Automated Customer Service System

Scope, in the source’s words

• Proposed definition. “‘Automated customer service system’ means any technology used to provide consumers with customer service where all or substantially all the communication to the consumer is made by such technology rather than by a natural person.” (proposed 6 RCNY 5-310)

• “Customer service” means “assistance by a merchant to a consumer, in response to an inquiry from such consumer, regarding a product, service, account, refund, billing matter, complaint, cancellation request, fraud claim, or dispute in relation to the sale or purchase of a good or service.” (proposed 5-310)

• The underlying law covers trade practices involving “consumer goods or services,” “consumer credit,” and “the collection of consumer debts,” with “consumer” meaning “primarily for personal, household or family purposes.” (NYC Admin. Code 20-700 and 20-701)

• Status. Proposed Oct 20, 2026; comments and a public hearing Nov 19, 2026. Not in effect.

• How these definitions apply to a given program is a question teams often take to their compliance team or counsel.

How it works

• Phone menus (IVR). DCWP’s notice says “interactive voice response systems in the 1970s first introduced automation to customer service.” These route calls by keypress or simple speech, and many take payments.

• Rules-based chatbots. In DCWP’s description they work “by employing if-then logic” and use “predefined combinations of question-and-answer options.” The CFPB’s 2023 report describes them as using “either decision tree logic or a database of keywords to trigger preset, limited responses.”

• LLM-based chat and voice agents. DCWP describes large language models as “statistical machines” that “make predictions by generalizing on the basis of learnt associations, leading them to produce answers that are plausible, but sometimes wrong.” Many AI agents in servicing and collections use these models, often with connections to account systems so they can look up balances or take payments.

• The “substantially all” line. The proposed definition turns on how much of “the communication to the consumer” comes from the technology. A system that drafts suggested replies for a human agent to send works differently from one that talks to the consumer directly. Where a specific setup falls is a question for compliance.

• What the proposal attaches to it. Disclosure “at the beginning of a covered interaction,” a single-step path to a human or a listed alternative, and three years of records including “Logs of all covered interactions.” (proposed 5-312 and 5-313)

What it means for AI agents (operations questions)

• Does the team have one list of every automated channel, including vendor-run ones, and the languages each runs in?

• Is the opening disclosure a configured message per channel and language, logged with each interaction, or written into each flow?

• When a consumer asks for a person, where does the conversation go, and does the history go with it?

• Can the system’s logs show, per interaction, how long it lasted, whether a transfer was offered, whether it was used, and what happened next? Those are the fields the proposal lists.