The short answer
The comparison people want, AI SDR against human SDR, has never been run as a controlled trial. What is measured is adjacent and more useful: acquisition economics are worst in the $25,000 to $50,000 deal-size band, and a fully loaded SDR costs more than most small-deal programmes return. Below that band the question is usually not which kind of outbound to buy, but whether to run outbound at all.
Why does deal size decide this before anything else does?
Because the cost of running outbound is close to fixed and the revenue it produces is not. Here is the model, with every input named and every assumption marked.
The cost side, sourced. The Bridge Group's 2025 survey of 351 B2B companies puts median SDR base salary at $55,000 and median on-target earnings at $80,000. The US Bureau of Labor Statistics Employer Costs for Employee Compensation series, March 2026, puts the employer load at 1.33 to 1.46 times wages depending on occupation. That gives roughly $106,000 to $117,000, before tooling and seats. Call it $110,000 to $127,000 fully loaded. No independent body publishes a fully loaded SDR figure; every number in circulation is produced by a company selling outbound services.
The output side, less solid. Bridge Group puts median monthly SDR quota at 10 meetings — but records 60% quota attainment, the lowest in the survey's history. Ten is the target, not the delivered result, and the report does not publish what is actually delivered.
The show rate, unmeasured. No independent source publishes a no-show rate for booked B2B sales meetings. Chili Piper analysed roughly four million form submissions in 2024 and stopped at the booking stage. The platforms that hold this data have not released it. The 80% below is an assumption, and you should replace it with your own.
Ten meetings booked a month is 120 a year. At an assumed eight-in-ten held, that is 96 meetings. At an assumed 10% close rate, 9.6 closed deals.
Where does the break-even actually sit?
The table holds meeting count and close rate constant and varies only contract value. That is generous to the large-deal rows, because meeting counts and close rates usually fall as deals get bigger. It counts the SDR only. Account executive time to work 96 meetings is real and additional, and no independent source publishes a figure for it.
| Annual contract value | First-year revenue from 9.6 deals | SDR fully loaded | Before AE time |
|---|---|---|---|
| $5,000 | $48,000 | ~$118,000 | Loses $70,000 |
| $10,000 | $96,000 | ~$118,000 | Loses $22,000 |
| $15,000 | $144,000 | ~$118,000 | Clears $26,000 |
| $25,000 | $240,000 | ~$118,000 | Clears $122,000 |
| $50,000 | $480,000 | ~$118,000 | Clears $362,000 |
| $100,000 | $960,000 | ~$118,000 | Clears $842,000 |
Is there a hard floor, and does the evidence support one?
Not as a cliff, and the measured data says something more specific. Benchmarkit's 2025 benchmarks, from 583 companies reporting full-year financials, put the new-customer acquisition cost ratio worst in the $25,000 to $50,000 band at about $2.40 spent per dollar of new annual recurring revenue. The $10,000 to $25,000 band below it performs better, at about $2.20. Efficiency improves above $50,000 and improves again above $250,000.
So the shape is not a floor. It is a trough, and it sits where deals are large enough to need a human selling motion and too small to pay for one.
Two things follow. The widely repeated claim that outbound does not work below $50,000 in contract value has no independent source — it appears to originate in unsourced arithmetic published by a consultancy in 2025, and the underlying intuition traces to a 2019 venture-capital essay that asserts a relationship without quantifying a threshold. And Bridge Group's own sample, the largest study of working SDR programmes available, has a median deal size of exactly $50,000 — meaning half the companies successfully running outbound sit below the number.
Run the model on your own close rate, your own show rate and your own retention. The arithmetic is transferable. The threshold is not.
One caveat runs the other way. This model counts first-year revenue only.
A $15,000 product that renews for three years is worth far more than $15,000, and a company with genuine retention can justify outbound well below where a first-year model says stop. Run the model on retained value if you have the retention data to support it.
Does AI change the cost side or the revenue side?
The cost side, sourced. The Bridge Group's 2025 survey of 351 B2B companies puts median SDR base at $55,000 and median on-target earnings at $80,000. Applying the US Bureau of Labor Statistics employer-cost load of 1.33 to 1.46 gives roughly $106,000 to $117,000 before tooling. Add seats and software and a fully loaded SDR lands near $110,000 to $127,000. No independent body publishes a fully loaded figure; every one in circulation is agency-produced. Published AI SDR pricing runs from $49 a month at Regie.ai to $3,750 a month at 11x, with AiSDR publishing $250, $900 and $2,500 tiers.
On the cost line the comparison is not close.
The revenue side is where the evidence collapses. Fang et al.'s pooled analysis of seven separate experiments splits on who started the conversation.
AI that responded to somebody who had reached out raised sales 16.3% among 44,614 consumers. Firm-initiated AI, meaning AI that contacts strangers, produced nothing measurable. Cold outbound is firm-initiated by definition.
The personalisation premium is real and small. Across 76,977 outreach recipients, personalisation bought about 0.43 percentage points of additional response. On a thousand emails that is four extra replies.
The effect is genuine. It will not rescue an economic model that is already under water.
One more finding belongs here. Luo et al., writing in Marketing Science in 2019, found that disclosing AI identity before a sales call cut purchase rates by 79.7%. Non-disclosure is not a strategy available to you.
What does the comparison table actually look like?
| Human SDR | AI SDR | No outbound at all | |
|---|---|---|---|
| Cost | $110,000–127,000 fully loaded for the SDR, plus AE time | $49 to $3,750 a month published; most vendors publish nothing | Nothing |
| Time to first meeting | Weeks of ramp | Days | Not applicable |
| Evidence of revenue effect | Decades of practice, no controlled trial against AI | Push messaging to 13.7 million showed no significant effect | Untested against either |
| Where it helps most | Complex deals with real discovery | Volume and coverage of a long list | Products that sell themselves |
| Failure mode | Slow ramp and attrition | Deliverability collapse, brand damage at volume | Growth capped by inbound |
| Fit | Clears the model most easily above the $25,000 to $50,000 band | Same | Where first-year revenue per deal does not cover a selling motion on your own numbers |
The fit row does not distinguish the first two columns, because deal size moves both of them the same way. It gives no cut-off for the third, because the model above is the cut-off and it is built from your close rate, your show rate and your retention rather than from a published threshold.
Who does AI actually help, according to the evidence?
The inexperienced, consistently. Brynjolfsson, Li and Raymond, publishing in the Quarterly Journal of Economics in 2025 on 5,179 support agents, found a 14% average productivity gain that broke down to 34% for novices and close to nothing for experienced staff. Dell'Acqua et al.'s study of 758 consultants found a 43% gain for below-average performers.
The same consultant study carries the warning. On tasks that required judgment, AI users were 19 percentage points less likely to be correct, while the work they produced looked better presented.
A founder who has never run outbound is the novice case, which is the case where the gains are largest. That founder is also the least equipped to notice when the output is confidently wrong.
What has nobody measured?
Nobody has run a randomised trial of an AI SDR against a human SDR doing the same job. Nobody has measured what happens when a company that has never done outbound starts doing it with AI.
That second gap is the one that matters, because it is the situation almost every buyer is actually in. Very few companies buying an AI SDR are firing a human SDR. Most have nobody doing the job at all, so the comparison in the product name is not the comparison they face.
A human SDR replaced is a cost you can measure. A human SDR never hired is a counterfactual nobody has run.
The closest available read on the outcome is Gartner's early-2026 survey of 210 chief sales officers, where 25% reported a return of 50% or better and 20% reported a negative return of 50% or worse. The spread is the finding. Roughly the same number of organisations went forwards and backwards, which is what a coin flip looks like when it is priced annually.
So which one should you buy?
If your contract value is comfortably above the $25,000 to $50,000 band where acquisition economics are worst, buy the cheaper of the two and spend the difference on the people who work the meetings, because at that value the meeting is where the money is won.
If it sits inside that trough, buy neither and spend the money on responding to people who contacted you first. That is the side of the split with a positive randomised result behind it.
If you are between the two, run the model on retained value rather than first-year revenue, and make the decision on that number.