These are not three alternatives to each other. AI SDR and human SDR are two ways to staff the work; signal-based prospecting is a way of deciding who the work is aimed at, and it sits underneath either. Most teams comparing all three are really making two separate decisions.
Disclosure: Signl sells the third thing, so the honest framing above is also the convenient one for us. Every competitor and category figure below is quoted from a named page with the date it was read, and where the numbers conflict that is said plainly rather than resolved in our favour.
Two decisions, not one
Decision one: who do we contact? A filtered list, or people who did something observable. That is the subject of intent-based outbound vs cold outreach.
Decision two: who or what does the contacting? A person, software, or a service.
The reason these get tangled is that AI SDR vendors answer both at once, usually by pairing automation with a large bought list, because that combination demonstrates well. The two decisions have different answers for most teams, and separating them is the useful thing this article can do.
What the published costs actually say
Here is the part worth bookmarking, and the reason this article exists: the numbers in this category contradict each other, and nobody collates the contradictions.
| Figure | Source | Read | Note |
|---|---|---|---|
| Human SDR, $116,500 to $210,000 a year fully loaded | Instantly, page dated 19 June 2026 | 25 Oct 2026 | Includes benefits, taxes, recruiting, ramp and management |
| Human SDR, about $139,120 a year | Autobound, page dated 8 Feb 2026 | 25 Oct 2026 | Cites AiSDR's analysis, not its own |
| AI SDR, "a fraction of one human SDR seat" | Instantly, same page | 25 Oct 2026 | No annual figure given. Components listed: domains $15/year, $5 to $10 a month per email account, credits from $9 a month, dedicated IP pools $358 a month |
| AI SDR platform fee, $1,000 to $3,000 a month | Autobound, same page | 25 Oct 2026 | Cites Valley. $15,000 to $35,000 a year for enterprise tiers |
| Cost per lead, $262 human against $39 AI | Autobound, same page | 25 Oct 2026 | Cites MarketsandMarkets. A second-hand citation we could not trace to a methodology |
| AI SDR tool churn, 50 to 70 per cent annually | Autobound, same page | 25 Oct 2026 | Cites UserGems. Described as roughly double the human SDR average |
Three things to take from that table.
The human range spans $93,500. Both ends are presented as the true fully loaded cost. They cannot both be, and the difference is mostly whether management overhead and ramp are counted, which is a modelling choice rather than a fact.
The AI side is quoted as a platform fee, and platform fees are not the cost. Domains, inboxes, warming, data and the person who supervises the thing are all real and are usually outside the number.
The churn figure is the most useful line in the table and it is on the vendor's own page. A category with 50 to 70 per cent annual churn is a category where most buyers conclude, within a year, that it did not do what they wanted.
A cost model with your numbers in it
Rather than pick a side of the above, work out your own. The only figure that compares across all three models is cost per qualified conversation, because it is the only one that survives different definitions of a lead.
(Annual people cost + annual tooling cost + annual data cost) divided by qualified conversations a year
Fill in four inputs:
| Input | Human SDR | AI SDR | Signal-led, run in-house |
|---|---|---|---|
| People cost | Salary, plus roughly 30 to 40 per cent for employer costs, plus management time | Supervision: usually a fraction of one person | The hours someone spends reading and writing |
| Tooling | Sequencer, dialler, data | Platform, domains, inboxes, warming | Listening tool and a sequencer |
| Data | Contact database subscription | Usually bundled, check the row limits | Included per person or per lookup |
| Conversations | From your own CRM | Pilot it, do not take the vendor's number | Pilot it the same way |
The denominator is where the argument is settled, and it is the one nobody can give you. Run a bounded pilot for both of the models you are weighing, on the same segment, and count the same thing.
Where each model is actually good
| Human SDR | AI SDR | Signal-based | |
|---|---|---|---|
| Strongest at | Discovery, objections, anything unusual | Volume and consistency at low cost | Timing and message relevance |
| Weakest at | Cost per attempt, and consistency on a bad week | Judgement, and anything off-script | Absolute volume. It has a hard ceiling |
| Breaks when | You need 10,000 touches | The market is saturated and the list is bought | Your buyers say nothing in public |
| Deal size suited | Larger, where one conversation pays for a lot of salary | Smaller and transactional | Any, but it caps the number of conversations |
| Ramp | Weeks to months | Days, plus domain warming | Days, plus a fortnight of learning the sources |
| Failure is visible | Immediately, in the pipeline review | Late, when reply rates decay quietly | Early, as sources that produce nothing |
The row that decides it for most teams is deal size. A human SDR costing $150,000 needs to produce a lot of conversations to pay for itself at a $5,000 contract value and very few at $80,000. That arithmetic has not changed because the software improved.
The combination most teams land on
Not a compromise, a division of labour:
- Signals decide who. A smaller list where every person did something, which is the input quality problem solved.
- Software does the mechanical part. Collecting, qualifying against the ICP, drafting a first line, routing into the sequencer. This is the part that is genuinely automatable now.
- A person does the conversation. From the reply onwards, where judgement actually earns its cost.
That is the shape Signl is built for: it prepares, qualifies and routes people, and the sending happens in your own tool while the conversation happens with a human. It is deliberately not an AI SDR, and the reason is the churn line above. The tools in that category that disappoint usually disappoint because they automated the part that needed judgement, not because the automation was bad.
If you want the work done rather than the software, our managed service is $2,000 a month, with a founding offer of $1,000 a month plus $250 per booked meeting; the honest comparison against agencies is in done-for-you LinkedIn lead generation.
Questions to ask any AI SDR vendor
Five, and the fourth is the one that produces the most informative silence.
- What is included in the price, and what are domains, inboxes and warming on top?
- Where does the list come from, and what happens when that segment is exhausted?
- What does the agent do when someone replies with an objection?
- What is your annual churn, and what is the most common reason given?
- Can I see the messages it would send to twenty of my real prospects, before I sign?
The full evaluation checklist, for agents generally rather than AI SDRs specifically, is how to choose an AI prospecting agent.
Limits
- Every figure in this article is vendor-published. We have verified what each page says and the date it said it, not whether it is true.
- We publish no performance figures of our own, because we have not measured any reproducibly across customers.
- Cost per conversation ignores quality. Twenty conversations with the wrong people is worse than five with the right ones, and no cost model catches that.
- None of these models fixes positioning. If the answer is "we do not have that problem", the staffing question was never the issue.
