seed-funded last year. And so we have to exploit that.

We’re seeing all the things that we talked about today, but I think operationally, one added point is because we’re still in the transition moment.

you almost run, like, two parallel processes, where it’s like, okay, we have this company and product, and we found product market fit and go-to-market fit, and if we were scaling this in 2019, we would add account executives and SDRs and AMs and scale it, and we’re gonna do that. But in parallel, we’re going to attempt the new model.

by hiring two full-cycle AEs. And see how it does. And keep them separate. And our hope is that the right side here actually outperforms, but we’re just not sure. It’s gonna take some experimentation.

So that’s kind of an operational muscle that we hadn’t talked about yet that I… we like to run in parallel, so that we don’t wait to scale to, like, try to catch up with this thing, but at the same time, we continue to experiment with what we believe is the future.

Julia Nimchinski:

Love it. And last question, speaking of parallels, what would be your advice to, you know, mid-market enterprise companies and their leadership? How would you combine this, you know, future, agent to agent.

Mark Roberge:

Yeah.

Julia Nimchinski:

Where everything’s heading in the today and the possibilities.

Mark Roberge:

It’s really hard, you have to disrupt yourself, which probably means, like, some really tough quarters where you have bad revenue numbers, and you have to set very realistic expectations with investors and your employees that you’re playing the long game. That’s very difficult.

I kinda hope we see more companies go private, cause some of them are a really good buy right now, and that would help. And when… with regard to disrupting yourself, I would probably double or triple, and some of them are doing this, their corp dev arms, because you are gonna need a new set of talent.

And the ground-up companies are gonna be able to get there faster, so you need to stay close to them and probably acquire them, both for their talent and the footprint of tech that they’ve built.

Julia Nimchinski:

Mark, phenomenal featuring you, as always. What’s the best way for the community to support you?

Mark Roberge:

Yeah, I’m most active on LinkedIn, and in terms of my own disruption, I’m… trying to build out an abysmal TikTok existence, so if you want to humor me over there, it’s probably easier to get in touch with me. But a lot of my, you know, content I share on LinkedIn.

Julia Nimchinski:

Love it. Thank you again.

Mark Roberge:

Okay, thanks, Julia. Good luck.