What is the ROI of an AI SDR compared with a human SDR?

Compare cost per held meeting, not cost per seat. Our estimate for one US human SDR is about $114,000 a year before tools and management, or roughly $950 to $1,900 per held meeting. An AI SDR starts at $198 a month for one LinkedIn agent, plus the time of the person who runs it.

Every figure marked estimate below is our own arithmetic on the inputs we list. Prices are Sellinger's list prices as of 19 October 2026. The salary and cost data are US figures from the sources linked. Replace the inputs with yours.

Why cost per meeting and not cost per seat?

A seat is an input. A meeting is the output you wanted. An SDR who costs $9,500 a month and books ten held meetings is a different purchase from one who books three, and an agent at $198 a month is a different purchase if it books none. Cost per seat also hides the work around the seat: someone has to pick the audience, write the offer and take the calls, whether the seat is a person or software.

The unit we use is the held meeting: a meeting where a prospect and a host were both there, with evidence. We explain why in Count meetings honestly. If you compare booked meetings on one side with held meetings on the other, the comparison is rigged.

What does a human SDR cost? (estimate)

We build the estimate from two published sources.

Estimate: if we treat the full $80,000 as wages and wages as 70% of total compensation, the employer cost is $80,000 divided by 0.70, or about $114,000 a year, about $9,500 a month. Two cautions. First, the BLS share is an average across all private industry, not a sales team. Second, we leave out recruiting, software, the manager's time and equipment, so this is a floor. The Bridge Group reports a median SDR manager OTE of $146,000, a cost this model does not allocate to the SDR.

We looked at the Bureau of Labor Statistics page for wholesale and manufacturing sales representatives as a cross check. It reports a May 2025 median of $72,080 for non technical products and $104,920 for technical and scientific products. That is a different job from an SDR, so we do not use it in the model. It only shows that the Bridge Group figure sits in a believable range.

How many meetings does a human SDR book? (scenarios)

The Bridge Group reports a median monthly quota of 10 held meetings, and that only 60% of SDRs are at quota, the lowest in the study's history. It does not publish the average meetings actually held, so we show three scenarios. They are estimates, not benchmarks.

The math: $9,500 a month divided by held meetings for steady state; $114,000 divided by nine productive months of meetings for year one. Because average tenure is 1.9 years, ramp matters: about an eighth of an average SDR's time in the job is ramp.

What does an AI SDR cost? (price plus a person)

The software price is a fact. The person is an estimate.

• Software. One LinkedIn agent is $198 per agent per month for one to two agents, with lower per agent prices at higher counts. Bundles run from $249 to $1,199 a month. Email starts at $39 a month. See the pricing page for current numbers. Sellinger offers 30 days free on messaging modules and bundles, no card.
• The person. Someone sets the audience, writes the offer, reads a sample of conversations and takes the calls. We assume 5 hours a week, an estimate, and cost it at the BLS private industry average of $46.89 an hour for total compensation: about 21.7 hours a month, about $1,015 a month.
• Data. If you use your own list, nothing extra. A purchased list is a separate line.

Estimate: one LinkedIn agent plus the person is about $1,215 a month. With a $249 bundle it is about $1,265. We use $1,215 below.

Note what this says. The person costs five times the software. That is the honest shape of an AI SDR budget: the software is cheap, and the attention is the cost.

What is the cost per held meeting for the AI side? (estimate)

Break even (estimate). Against a human SDR at quota ($950 per held meeting), the AI side needs about 1.3 held meetings a month. Against a human at half of quota ($1,900), it needs about 0.6. Those thresholds are low, and that is why the right question is whether you get any meetings at all and whether they convert, not which seat is cheaper.

What should you not conclude from this?

• That an AI SDR books as many meetings as a human. It might book more or fewer. It depends on your audience and offer. On our own account, meetings per person messaged ranged from under 0.4% to 3.7% depending on the audience, with the same agent (Sellinger case study).
• That booked equals held. Of 351 showcase meetings booked through our agents, 33 are confirmed held with a recording or a host mark. The rest are unconfirmed, not necessarily no shows, but we cannot prove them. If you apply a haircut like that to your own agent's bookings, cost per held meeting rises in proportion.
• That the seat decision is binary. Many teams use an AI SDR beside people. A person still sets the offer, takes the call and closes. See AI SDR vs human SDR.
• That meetings are revenue. Pipeline depends on show rate, win rate and deal size, which are yours. Our ROI calculator lets you enter deal value and shows a projected pipeline. Its human SDR cost is a rough US estimate with a different set of assumptions from this post.

How do you measure cost per meeting for your own account?

1. Fix the period. Use a calendar month, not a week.
2. Add up the full cost. Software, the person's hours at a loaded rate, and any data you bought.
3. Count held meetings. A recording, a transcript or a host mark. Keep booked and held as separate columns.
4. Divide. Total cost over held meetings.
5. Repeat for three months. Bookings can take weeks. In one study a solo founder's median from first touch to booking was 22 days (ShortVibe case study), so month one understates the steady state.
6. Compare with the next best use of the same money and hours. For some teams that is a human SDR. For others it is a partner programme or doing nothing.

If you want to test this on your own numbers, start with a defined audience and one channel, run it for 30 days, and read Do AI SDRs actually work? for the weekly checks.

Where does this model break?

• It uses US pay. Markets differ a lot.
• It uses an average benefits share and a median quota, not your company's.
• It does not price the cost of a bad hire, an SDR leaving at month eight, or an AI SDR sending the wrong message. Both have risks, and the controls are different: a manager and a coaching plan for a person, and approval mode and stop rules for the agent.
• It does not value speed. An agent can reply within minutes at night and at weekends, which may matter for your deal size.
• The 5 hour a week assumption is a guess. Measure your own hours in the first month and replace it.