Topline · 8 Feb 2026 · From the week of 2 February
Mark Roberge (Ex-HubSpot CRO): "AI Startups Will See the Highest Failure Rate in History"
These are notes on the conversation, checked against its transcript. The episode itself has the full discussion.
In brief
Mark Roberge, former HubSpot CRO, author of The Science of Scaling and co-founder of Stage 2 Capital, joins Topline hosts Sam Jacobs, Austin and A.J. Bruner to discuss scaling, go-to-market structure and AI. The conversation covers AI-native startup failure rates, founder turnover, how founders time revenue scaling, defining ICP by lifetime value, retention-based product-market fit, and how AI could reshape sales roles. Mark's central argument is that companies should scale revenue when they are ready rather than when venture money arrives, and that AI should move reps toward selling time and full-cycle rainmaker roles.
For founders
- Mark's sequence for scaling is product-market fit first, then go-to-market fit, then growth and moat, and he says the timing should follow readiness rather than when venture money arrives.
- Mark says ICP should follow where lifetime value is highest, not where inbound demand or CAC looks best.
- Mark recommends putting the share of customers hitting a leading indicator of retention on the first slide of every seed-stage board deck.
- Mark says product-market fit should be measured continuously, because he has never seen a business scale without it coming under risk.
- Mark expects a very high failure rate for the last two years of AI-native startups, though he is less certain the cohort's index return will be strong.
For revenue leaders
- Mark's first AI milestone is moving rep selling time from about 25% to 80% of the week by removing admin work.
- Mark's second milestone is collapsing SDR, AE, AM and CS into one role that owns the full cycle.
- Mark says to carve out a role when the first deal captures roughly 90% of a sale's lifetime value potential. If it doesn't, he says to keep the AE involved when the rest is complex to capture and to carve it out when it is not.
- Mark recommends running competitive experimental teams beside the core sales org, with the core serving as a control group.
- One of the hosts argues revenue leaders should hunt for rainmakers rather than design around B players, even though rainmakers are hard to find.
What was said 19, most useful first
ICP should be defined by where lifetime value is highest, not by where inbound demand is strongest or where CAC is lowest. Listen
Mark said that the two common answers, where the most inbound demand is and where CAC is lowest, are both brutally wrong. He said ICP should be correlated with the highest LTV, meaning the customers who succeed, want to retain and buy more. He said the business should then work backward to acquire those customers profitably.
“It should be correlated to where you have the highest LTV.”
Listen to the episode Strategy & market Link to this Report a problem
The first slide of every seed-stage board deck should show what share of customers are hitting a leading indicator of retention. Listen
Mark said Stage 2 has rolled this out with every portfolio company. He said he cares less about how much was sold and more about what percentage of customers hit the LIR. He said he reduces the LIR to three variables: what percentage of customers do an event every time period.
“I personally think it should be the first slide in every board deck at a seed -funded business.”
Listen to the episode Metrics & finance Link to this Report a problem
The first milestone for AI in go-to-market is raising rep selling time from about 25% to 80%. Listen
Mark defined selling time as the share of a rep's week spent face to face or on Zoom with a customer or prospect. He said this is possible today but not easy. He said that if admin work is removed while skills, demand and product-market fit stay the same, rep productivity could roughly triple.
“what is accessible to unlock massive efficiency improvements in every go to market org is first milestone drive selling time from 25 % to 80%”
Listen to the episode Sales process & deals Link to this Report a problem
Mark's test for carving out a role is whether the first deal captures most of a sale's lifetime value potential. Listen
Mark said if about 90% of lifetime value is captured in the first deal, the post-sale work should be carved out so AEs are not spent on it. If not, he said the next question is how complex it is to capture the rest, such as needing to reach Europe, Asia or other departments. He noted the trade-off that a rep who owns the customer after the sale does not sell poor-fit customers.
“What percent of the overall lifetime value potential of a sale is captured in the first deal? If it's 90%, then carve it out.”
Listen to the episode Sales process & deals Link to this Report a problem
Mark recommends setting up competitive experimental teams alongside the core sales organization to try to outperform it. Listen
Mark said that when he was a senior advisor at BCG, he helped companies with ten-billion-dollar on-prem businesses move into SaaS. He said the core should keep running while the company finds a sandbox, such as an open territory or a new product, and stands up an inside sales team with MQLs. He said the core then serves as a control group to measure the experiment against.
“I think that's going to be the key is to have these competitive teams to experiment and try to outperform the core.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Strong product execution can let companies run inefficient, order-taking sales teams for a time, but that advantage runs out. Listen
Mark pointed to the Google era, when the hiring profile was Ivy League with no sales experience, and said it was replicated at Slack and, to some degree, at OpenAI today. He said OpenAI is struggling a little in the enterprise because it never really fixed this. He said the product-led success covered the inefficiency until it ran out.
“you could get away with an inefficient sales org and people who didn't really know the basics of selling that looked really good”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Many founders time their push to scale revenue around when they raised venture money rather than whether they are ready. Listen
Mark said that after about eight years of post-HubSpot angel and VC investing and board work, this is the most common answer he hears when he asks founders how they decided when and how fast to scale revenue. He said that when asked whether they assessed readiness, founders often say no and that they had the money. He described a plan built in Excel around a triple, triple, double, double growth pattern.
“We chose the timing on when we were at scale revenue based on when we got money from a VC.”
Listen to the episode Strategy & market Link to this Report a problem
Even HubSpot, a master of inbound, had about 30% of its inbound leads within its ICP. Listen
Mark said HubSpot had a very broad market, and about 30% of its inbound leads were ICP. He said he has never seen a business where more than 50% of inbound leads should be sold to. A host added that with outbound you know a prospect fits and must check for pain, while with inbound you must check fit, and that an inbound lead that does not retain is no good.
“30 % of our inbound leads were ICP”
Listen to the episode Pipeline & demand generation Link to this Report a problem
Mark cited Slack's documented example of a leading indicator, where 80% of customers sent 2,000 team messages every month. Listen
Mark said Slack had one of the most famously documented examples of a leading indicator of retention. He said the LIR should be measured in the short term and expected to correlate with retention over time. He said that after a year, the company could calculate whether the LIR actually predicted retention.
“80 % of customers send 2000 team messages every month”
Listen to the episode Metrics & finance Link to this Report a problem
Product-market fit should be measured through retention, and that a team can use a survey test but he does not love surveys. Listen
Mark said that the Sean Ellis test, where 40% of customers would be very disappointed without the product, is one rigorous measure, but he said he does not love surveys because of human bias. He said that in his businesses, product-market fit is best represented in the long term by retention. He said the LIR should be defined as the early behavior that predicts renewal and expansion.
“just don't love surveys, right? And like the human bias that can come out of it.”
Listen to the episode Metrics & finance Link to this Report a problem
Product-market fit is always at risk, and he has never seen a business scale where it did not become at risk. Listen
Mark said product-market fit can erode because of technology, macro conditions, competitive landscapes or customer desires. He said he thinks LinkedIn Recruiter stayed in use for eight years because the other tools lost product-market fit. He argued that a business should measure product-market fit about three quarters before its peers do.
“I've never seen a business scale where product market fit does not become at risk because of movements in technology, macro conditions, competitive landscapes, customer desires.”
Listen to the episode Strategy & market Link to this Report a problem
The product owner should own product-market fit, though he said a wrong answer could be a problem for the CRO. Listen
Mark said the product owner's job is to define who the product is built for, which is another way of saying product-market fit. He said this could be problematic for the CRO if the answer is wrong or the market is too small for the growth goals. He said the market size can be calculated.
“So I do think it's the product owner.”
The best organizations recalculated ICP about once a quarter, and AI could make that continuous across many attributes. Listen
Mark said the best organizations today may have had their ops team re-run ICP once a quarter using four or five segment attributes. He said AI could allow this to happen hourly across an effectively unlimited number of attributes. He said ICP can shrink because of new competition or economic conditions, or expand because of product releases.
“We can do it hourly on an infinite number of attributes.”
Listen to the episode Strategy & market Link to this Report a problem
Mark's second milestone is collapsing SDR, AE, AM and CS into one role that owns the whole customer cycle. Listen
Mark said the SaaS era specialized roles, starting when SDRs were carved out from AEs and later when CSMs were carved out, and he said that went too far. He said AI should allow the field to reverse specialization and reach a much higher level of efficiency. He called this and the selling time target his two convictions.
“collapse your go to market org from SDR plus AE plus AM plus CS to one role that owns the whole thing”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
A host argues revenue leaders should look for rainmakers rather than designing their organizations around average B players. Listen
The speaker said SaaS go-to-market models were built around B players because they were easy to find and hire. He said the volume of hiring means many will be B players, but leaders should be willing to hunt for rainmakers and then work to make them more productive. He said the SaaS era penalized A-plus performers as territories shrank, and they often left.
“I think having the audacity to say, I'm going to go and look for these rain makers and it's going to be hard to find.”
Listen to the episode Hiring & team building Link to this Report a problem
The first phase of AI is about streamlining current workflows, and the second phase will be about reinventing them. Listen
Mark used the Web 1.0 analogy, when the internet was seen as putting a brochure online and people could not yet conceive of user-generated content or companies like Uber and Twilio. He said he is not bullish on the survivability of many companies funded today because they focus on workflow streamlining. He said AI 2.0 will be about workflow reinvention, which is hard to conceptualize.
“I feel like AI 1.0 is about current workflow streamlining.”
Mark expects a very high failure rate for the cohort of AI-native companies funded in the last two years, but is less sure the cohort will perform well as an index investment. Listen
Mark said he has high conviction that the failure rate for the last two years of AI-native investments will be massive. He said he is not sure the cohort will perform as an investment, because a few very large exits could still offset the failures.
“yes, I do think there's going to massive failure rate in the last two years of AI native investments that were done”
Listen to the episode Fundraising & investors Link to this Report a problem
Mark's framework says a company is ready to scale only after it reaches product-market fit, then go-to-market fit, then growth and moat. Listen
Mark said the book has two underlying frameworks, and this sequence is the first. He said he wants to quantify each stage rather than rely on qualitative answers. He said he would rather measure product-market fit than describe it in fluffy terms.
“the answer to when you're ready to scale, you have to get product market fit first, then you have to do go to market fit, then you have to do growth and moat”
Listen to the episode Strategy & market Link to this Report a problem
Mark describes some founders as stuck in overcapitalized walking zombies that missed the window and should do something else. Listen
Mark said many founders he knows are stuck, overcapitalized, have missed the window and should be doing something else. He said VCs often write these companies off. He noted some founders do pivot successfully but are simply tired, and that departing founders should make sure what they leave behind is taken care of.
“stuck -in walking zombies, which are way too overcapitalized, miss the window, and they should be doing something else”
Listen to the episode Fundraising & investors Link to this Report a problem