In-house SDRs, outsourced SDR shops, and AI Sales Departments, priced honestly per qualified op.

Every founder hitting the outbound problem at Series A walks into the same three-door room. Door one is the in-house SDR team. Hire two reps at eighty thousand loaded each, add tools, add a manager, add a year of patience while they ramp. Door two is the outsourced SDR shop. Belkins, Cience, Memory, CIENCE, Martal, SalesRoads, any of the hundred boutique shops selling meetings-as-a-service for six to twelve thousand a month. Door three is the fractional AI Sales Department. One retainer, one operator, agents running the function end to end.

Most founders walk through door one because it is the door the board expects. The plan sits on a slide, two SDR boxes under a VP of sales box, and nobody asks the cost-per-qualified-op question because nobody owns the ratio. The slide looks like a sales motion. The cash flow looks like a sales motion. The output looks nothing like a sales motion, but that takes a year to surface because the first six months are ramp and the next six are the burnout-and-recruit cycle.

A smaller group walks through door two. They have been burned by door one already, they know the recruiting tax and the ramp tax and the turnover tax, and they decide to externalize the whole problem to a shop that has run this playbook a hundred times. The economics look better on paper. Six to twelve thousand a month is cheaper than two loaded reps. The output is still gated by humans inside the shop, but at least those humans are someone else problem. That is the pitch and it works on a lot of founders.

The third door is newer. Door three is a function operated by AI agents under a single operator, priced per function and not per person, with output gated by machine time and not by human attention. The retainer is smaller than door one and competitive with door two, but the volume runs ten to twenty times higher because the labor ceiling that constrained the first two options is no longer there. Cost per qualified op collapses by an order of magnitude. We laid the underlying mechanics out in The 80-Email SDR Trap. This page is the head-to-head across all three doors, with the dollar numbers visible.

Two SDRs loaded run a hundred and sixty thousand a year in salary alone. Add Apollo, Outreach, LinkedIn Sales Navigator, ZoomInfo, the warm-up tooling, the calendar tools, the meeting recorders. That is another twenty to forty thousand on the tool stack. Add the recruiter fee on each hire, fifteen to twenty thousand a head, because you will hire each seat at least twice over twenty four months. Add the manager tax, because the moment you have three reps you need a head of SDR or you lose a quarter of the VP head calendar to one-on-ones.

Now look at the output. Eighty cold emails per day per rep is the labor ceiling. Sixteen hundred emails a month per rep, sixteen replies if a one percent reply rate is honest, eight real replies once you strip out the out-of-office bounces. Two discovery calls. One qualified opportunity per rep per month. Two reps, twenty four qualified opportunities a year. Divide two hundred thousand of loaded cost by twenty four and you get a number above eight thousand per qualified op on the conservative math, closer to forty thousand once you load the loss column, the no-shows, and the deals that ghost after demo.

The structural problem is not the dollar number. It is the volatility. At any given moment one rep is ramping, one rep is interviewing internally for an AE seat, one rep is recovering from a bad month, one rep is being managed onto a PIP. The function never runs at theoretical peak. The output line is bumpy in a way the salary line never is. Boards ask why pipeline is volatile and the answer is always the same. Pipeline is volatile because the labor is volatile, and the labor is volatile because SDR is the highest-turnover seat in modern software.

Founders who run door one for two years and survive it usually exit the cycle when the lead SDR quits in month seventeen, takes the playbook with them, and the runway model can no longer justify the rebuild. That is the moment most teams open the door two conversation, or call us. The cost-per-qualified-op number written down at that point is the same number they could have written down on day one if anyone in the building owned the ratio.

The outsourced shop pitch is clean. You pay six to twelve thousand a month per pod. A pod is typically two SDRs sitting in Manila, Bogota, or Krakow, plus a campaign manager stateside, plus a shared data and ops layer. They run your outbound out of dedicated sending domains on their infrastructure. They book meetings into your calendar. You get a weekly report and a monthly QBR. The math on the slide reads twenty to thirty qualified opportunities a year for nine thousand a month average, which is roughly five thousand per qualified op. Half the cost of door one. The pitch is honest as far as it goes.

What the pitch leaves out is what happens after month three. The pod that opened your account had your founder on the kickoff call. The pod that runs your account in month four has rotated two of three seats. The campaign manager you trained the messaging with has been promoted to a senior pod across four accounts. The new pod manager reads your one-pager on Monday morning and writes campaigns from it. The personalization that was forty percent custom in month one is twelve percent custom in month five. The reply rate that started at two and a half percent settles at nine tenths of one percent. The number of meetings stays roughly constant because the shop tunes volume up to compensate, which works until your domain reputation slips and the deliverability tank empties.

The other problem is the quality of meeting. Outsourced SDRs are paid on meetings booked. The contract reads net-new SQLs per month. The incentive structure rewards anything that looks like a meeting on the calendar. Your reps spend the first ten minutes of every call qualifying the lead out of the funnel, because the shop pushed a curious VP of an adjacent vertical onto the calendar to hit quota. The cost-per-qualified-op number on paper is five thousand. The cost-per-actually-qualified-op number after your reps filter the no-fits is closer to nine or ten. You only see this after two quarters of pipeline data.

The third issue is reversibility. Most shop contracts run six to twelve month minimums with two-month opt-outs. The data they generated about your ICP, your sending infrastructure setup, your sequence performance, lives in their stack and you do not get a clean export when you leave. The institutional knowledge of what worked in your vertical walks out the door with the campaign manager when the relationship ends. You start the next vendor relationship from zero, or you go back to door one with a budget the board will not approve a second time.

Door one and door two look like different products on the slide. They are the same product. Both are human SDRs sending cold emails, gated by the labor ceiling of about eighty real-personalization touches per day per rep. The only difference is whose payroll the humans sit on, which controls who carries the recruiting tax and who absorbs the turnover. The underlying unit economics, cost per touch and quality per touch, are the same. Door two prices the labor more efficiently because the shop runs the seats across multiple accounts. Door one carries the full burden alone.

Door three is not a cheaper version of either. It is a different shape entirely. The agents do the sourcing, the enrichment, the personalization, the sequencing, and the follow-up. The operator audits angles, refines the ICP, and pushes new sequences when the data says to. No human in the loop on the per-email decision. The labor cost of personalization, the variable that broke doors one and two, drops to roughly zero. What scales linearly is supervision overhead, which is what the operator running the department actually does. Volume becomes free and review becomes the constraint, which is the inverse of how doors one and two work.

That structural inversion is why the same monthly retainer ships five hundred touches a day or fifty. It is why reply rates settle at four to five percent instead of one or two, because every email is researched before it is sent. It is why the year-one output runs ten to twenty times higher than either door one or door two on a comparable budget. And it is why the full department offering reads as a single line item on the invoice, regardless of whether you are running a small ICP at three thousand touches a month or a broad ICP at fifteen thousand touches a month.

The other reason door three is different is reversibility. The institutional knowledge that the agents build, your voice profile, your ICP filters, your enrichment sources, your sequence performance bank, lives in a stack you can export at any time. If you decide in month nine that you want to bring the function in-house, you can. The new in-house team inherits a documented motion, a voice profile, and a year of reply data. That is not the off-ramp profile of door one or door two. Door one off-ramps as severance and an empty seat. Door two off-ramps as a contract expiry and a data wall. Door three off-ramps as a clean handoff with everything you built coming with you.