Fewer SDRs, more systems. The roles I cut, the ones I added, and what it did to CAC.
I put AI agents into the top of our pipeline, and the thing that changed most was not the pipeline. It was the org chart. The motion needed fewer people doing manual prospecting and more people building and supervising the systems that do it, and that shift rewrote our roles, our manager layer, and our CAC.
This is not a story about a magic tool. It is a story about org design. When the work changes, the boxes on the chart have to change with it, and most teams keep the old chart and bolt the new tools on the side. That is how you end up paying for both.
Our previous structure was the standard one. A wide base of SDRs reporting up through SDR managers, feeding AEs, with a thin RevOps function trying to keep the data clean enough to trust. The whole shape assumed that the binding constraint was human hours spent finding and touching accounts.
That assumption stopped being true. Once an agent could research accounts, draft the first touches, and keep records current, the constraint moved. The bottleneck was no longer how many emails a person could send. It was how good our targeting, data, and routing were, and whether a human showed up at the right moment with judgment.
So the old chart was solving for capacity we had already automated. We were paying a manager layer to supervise activity that no longer needed that much supervision, and paying a base of reps to do work that had become the cheapest part of the funnel.
I want to be precise about the difference, because it matters for how it felt internally.
I cut very little outright. Most of the reduction came from not backfilling. As SDRs left through normal attrition, I held the seats open instead of reposting them. Over a few quarters that took out roughly a third of the SDR seats without a single layoff in that group. That is the humane version, and it is also the version that protects you from cutting too deep before you know the new motion works.
What I did stop paying for, fast, was the manual ops work that sat underneath prospecting. The list-building, the enrichment-by-hand, the copy-paste between tools, the constant manual hygiene. Some of that was contractor spend, some of it was junior headcount whose whole job was feeding the machine. That work did not move to a person. It moved into the system.
Here is the honest part. The reps who stayed were not the highest-volume prospectors. They were the ones with the best judgment in a live conversation. Volume stopped being the thing I was selecting for, and that changed who I wanted in the seats. It also changed how I interviewed. I stopped asking how many dials someone could make in a day and started asking how they would handle an account that went quiet, or a buyer who pushed back on price. The skills I was hiring for moved up the value chain, and so did the cost of a bad hire.
The cuts paid for three investments, and these are the boxes most GTM charts are still missing.
A GTM engineer. This is the highest-leverage hire I made, and it does not fit cleanly in sales or in ops. This person wires the agents to the data, builds the workflows, owns the integrations, and turns "we should target accounts that did X" into a system that actually does it. Without this role, your AI investment stays a demo. With it, the motion compounds. I would rather have one strong GTM engineer than five more SDRs, and that is not a sentence I could have said two years ago.
An agent-ops owner. Someone has to supervise the systems the way you used to supervise people. Are the agents on-message. Are they touching the right accounts. Where are they getting it wrong, and what is the feedback loop to fix it. This is quality control and management for non-human workers, and it is a real job, not a side task you hand to a manager who already has a team. When I tried to make it a side task, quality drifted and nobody owned the drift.
More weight on RevOps. RevOps went from a thin support function to a center of gravity. When the agents run on your data, your data quality and your routing logic are not back-office concerns. They are the product. I grew that team and raised its seniority, because the decisions it makes now directly set how well the front of the funnel performs.
The manager layer got thinner, and the spans got wider in some places and narrower in others.
The SDR manager layer shrank the most, because there were fewer SDRs to manage and because a chunk of what those managers did (pushing activity, checking that the work got done) is now handled by the agent-ops function watching the systems. I did not need a manager for every seven reps when half the activity is machine-run.
On the AE side, spans actually tightened a little. The remaining humans are doing harder, higher-judgment work, and that work benefits from real coaching, not activity policing. So I traded breadth of supervision for depth of it.
The new shape is flatter and lumpier than the clean pyramid I started with. Fewer layers, a smaller base of human prospectors, and two or three specialized technical roles sitting next to the sales org instead of underneath it. If you drew it, it would look less like a triangle and more like a small sales team with a systems team bolted directly to its side, both reporting up to the same place so the priorities do not split.
That last point is the one I would underline. The systems roles cannot report into IT or sit in a shared services pool, because then their backlog gets set by someone who does not own the number. The GTM engineer and the agent-ops owner work for the revenue org, full stop. When the agents misfire, the person who fixes them feels the same pressure I feel, and that alignment is most of why the motion stays sharp.
CAC came down by about a quarter. That is directional, and the path to it was not clean. The savings came from two places. The obvious one is fewer fully-loaded SDR seats and less contractor ops spend. The less obvious one is that the remaining funnel got more efficient, because better targeting and faster follow-up mean you waste less pipeline on accounts that were never going to close.
Now the honest accounting of what broke.
The work the remaining humans do is better. Less manual prospecting, more closing and judgment. The people who stayed are happier, because they spend their day on the part of the job that uses their brain. That is the quiet win that does not show up in CAC but shows up in retention.
The chart you inherited was designed for the constraint you used to have. If AI moved the constraint, the chart is now wrong, and no amount of new tooling fixes an org built for the old problem.
If you are putting AI into your motion and leaving the org chart alone, you are getting a fraction of the return and paying full price for it. Decide what the new constraint is. Staff for that. Then have the discipline to let the old roles go as the new ones prove out.
The GTM × AI signal, every morning. Free.