Ten cases. About a dollar.
This is a constructed scenario, not a single account. We assembled a mix of cases designed to simulate roughly a month of usage for a large customer — reports arriving across all four intake modalities, with a deliberate spread of light and heavy back-and-forth, plus investigator AI-helper sessions of varying depth on the cases that warranted them. Every figure is modeled on the actual prompt paths in the app, priced on Claude Sonnet 4.5.
That works out to roughly $0.11 per case.
What each AI job costs
Per-unit rates derived from the live intake and investigator prompt paths.
“Light / heavy” reflects the depth of back-and-forth. Chat intake cost scales fastest because the whole conversation is re-sent on every turn; the case AI helper scales with how many investigator questions are asked and how often it pulls the full case.
Ten reports, a mix of depths
| Case | Modality | Intake load | AI helper | Tokens ↑ | Tokens ↓ | Est. cost |
|---|---|---|---|---|---|---|
| CH-1042 | Chat | Heavy~14 turns | Heavy~7 sends, full pull | 75,700 | 4,830 | $0.30 |
| CH-1043 | Chat | Light~5 turns | Light~3 sends | 20,700 | 1,630 | $0.09 |
| CH-1044 | Webform | Heavy3 passes | Heavy~7 sends, full pull | 47,800 | 3,630 | $0.20 |
| CH-1045 | Webform | Light1 check | — | 1,400 | 230 | $0.01 |
| CH-1046 | Phone | Humanno AI intake | Light~3 sends | 12,700 | 1,030 | $0.05 |
| CH-1047 | Phone | Humanno AI intake | — | 700 | 30 | $0.003 |
| CH-1048 | Humanno AI intake | Heavy~7 sends, full pull | 45,700 | 3,030 | $0.18 | |
| CH-1049 | Chat | Heavy~14 turns | Light~3 sends | 42,700 | 2,830 | $0.17 |
| CH-1050 | Chat | Light~5 turns | — | 8,700 | 630 | $0.04 |
| CH-1051 | Humanno AI intake | Light~3 sends | 12,700 | 1,030 | $0.05 | |
| Month total — 10 cases | 268,800 | 18,900 | $1.09 | |||
Where the money goes
Two patterns dominate spend: heavy chat intakes (full conversation re-sent each turn) and heavy AI-helper investigations (the agent re-pulls the full case on every question). Phone and email reports cost essentially nothing until an investigator opens the helper.
Illustrative model. Actual per-message token counts are captured in the platform's AI usage records.