Salient
Salient builds AI agents for US consumer loan servicing at banks, credit unions, captive lenders and specialty finance companies. Taylor handles inbound and outbound servicing and collections by voice, text, email and chat, carrying context across channels and across product lines such as auto and deposit. It negotiates and processes payments, captures promises to pay, sets up arrangements and hands off to a human agent when escalation is warranted, with built in guardrails for FDCPA, Reg F, UDAAP, TCPA and CAN-SPAM.
Marshall is an automated audit engine for compliance teams. It compiles a maintained repository of federal and state law into rules the lender approves, then checks every call, interaction and account event, from payment posting and notices to repossession triggers and right to cure windows, and routes violations to remediation queues with exam ready reporting. Flyn handles total loss and GAP claim recovery, Alex handles chargebacks and payment disputes, and Melanie handles chargeoffs.
Westlake Financial raised its right party contact rate 34% within 60 days of using Taylor for outbound collections, and American Credit Acceptance moved from a 3% manual sample to Marshall reviewing every interaction. Ally Financial and Exeter Finance are also listed as customers. The company, Skwid Inc. of San Francisco, was founded in 2023 and has raised $75 million from Andreessen Horowitz and Matrix. Most deployments reach production 60 to 90 days after signing, and pricing comes through a demo.
Capability Axes
Capability grades
15 of 15 axes rated · 10 graded A or B
Taylor holds voice, text, email and chat conversations with borrowers and completes payments and arrangements, and Marshall applies compiled legal rules to every call, interaction and account event. Remove the models and neither agent has a product left, so the AI is what the company sells.
Taylor negotiates and processes payments, captures promises to pay and sets up arrangements without a person reviewing every call, under real time guardrails for FDCPA, Reg F, UDAAP, TCPA and CAN-SPAM, and hands off to a human agent when escalation is warranted. Marshall rules need lender approval before they run. The thresholds that trigger a handoff, and approval rules for each servicing action, are not published.
Marshall keeps an audit trail for every account event with exam ready export, and release notes record specific changes to Taylor's contact detection and outreach controls. A lender can inspect events, flags and product changes. An independent evaluation set or a version and validation record for the underlying models is not published.
Westlake Financial raised its right party contact rate 34% within 60 days of putting Taylor on outbound collections, with lower cost per contact. American Credit Acceptance moved from a 3% manual sample to Marshall reviewing all of its interactions. Each figure belongs to its own customer and product, and the second measures review coverage rather than an exam result.
Identifiable consumer interaction data is barred from model training under the privacy terms, while aggregated, deidentified information may be used for it. Retention of customer records follows the agreement, lenders decide how the service is used, and every AI voice agent identifies itself at the start of a call.
For borrower calls and related records handled for lenders, the company acts as processor, with retention set by the customer agreement and return or deletion at termination. Affected business customers get incident notice within 48 hours. The commercial DPA, a subprocessor list and specific GLBA service provider terms are not public.
Encryption, access controls and security assessments are described in the privacy terms, and release notes add SAML single sign on and audit log export. The site names no SOC 2 or PCI attestation, the trust subdomain does not load, and certification claims found elsewhere carry no auditor report for this service.
Taylor builds in guardrails for FDCPA, Reg F, UDAAP, TCPA and CAN-SPAM, and Marshall compiles CFPB, OCC and state servicing law into rules a lender approves before enforcement. These are presented as software controls the lender uses, with no lender or collection agency license claimed, which fits the product.
Every AI voice agent identifies itself at the start of a call, and Marshall rules pass a lender approval workflow before they deploy, backed by model risk governance documentation and SOP drift detection. Reviewers work flagged activity through event logs and remediation queues. Results by borrower group, such as escalation or error rates, are not published.
Taylor can hand a conversation to a human agent, and borrowers are directed to their lender for account privacy requests. The only public terms on the site cover recruiting text messages, so any remedy for an AI error in servicing sits in the customer contract, which is not public.
AI runs Taylor's conversations and authentication and Marshall's rule checks, and the privacy terms say when consumer data may enter model training. Production model providers, model versions and how lenders hear about a model change are not identified.
Taylor connects directly to existing core servicing, telephony and messaging systems, and Marshall reads account events at the loan management system level, from payment posting to repossession triggers. Public API documentation and a status page are linked, and most deployments reach production 60 to 90 days after a signed statement of work. Specific loan management or telephony platforms are not named.
Taylor and Marshall run as a connected service against the lender's existing servicing systems, and enterprise releases add audit log export to customer owned S3 storage and SAML single sign on. Hosting regions, tenant isolation and the locations where voice and model requests are processed are not stated.
Lenders book a demo for a single product or for a consumer lending portfolio. Agent subscriptions, per conversation rates, minimum commitments and implementation charges are not published, and the only terms on the site cover recruiting texts.
Banks, credit unions, captive lenders and specialty finance companies in the US are the target, across servicing, collections, compliance audit, total loss recovery, disputes and chargeoffs. Westlake Financial, American Credit Acceptance, Ally Financial and Exeter Finance appear as customers, and JPMorgan Chase selected Salient for its Hall of Innovation award.
Alternatives to Salient
Other products that do the same kind of work as Salient, and what each one does.
Skit.ai sells AI agents that run consumer debt collection conversations for lenders, banks, collection agencies, debt buyers and collection law firms in the United States, and it also serves banks and other enterprises in India.
Domu AI runs generative AI agents that conduct financial servicing conversations across voice, SMS and email for banks, fintechs, buy now pay later providers, insurers, loan servicers and business process outsourcers, covering loan servicing, collections, recovery and customer operations.
Floatbot builds voice-first conversational agents for collections, lending, banking and insurance, with its collections agent designed to run debt recovery conversations end to end.
AI Rudder sells voice and chat agents that call borrowers and customers for banks, lenders and other consumer businesses.
Prodigal builds agentic AI for consumer loan servicing and debt collection, running on the Prodigal Intelligence Engine, an internal platform the company states is trained on half a billion consumer finance interactions and used to carry context across its applications.
Kompato AI runs generative voice agents for debt collection, serving first party lenders recovering accounts before charge off and debt buyers pursuing recovery after charge off.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | Data Protection Terms | Implementation | Source |
|---|---|---|---|---|
|
Contact sales
|
Subscription, usage and implementation rates are quoted through a demo for a single product or a lending portfolio. | A data processing agreement is not published, and no pricing tier is tied to one. | Implementation fees are not published. Most deployments reach production 60 to 90 days after a signed statement of work. | Vendor Published |
The product and customer pages direct lenders to a demo. Subscription, usage and implementation rates are unavailable publicly.