AI / ML Disclosure
Recovery Console uses machine-learning systems to summarize phone calls, score lead intent, and draft follow-up message copy. This page tells you which models we use, what data we send them, and what controls you have.
1. Where AI is used in the product
| Feature | What the model does | Provider / model family |
|---|---|---|
| Call intelligence | Transcribes and summarizes inbound and outbound voice calls; tags intent (e.g. emergency, quote request). | Anthropic Claude (text); third-party speech-to-text provider for transcription. |
| Follow-up drafting | Suggests SMS / email copy for missed calls and stale estimates; the operator reviews before sending. | Anthropic Claude. |
| Lead scoring | Ranks recovery candidates so the operator works the highest-value first. | In-house heuristics; no third-party model. |
2. What data we send to model providers
For call intelligence: redacted call transcripts and minimal context (job ID, account timezone). We send no payment information, no SSNs, and no health information. For follow-up drafting: customer first name + the relevant job context only.
3. Training
We do not authorize our model providers to train their foundation models on Customer Data. Anthropic processes our requests under their commercial terms, which by default exclude API inputs and outputs from training data sets. We have confirmed this in writing. If a provider changes its default behavior, we will switch providers or disable the affected feature before relying on the new behavior.
4. Human review
AI outputs are surfaced to the operator inside Recovery Console for review. Outbound messages are not sent automatically without operator review unless the operator has explicitly enabled an automation rule for that channel. The operator is responsible for the content of any message that is sent, including TCPA-relevant content (see SMS Terms).
5. Accuracy + limitations
- Call summaries can omit details, mishear names, or misattribute sentiment. Treat them as a starting point, not the source of truth.
- Lead scores are heuristic ranks; they do not reflect any prediction about the end-customer’s creditworthiness, employment, or housing eligibility, and must not be used for any of those purposes.
- AI-generated copy may include factual errors or inappropriate tone. Edit before sending.
6. Bias + fairness
We don’t use AI to make any decision that would constitute an “adverse action” under fair-housing, fair-credit, or employment law. The Service is designed to recover trade-services revenue from existing leads — it does not screen tenants, score credit, or rank job applicants. Customers who attempt to repurpose the Service for any of those uses are violating the Terms of Service.
7. Opting out
Operators can disable AI features in Settings → Integrations. Disabling AI suppresses summaries, drafting, and lead-scoring; the rest of the Service (job ingest, dispatch, audit) continues to function. End-customers who want their interactions excluded from AI processing should ask their provider (the operator) to suppress their record; the operator can do this from the conversation view.
8. Logging + retention
AI request and response payloads are logged for observability and incident response, with the same retention as other operational data (see Data Policy). The logs are encrypted at rest and accessible only to the engineering on-call team.
9. Contact
Questions about how AI is used in your account? privacy@dorsey-worx.com