I hosted a fireside chat recently with Mark Roberge, the founding CRO at HubSpot and now a lecturer at Harvard Business School, and Sahil Aggarwal, Von's CEO. Sahil opened with a stat most companies have never measured about their own team.
The Von team ran an analysis across 1,200 sellers at 50 companies and found that the average AE spends 1.2 hours a day selling. It is the same number our Rep Time Allocation product surfaces every time a customer connects their calendar and call data. Sellers everywhere are running at a fraction of their available selling time, and most have never checked.
"That's crazy," Mark said.
Most sales leaders assume their own team is the exception to that number. The formula they run instead is the one that has held for decades: if you need to grow revenue by $10 million, hire ten more reps. That math assumes every existing rep's day is already spoken for. It rarely asks whether it is.
Once you accept that most reps are sitting on a lot of unused capacity, the conversation gets more complicated than "hire less, automate more." Mark and Sahil spent the rest of the hour arguing through two harder questions, and neither one landed on a clean answer. I think that is the right place to leave it, and I think more revenue leaders should be having this argument out loud instead of assuming they already know how it ends.
Which Lever Moves
Mark teaches a formula at HBS called revenue velocity. It breaks a rep's output into four variables:
- Opportunities actively being worked
- Close rate
- Average deal size
- Sales cycle length
Multiply the first three, divide by the fourth, and you get quarterly output per rep.
Deal size is a pricing decision, not an efficiency lever. Close rate and cycle length depend mostly on buyer behavior; a buyer looping in three more stakeholders will not move faster because you did. That leaves one variable entirely within your control: how many opportunities a rep can actively work at once, which is a direct function of selling time. That is the lever AI moves right now.
Most of what eats a rep's day is not selling. It is the work required to be ready to sell: building a briefing before every call, reconstructing where a deal stands from notes and memory, updating the CRM, drafting the follow-up. None of it happens in front of a customer, and none of it is optional. AI can absorb that work directly, pulling deal history, the last call, and open risks into something a rep can act on instead of assemble from scratch. That is what hands hours back to the day.
So the capacity is there. What you do with it is where the argument starts.
Question One: How Does the AE Role Change?
The real shift is not that reps get two free hours. It is that AI can now absorb a meaningful piece of the AE role itself.
Take context. A rep with 15 open deals used to hold the state of every one in their own head, or dig it out of the CRM before every call. A GTM intelligence layer like Von already knows the deal history, the last call, and the account context for every opportunity a rep owns, so that synthesis no longer has to happen from memory or from scrubbing back through notes and recordings.
Once AI is carrying that weight, the question changes. It is not "what does a rep do with two free hours." It is "what does the AE role become?"
Mark's answer was role collapse. The standard structure for two decades has split the cycle: SDRs book, AEs close, AMs and CSMs manage the relationship after. That specialization was a workaround for a time constraint that is starting to disappear. Mark pointed to engineering as early evidence: the line between frontend, backend, and product roles has blurred as AI closed the gaps that used to justify separating them. He argued sales is a year or two behind that same curve.
Sahil had a different perspective. A fully collapsed role only works if AI also absorbs enough of the SDR and CSM workload that the freed-up time does not just get eaten by a different set of tasks. But he did acknowledge that he's already seeing this happen with his own engineering team. His PMs now fix small bugs themselves, and his frontend engineers write backend code, because the specialization that used to define those roles has already eroded.
Both things can be true. What the AE role becomes once AI is carrying real weight, and whether the org chart underneath still makes sense, is a decision every revenue leader needs to make deliberately instead of by default.
Question Two: How Do You Get There
Mark's answer was to stop trying to convert the whole org at once. Pull two or three people out and let them run a different model while the rest of the team keeps operating on the playbook that is already hitting its number. He was specific that this should not be your top performers. Your best closers are great at executing a known process, not at the kind of experimentation this requires. Give the test team room to figure out what works, without a quota attached, and only roll the model out once you have real proof it holds.
Underneath both arguments is the same requirement: a system that already understands your business, with deal history, account context, and risk signals sitting in one place instead of scattered across a CRM, a call recording tool, and someone's inbox. At Von, we call that the revenue brain, and it works differently than the agent layer most companies are building. An agent that drafts a follow-up still has to go looking for context every time it runs, querying the CRM and stitching an answer together on the spot. A revenue brain has already spent time learning your business before anyone asks it anything, the same way a tenured RevOps leader would after a year on the job. That difference is what determines whether a rep gets meaningful time back or just a faster version of the same manual work.
Where This Leaves You
Mark and Sahil did not land on a finished answer, and we are not going to pretend we have one either. Most revenue leaders have not run this number on their own team, have not asked what they would do with the time back, and have not tested any of it in a way that would tell them something real.
Pull your own calendar data. Have the conversation about specialization versus full-cycle ownership before AI forces the decision on you. And if you decide to test a new model, do it the way Mark suggested: small team, no quota, real evaluation before you scale it.
Keep Going
Mark and Sahil covered more ground than fits here, including a longer exchange on why most companies are better off buying this kind of intelligence layer than building it in house. Watch the full conversation.
And if you want to support a good cause while you're at it, Mark's book, The Science of Scaling, walks through the same first-principles thinking he brought to our conversation. Every dollar of proceeds goes to McLean Hospital's mental health research.
