Replace your SDR team with AI, before the next ramp cycle.

Two SDRs cost you eighty thousand a year fully loaded, each. That is one sixty in salary alone before you add HubSpot, Outreach, Apollo, ZoomInfo, the sales manager taxing a third of the VP head's calendar, the laptop, the offsite, the SPIFFs, the QBR dinners. Round the all-in number to two hundred thousand a year for the function. That is the ledger entry. Now look at the output and run the division everyone in the room is too polite to do out loud.

Each rep sends sixteen hundred emails a month. Sixteen replies if a one percent rate is honest, which it usually is not because half are out-of-office. Eight real replies. Two discovery calls. One qualified opportunity per rep per month. Two reps, twenty four qualified opportunities a year. Two hundred thousand divided by twenty four is a hair over eight thousand per qualified op. Add the loss column, the no-show rate, the deals that ghost after demo, and the real cost per won deal is closer to forty thousand on a healthy quarter and seventy on a bad one. The forty thousand number is the conservative one. It is the number you can defend to a board. The seventy thousand number is what your CFO sees when she models the reality across the no-shows and the ghosts.

No founder writes this number down. They write down the salary, the tool stack, the recruiter fee, and the ramp budget separately, on separate spreadsheets, owned by separate people. Finance sees a payroll line. Sales sees a quota attainment number. Nobody divides the first by the second because nobody owns that ratio. The function is structurally protected from its own efficiency math by the way most early-stage teams are organized. The full breakdown of how the math compounds lives in The 80-Email SDR Trap, and the cure for it is the same five-step engine that powers our AI Sales Department.

Once you write the number down, you cannot unwrite it. The only question left is whether you keep paying it because the alternative scares you, or whether you rebuild the function around agents that do not have an eighty-emails-a-day ceiling. Most founders pick rebuild within a quarter of running the math. The ones who delay another quarter usually do so because they are about to renew an annual Outreach contract, or because the head of sales is six months from a promotion to VP and the rebuild conversation feels like a vote of no confidence. Both are bad reasons to keep paying forty thousand a qualified op.

SDR is the highest-turnover role in modern software. The average tenure across industry data sits between fourteen and eighteen months, and the modal first-job SDR tenure is closer to twelve. Half of that window is ramp, which means a rep is at full quota productivity for roughly seven months of their tenure before the burnout, the promotion, or the resignation. Then you start the search again. Recruiter fees of fifteen to twenty thousand a hire are normal in this market. Three weeks of interviews. Six weeks before the new rep can write a coherent cold email on your product. Three more months before the sequences they own personally start to perform.

Multiply that across a two-rep team. There is almost no quarter in which both reps are running at full output. One is always ramping, one is always interviewing internally for an AE seat, one is always recovering from a bad month, one is always being managed onto a PIP. The function never sees its own theoretical peak. The org chart shows two seats. The output shows one and a half on a good week and zero point eight on a bad one. The variance is invisible on the salary line and painfully visible on the pipeline line. Boards ask why the number is volatile and the answer is always the same. The number is volatile because the labor is volatile.

Compare to the fractional alternative. The agents do not burn out. They do not interview at competitors. They do not lose their motivation in the second half of the quarter when the pipeline thins. They do not need a manager to coach them through a slump because slumps are an artifact of human attention, not of the work. The cost line stays flat. The output line stays flat. Variance disappears, which is the part most founders underestimate when they finally make the switch. The function becomes a fixed input feeding a predictable output curve, which is what every other infrastructure function on the company looks like already. Sales is the last team to get there because sales has always been the last team to admit the labor was the problem.

A good SDR writes maybe three genuinely personalized emails an hour, on the days they are focused and rested. That is a hard ceiling. Reading the prospect company page, finding two recent signals, opening the LinkedIn profile, scanning a podcast appearance, then writing a coherent four-sentence opener takes twenty minutes if you are fast. Multiply across a workday and the upper bound is twenty real touches if nothing else gets in the way. Nothing else never gets in the way. There are standups, pipeline reviews, manager one-on-ones, a CRM that crashed, a sequence that needs rewriting, a list that needs cleaning. The twenty number is theoretical. The lived number is closer to eight.

So reps drop into the only mode the labor math allows. Templates with name tokens, role tokens, company tokens. Eighty emails a day, written in six minutes, sent in batches. Prospects spot the template inside one line and delete. Reply rates collapse to half a percent. The rep grinds for another quarter and quits. Or worse, they stay and the quality bar of the function collapses with them, because the floor of what the team sends defines the ceiling of what the team expects to ship next month.

AI agents are not constrained by the same ceiling. They can hold every prospect's last six months of public activity in working memory while drafting. They can scan three thousand prospects an hour against ICP criteria, pull enrichment on the hundred that pass, and draft personalized copy on each. The labor cost of personalization, the thing that broke the human motion, goes to zero. What scales is the supervision overhead, which is what the operator running the department actually does. The math flips. Personalization was the constraint and review was free. Now review is the constraint and personalization is free.

That is why the same monthly retainer ships five hundred touches a day or fifty. The agent does the writing. The agent does the research. The operator audits angles, refines ICP, and pushes new sequences when the data says to. Your reps do not write outbound anymore. They close the warm queue. The shift in calendar shape alone is worth the switch. Mornings move from cold drag to warm conversation. Afternoons move from list cleaning to discovery. Everyone in the function gets to do the work they signed up for.

Most teams misread what "AI does outbound" means. They think it is a copilot. A rep opens a tool, types in a prospect name, gets a draft, edits it, sends it. The labor math barely improves because the human is still the bottleneck on every touch. Speed per email goes from twenty minutes to ten. Volume doubles, quality stays mediocre, the rep still burns out at month six. The unit economics shift by a small constant and the structural problem stays in place.

End to end is different. The agents identify the prospects, pull the enrichment, draft the email, send it, track engagement, schedule the follow-up, branch the sequence on reply, and pass the warm conversation into a human inbox already qualified. No human touches any of those steps in the normal case. The operator audits a sample of outbound each day, reviews the angle dashboard each week, refines the ICP each month. Your reps are downstream of the warm reply, not upstream of the cold one.

That shift changes the shape of the sales function in three ways. First, the headcount required to support a target pipeline number drops by a factor of three or more, because you no longer pay humans to do the work that does not require humans. Second, the variance in monthly output collapses, because the inputs to the system are stable. Third, the work the remaining humans do gets meaningfully more interesting, which compounds into better closing rates and longer tenure on the rep seats you do keep.

It also changes what you measure. The old motion measured activity. Emails sent, dials made, sequences enrolled. Activity was a proxy for output because the labor ceiling made volume the binding constraint. With agents, volume is free, so activity stops being a useful metric. You measure outcomes only. Warm replies, meetings booked, qualified opportunities, won deals. The metrics page on Monday morning becomes shorter, more honest, and harder to game. That alone is worth running the function differently.