Four published estimates of a fully loaded US SDR: $134,000, $142,500, $154,000, and one source giving a range of $102,000 to $210,000. The middle three sit inside a 15% band on what should be simple addition; the fourth is twice as wide as the other three combined.
The temptation is to average them. Do not. Every one of these figures is published by a company selling an alternative to hiring, which means the errors do not cancel out, they all lean the same way.
Do not average them. The spread is not measurement error, it is a disagreement about which costs belong in the total, and the disputed line items are the interesting part.
The itemisation
The most detailed published breakdown looks like this. The first two rows are uncontroversial. Everything below them is where the estimates diverge.
| Line item | Annual | Disputed? |
|---|---|---|
| On-target earnings | $65,000 | No |
| Benefits and payroll taxes | $16,250 | No |
| Tools and software | $9,000 | Sometimes |
| Management overhead | $18,000 | Often |
| Annualised ramp cost | $20,250 | Usually excluded |
| Annualised turnover cost | $14,000 | Usually excluded |
| Total | $142,500 |
Base pay and benefits are 57% of the total. The other 43% is the part people argue about, and the two largest disputed items — ramp and turnover — are the ones a first-time buyer never thinks to include.
Why ramp and turnover are real costs, not accounting theatre
The instinct is to treat ramp cost as fictional. You are paying the salary anyway; calling the first four months a separate expense feels like double-counting.
It is not, and the reason is tenure. Median ramp to 80% of quota is 3.9 months. Median tenure is 17.6 months. So 22% of the average rep's entire time at your company is spent producing partial output, and because they leave after roughly a year and a half, you pay that 22% again on the replacement, and again on the one after that.
The number that actually matters
Annual cost is the wrong unit. What you are buying is held meetings, so price those.
Take the median 14.6 booked meetings a month, apply the 20% no-show rate, and you get about 11.7 held. Over a year that is roughly 140 held meetings. Against the mid-range cost estimate:
| Cost estimate | Held meetings / yr | Cost per held meeting |
|---|---|---|
| $134,000 | 140 | $957 |
| $142,500 | 140 | $1,018 |
| $154,000 | 140 | $1,100 |
Published figures for in-house cost per held meeting run $821 to $1,150, which brackets our arithmetic and suggests the model is roughly right.
Why this number gets misquoted in every AI comparison
Vendors selling AI SDR software routinely compare their cost per meeting against a human figure around $500. That number is not wrong exactly, it is what you get if you divide base salary alone by booked rather than held meetings, and exclude ramp, turnover, management and tooling.
In other words the comparison uses the narrowest possible human cost against the broadest possible software output. The honest comparison is roughly $960 to $1,100 per held meeting for a human, and you should demand the software's figure be computed the same way — held, not booked, with the data and infrastructure costs included.
That is a genuinely favourable comparison for the software in many cases. It just is not the one being advertised, and a buyer who accepts the advertised version has no way to check it.