AI SDR cost
An AI sales-development agent burns tokens twice — researching every lead and drafting for the ones that engage — plus verification and sending tooling that bills per lead regardless.
The formula
Token prices as of , pulled daily from OpenRouter's API. Verification and tooling defaults are volume-tier estimates — paste your vendor's actual rate. Deliberately excluded: domain/mailbox infrastructure and human review time.
Two token bills, and only one of them follows your list size
The research pass bills on every lead you touch. The drafting pass bills only on the fraction that replies. That asymmetry is the whole cost structure, and it is invisible if you only look at a blended "cost per lead" number.
| Bill | Billed on | Scales with |
|---|---|---|
| Lead research | Every lead | List size × depth per lead |
| Personalised drafts | Engaged replies only | List size × engagement rate |
| Verification | Every lead | List size, flat per lead |
| Sending and data tooling | Every lead | List size, flat per lead |
At a 15% engagement rate the drafting line carries roughly one seventh of the token volume of the research line. If you are hunting for savings, that is where they are not.
Cost per lead is the wrong metric — cost per meeting is the right one
A low cost per lead is easy to manufacture: shallow research, generic copy, verified-but-cold lists. It optimises the numerator while destroying the denominator. Work the arithmetic in the other direction instead: divide the total monthly spend by meetings actually booked, then compare that to what a booked meeting is worth to you. That is the only comparison that survives contact with finance.
It also explains an uncomfortable result — two campaigns with very different per-lead costs can land on almost identical cost per meeting, and the more expensive one is often the better business, because it wastes less of your team's attention.
Where teams get this wrong
- Budgeting tokens and forgetting infrastructure. Mailboxes, domains, and a sending reputation that has to be warmed for weeks sit outside the token bill and are usually larger.
- Treating verification as deliverability. They are different problems with different vendors. Verification says the address exists; nothing on that invoice says a mailbox provider will accept your mail.
- Maximising research depth. Doubling research tokens rarely doubles reply quality, but it always doubles the cost line. Find the minimum context that lets you say something specific and true.
- Leaving human review out of the model. Many organisations require a human to approve outbound copy. If a person spends two minutes per message, that labour dominates every token in this calculator.
- Scaling volume before fixing targeting. Ten thousand badly-targeted leads cost ten times as much as one thousand and produce the same pipeline.
Cutting the bill without losing personalisation
- Screen first, research second. Run a cheap qualification pass on the whole list, then spend research tokens only on leads that survive it.
- Share research across accounts. Several contacts at the same company share one company-level research pass — cache it instead of paying for it repeatedly.
- Cap output length. Set a hard maximum on research output. Padding is what turns a cheap model into an expensive one.
- Move drafting behind engagement. Never draft for a lead that has not signalled interest.
The four numbers this estimate needs before it means anything
A cost per lead on its own cannot tell you whether outbound is working, because it is the numerator of a fraction whose denominator is the only part that generates revenue. Pair it with four figures from your own funnel:
| Figure | Why it matters |
|---|---|
| List size per month | Sets the fixed lines — research, verification, tooling |
| Engagement rate | Determines how much of the drafting line you actually pay |
| Meeting rate | The denominator that turns cost per lead into cost per meeting |
| Value of a meeting | The only number that says whether any of the above is a fair price |
Once you have them, divide total monthly spend by meetings booked and compare that to what a meeting is worth to your business. That comparison stays valid when model prices move, when vendors change and when list quality shifts — which is precisely why it is the number worth putting on a dashboard.