Billing and balance
How AI Desk is priced, what is and is not charged, volume discounts, and keeping an unattended service funded.
Billing and balance
AI Desk is prepaid. You hold a balance in US dollars, and each report deducts its price from it. There is no monthly subscription and no minimum commitment.
If you also hold a MarketCheck API plan, the two are separate systems with separate accounts. AI Desk runs never appear on your API subscription invoice.
How reports are priced
Two shapes:
Flat price. A fixed amount per run, whatever the vehicle. A Window Sticker is $1.99 whether it is a base model or fully loaded.
Per unit, quoted first. Reports that take a list are priced per vehicle with a minimum. These never run a large job without telling you the total first: call once without a cost confirmation to get a quote, which charges nothing, then call again with the quoted total to run it. A batch cannot surprise you.
Every report's page in the catalog states its price and which shape it uses.
What is never charged
The rule is that you pay for work that produced something you can use:
- A report that fails is not charged.
- A VIN that cannot be valued, usually for want of a usable odometer reading, is listed with its reason and excluded from the bill.
- A listing page we could not read is not charged.
- A page that turns out to show several vehicles, so no figure can be attributed to one car, is not charged.
- A quote call is not charged.
- Re-reading a page already read on the same report is not charged again.
This is deliberate and it matters for retry logic: a failed call can be retried without risking a double charge.
Volume discounts
Larger top-ups earn a discount on the effective per-report cost, applied automatically. The tier is set by the size of the top-up:
| Top-up | Discount |
|---|---|
| Under $250 | none |
| $250 to $500 | 5% |
| $500 to $1,000 | 10% |
| $1,000 to $2,500 | 18% |
| $2,500 to $5,000 | 25% |
| $5,000 and above | 30% |
Where a discount applied, the run's billing block reports both the list price and what you were actually charged, so the discount is visible per report rather than buried in an invoice.
Auto top-up
Set a threshold and an amount, and the balance refills itself when it drops below the threshold.
For anything unattended this is not optional in practice. A backend service that stops at a zero balance overnight is the single most common way a server-side integration fails, and it fails silently from the user's point of view.
Caps on agents
An agent, the machine identity a backend service uses, carries its own limits independent of your balance:
- Per-call maximum, which refuses any single call priced above it
- Daily cap
- Monthly cap
Caps protect the balance from your own code. A loop in an integration is a more likely cause of unexpected spend than anything else, and a monthly cap bounds it.
Be aware when debugging: a cap doing its job and a broken integration look identical from the outside. Alert on cap refusals specifically.
Seeing what was spent
Each run reports its own cost and the balance remaining, both in the response to a server-side call and in the assistant's reply in a chat client. Amounts are always stated in dollars.
Your account carries the full history: what ran, when, what it cost, and which agent or person ran it. Agent activity is attributed to the agent, so a machine integration's spend is separable from a person's.
Refunds
Failed runs are not charged, so there is usually nothing to refund. If a report ran, was charged, and was not usable, raise it and we will look at it. The refunds policy is published on the site.