Insights
What does one qualified lead actually cost on an AI voice agent?
Callable ·
Every AI voice vendor will quote you a subscription price. Almost none will tell you what a lead costs, because that depends on things they don't control and would rather not discuss.
Here is ours, measured over seven days of continuous production calling.
The answer
881 qualified leads against $16,638 of platform spend — $18.89 each.
Measured 11–17 August 2026, across three anonymised outbound tenants.
But the blended figure is the least interesting number in the study.
The spread is the finding
| Tenant | Sector | Qualified leads | Platform spend | Cost per lead |
|---|---|---|---|---|
| A | Superannuation | 818 | $14,000 | $17.11 |
| B | Residential solar | 21 | $772 | $36.78 |
| C | Financial services | 42 | $1,866 | $44.43 |
| Combined | — | 881 | $16,638 | $18.89 |
Three tenants. Same platform. Same week. A 2.6× spread in what a lead costs.
Whatever is driving that, it isn't the platform fee — and it follows that choosing an AI voice vendor on subscription price is choosing on the variable that matters least.
What actually drives it
Two numbers, and they don't move together.
| Tenant | Dials | Connected (25s+) | Contact rate | Leads per 100 conversations |
|---|---|---|---|---|
| A | 317,429 | 43,451 | 14% | 1.88 |
| B | 8,068 | 1,488 | 18% | 1.41 |
| C | 9,096 | 5,334 | 59% | 0.79 |
Tenant C reaches a human on 59% of dials — four times Tenant A's contact rate, and the best in the set by a distance. It also has the worst cost per lead.
Tenant A reaches a human on only 14% of dials, but converts those conversations best, and buys absolute volume with scale: it places eighteen times more dials than the other two combined.
So cost per outcome tracks the product of contact rate and conversion, not either one alone, and certainly not contract size. The largest contract in the set is the cheapest per lead. The middle contract is not the middle cost.
The practical implication for anyone evaluating this category: your list and your offer will move your cost per lead further than any vendor's pricing page will. If a vendor tells you their platform will fix your unit economics, this is the table to hold them to.
What counts as a lead
The strictest definition available, deliberately.
A call counts once, and only when the post-call extractor classifies it as that tenant's configured primary outcome and the record survives a required-field validation gate. Calls lacking captured qualifying data get downgraded out of the success count.
Not counted: connected conversations, positive sentiment, callbacks, transfers that dropped, or any secondary outcome.
That gate fails closed, which means residual classifier error pushes the success count down and the cost per lead up. The number is conservative in the direction that matters.
Why seven days
All three tenants were fully live and calling for the entire window, so no tenant is charged for days it wasn't operating, and nothing here reflects ramp-up. Cost basis is the contracted monthly fee apportioned at 7/30 of a month, with no discount, credit or ramp allowance applied.
What this doesn't tell you
Platform cost only. The denominator excludes list acquisition, human follow-up, CRM licences and any commission on closed business. A full cost-per-acquisition figure is higher for every tenant here.
A lead, not a sale. Downstream close rates differ by sector and sit outside this dataset. You cannot read these as cost per customer.
One week, three tenants. A seven-day window removes ramp-up distortion but is more exposed to weekly variance than a monthly one, particularly for the two smaller tenants whose outcome counts are in the tens. Over the preceding 30 days the same three tenants blended to $22.38 per lead — so the weekly figure isn't an outlier in aggregate, but individual tenants do move. Tenant B ran materially below its monthly average this week; Tenant A ran above it.
These are managed contracts, between $3,310 and $60,000 a month — not the self-serve plans on our pricing page. Read the mechanism, not the headline.
The full note
Every figure derives from read-only SQL against the production call store. The complete method, per-tenant detail and limitations are in Research Note CA-2026-02, and every published CallableAI number lives on our research page with the same treatment.
Deployed white glove. Backed by an SLA.
CallableAI works with organisations that need quality, volume and compliance to hold at once. Bring us your requirements — we'll scope the deployment, the rollout and the SLA with you.