AI Voice Agents for Loan Collections: Staying Inside TCPA, Regulation F and RBI Rules

Outbound collections is the phone call with the most rules attached, in every market. That is exactly why it suits automation: the rules are specific, and specific rules can be encoded so that a non-compliant call cannot be placed. The lenders getting this right treat the regulations as the design spec for the agent, not as a disclaimer read before launch.

Start with what changed: AI voices are regulated voices

On 8 February 2024 the FCC issued a declaratory ruling confirming that AI-generated voices fall within the Telephone Consumer Protection Act's restrictions on "artificial or prerecorded" voices. The practical effect is that an outbound AI voice call to a mobile number requires prior express consent for informational calls, prior express written consent for marketing calls, clear identification of the calling entity, and a working opt-out mechanism. There is no loophole for an AI that sounds human.

The stakes are set by the statute: $500 per violating call, up to $1,500 if wilful, with no aggregate cap. A campaign of 10,000 non-compliant calls is a seven-figure exposure. Consent records, therefore, are not paperwork. They are the gate the dialer checks before every call.

Regulation F and the FDCPA: the rules the agent enforces

For third-party collectors, and as a benchmark many first-party lenders adopt, the Fair Debt Collection Practices Act and the CFPB's Regulation F set the conduct rules:

• Frequency. Regulation F establishes a rebuttable presumption that a collector violates the law by calling a consumer about a particular debt more than seven times within seven consecutive days, or within seven days after having a telephone conversation about that debt. The presumption applies per debt and to telephone calls specifically.
• Time of day. Calls are limited to 8 am to 9 pm in the consumer's local time, absent consent. An agent needs the consumer's time zone, not the lender's.
• Disclosure. The collector must identify itself, state that the communication is from a debt collector, and provide validation information within the required timeframes.
• No third-party disclosure. The debt is not discussed with anyone other than the consumer or authorised parties, which means identity verification before any specifics.
• No harassment, false statements or unfair practices. Scripts must be built so that threatening, misleading or pressuring language is not available to the agent.
• Cease and dispute handling. A request to stop calling, a dispute, or a statement that the consumer is represented by an attorney must be recognised, honoured and recorded immediately.

State laws layer on top: two-party consent states for recording, state licensing for collectors, and state-specific frequency or hour limits that are sometimes stricter than the federal floor.

What an AI collections agent does well

The sweet spot is early-stage and pre-delinquency contact: payment reminders before a due date, first missed-payment outreach, promise-to-pay confirmation, and payment-plan setup within pre-approved parameters. These are high-volume, low-judgment conversations where a courteous, consistent agent that respects every rule outperforms an overworked human queue.

A well-built agent:

• Verifies identity before discussing anything specific, using the lender's approved verification steps.
• Follows the approved script and disclosures exactly, in the consumer's language.
• Offers payment options within policy: pay now by link, schedule a payment, set up a plan from an approved menu.
• Records promises to pay with dates and amounts, and schedules the follow-up inside frequency limits.
• Detects hardship, disputes, attorney representation and cease requests, stops the collection path, records the event, and routes to the right human team.
• Escalates anything ambiguous to a person, with the transcript attached.

Late-stage collections, legal escalation, settlement negotiation and any conversation about a consumer's hardship are human work. The agent's job is to make sure those humans spend their time there.

Compliance as architecture

The controls that matter live in the system, not the script:

• Consent gate. No call is placed without a current, recorded consent basis for that number and that call type.
• Frequency ledger. Every attempt, connection and conversation is logged per debt and per consumer, and the dialer checks the seven-in-seven window before dialling.
• Time-zone-aware scheduling from the consumer's number and address, not the office clock.
• Immutable event log of disclosures made, opt-outs received, disputes raised and promises recorded, with timestamps and recordings.
• Script governance. Approved language only, versioned, with prohibited phrases blocked at generation time.
• Human handoff paths that are tested, staffed and measured.

An auditor should be able to reconstruct any call from the log and see the rule check that permitted it.

India: the RBI layer

For banks, NBFCs and fintechs operating under the RBI's digital lending guidelines, and for vendors serving them, recovery calls run under the RBI Fair Practices Code and recovery agent norms, alongside TRAI's DLT registration and DND rules for outbound calls and the DPDP Act for data. Recovery contact is permitted only between 8 am and 7 pm, including retries and callbacks. The agent must identify the lender and itself, avoid any threatening or abusive language, never disclose the debt to third parties, limit frequency, provide grievance redressal information and offer a human escalation path. The lender remains liable for the conduct of outsourced agents, whether human or automated, so vendor contracts should state who enforces what. We cover the Indian language and telecom requirements in detail in our guide to AI voice agents in India.

Measuring the program

Beyond dollars collected, track: right-party contact rate, promise-to-pay rate and kept-promise rate, complaint rate per thousand calls, opt-out and dispute handling time, and, above all, zero rule-check failures in the log. A collections agent that collects more but generates complaints is a liability, and the log will show it long before a regulator does.

Claudeter builds collections and servicing agents for lenders in the US and India with the consent gate, frequency ledger and conduct controls implemented in the workflow, and a discovery sprint that starts from your compliance team's requirements rather than a demo script.

Sources: NCLC, top TCPA and robocall developments 2024/2025; FCC, Declaratory Ruling FCC 24-17; Mayer Brown, FCC declares authority over AI-generated calls under the TCPA.