# Automated Customer Service Gets a 15-Minute Human Deadline > California's AB 1609, signed Sept. 28 and effective Jan. 1, 2027, lets large consumer businesses keep AI customer service but requires them to disclose the bot, offer a simple way to ask for a person on every service channel, and connect customers within 15 minutes or book a callback within one business day. Here is what the law requires, why shoppers pushed for it, and how retail teams can build a handoff that meets the deadline. Content type: article Source URL: https://www.agentpmt.com/articles/automated-customer-service-gets-a-15-minute-human-deadline Markdown URL: https://www.agentpmt.com/articles/automated-customer-service-gets-a-15-minute-human-deadline?format=agent-md Updated: 2026-10-06T12:14:28.548Z Author: Pancakes Tags: Successfully Implementing AI Agents, Controlling AI Behavior, AI Agents In Business, Enterprise AI Implementation, News --- # California Gives AI Customer Service a 15-Minute Deadline to Reach a Human Gov. Gavin Newsom signed AB 1609, the Right to Human Customer Service Act, on Sept. 28. The deal it offers large consumer businesses is simple to state. Keep your AI customer service agent. Tell people it is an AI. Give them an easy way to ask for a person, and then either get them to one within 15 minutes or book a specific time for a human to call them back within one business day. The law takes effect Jan. 1, 2027, the week after the holiday shopping peak. It does not ban, cap, or tax the chatbot. What it regulates is the exit from the chatbot, which is where customers say AI service breaks down. In an Alchemer survey of U.S. consumers released Sept. 29, the single most-cited harm from AI in customer service was that it made a person harder to reach, named by 44% of respondents. That is precisely the behavior AB 1609 now puts a clock on, for every large company with California customers. For retailers, grocers, delivery platforms, subscription services, and consumer brands that run their own support, the practical reading is short. Every channel where an AI talks to a customer now needs a recognizable "get me a person" path, a clock, and a fallback booking. The brands that treat that path as part of the agent's design, instead of the place the bot gives up, will find the law easier to meet and the customers easier to keep. ## What AB 1609 asks of a covered business The law reaches businesses with more than $500 million in gross annual revenue nationally that sell goods or services to customers in California. That threshold is measured across the entire company, so a national chain with modest California sales is still in, and so is a direct-to-consumer brand whose support desk sits in another state. Assemblymember Rick Chavez Zbur, a Los Angeles Democrat, wrote the bill. He has said it grew partly out of trying to get medication for his mother during the pandemic and spending hours on hold before driving to the pharmacy with the call still connected. "Technology should make life easier, not block people from getting help," he told lawmakers. As summarized by Fox Rothschild's Odia Kagan and by Law Commentary, the obligations break into five pieces. **Disclosure.** A business cannot present a chatbot as a human. Where a reasonable person could be misled, it has to give a clear and conspicuous disclosure that the bot is artificial, in the same medium the bot uses. A voice bot discloses by voice; a chat bot discloses in the chat. **A way out on every channel.** During regular business hours, the business must offer a simple feature that lets a customer ask for a customer service agent through "commonly understood commands, prompts, menu selections, or verbal requests." Fox Rothschild's summary says the feature has to exist on every platform where the company provides customer service, online chat and phone included. **The clock.** Once a customer asks, the business must make a good-faith effort to connect them with an agent within 15 minutes, or give them a specific appointment within one business day when a person will call or otherwise connect. Fox Rothschild reads the law as also accepting a callback within one business hour. Requests that arrive by email, web contact form, or voicemail fall outside the 15-minute window, according to Law Commentary. **Hold limits after the handoff.** Once a customer is connected, the business must make a good-faith effort to keep any single hold under 15 minutes and total holds, waits, and escalation queues under an hour. Businesses that offer support online and also run a phone line have to post that phone number clearly on their website. **Enforcement.** Public prosecutors enforce the law through civil penalties, with a higher penalty for each violation after the first. Consumers get no new private right to sue over a violation. Hospitals' patient-care communications, consumer reporting agencies, certain regulated utilities, and some business-to-business lines are carved out, and emergencies outside a company's reasonable control can excuse a missed window. Two details soften the edges. Law Commentary notes that commercially reasonable and practicable measures designed to comply satisfy the good-faith standard, so the law asks for a working system, with room for the occasional bad Tuesday. And the clock used to be much tighter. Zbur's first draft gave businesses five minutes, and business groups, including the California Chamber of Commerce, argued that earlier versions imposed high costs and rigid rules across very different support operations. Internet.Works, a tech industry coalition, raised its own concern in a June bill analysis reported by SF Gate: "These pressures push agents to prioritize speed over actually solving the customer's problem." Fifteen minutes plus an appointment option is the compromise that came out of that negotiation. AB 1609 joins a long run of state AI rules, which we tracked earlier this year in [our roundup of new state AI laws](https://www.agentpmt.com/articles/ai-regulation-accelerates-19-state-laws-passed-in-two-weeks). Consumer brands will meet a sibling law in their marketing departments too: SB 1050, also newly signed, requires a disclosure when an ad features an AI-generated performer. One scope question deserves a conversation with counsel now instead of in January. Coverage describes the law as reaching every platform a business uses for customer service. Retailers are also putting conversational shopping assistants in front of customers before they buy; Old Navy and Banana Republic switched theirs on this week. Does a pre-purchase shopping assistant that answers "will this ship by Friday?" count as a customer service chatbot under the statute? The statutory definition will decide it, and a brand launching an assistant this quarter should get its answer from counsel before the law takes effect. ## Shoppers filed this complaint first The Alchemer numbers explain why legislators aimed at the exit. Customers named two failures above the rest. AI made it harder to reach a person, and once they did reach one, many had to repeat information they had already given the bot. Jeannie Walters, CEO of Experience Investigators, gave CX Dive the mechanism behind the first complaint. "A bot's 'scorecard' often rewards keeping customers away from live agents, so that's what the bot is literally trying to do," she said. Grade a bot on containment, and it will get very good at containing. For a covered business, a bot tuned to block the exit is now a legal liability as well as a customer one, which makes this a good season to swap containment for two other measures: how often the customer's issue got resolved, and how long it took to reach a person when they asked. The second complaint is a plumbing failure dressed up as a bot failure. Jon Picoult of Watermark Consulting told CX Dive that forced repetition usually reflects back-end systems that do not talk to each other, such as the CRM, ordering, and billing. The bot knew the order number. The human's screen did not. None of this reads as customers rejecting AI. The same Alchemer survey found that what many people want from a company's AI is help resolving their issue faster. A separate survey from Synchrony and Oxford Economics found shoppers are more willing to trust an AI shopping assistant from a brand they already deal with, and that a bad recommendation from that assistant would make many of them shop with the brand less. Brand trust is the advantage a retailer brings to its assistant, and a dead-end conversation spends it. ## Grocers are adding assistants anyway If AB 1609 reads like a reason to pull the chatbot, the operators with numbers are moving the other way. Albertsons reports that shoppers using its conversational search spend about 10% more per order, with a larger lift for shoppers who use the fuller assistant that builds recipes and ingredient lists around dietary preferences. That is the kind of result that funds the next phase, and Albertsons is now expanding its OpenAI work into merchandising recommendations, employee tools, and shopping inside ChatGPT. Jill Pavlovich, the grocer's senior vice president for digital customer experience, first disclosed the order-value figures in August; PYMNTS re-reported them alongside the expansion on Oct. 5. "When they're using a more comprehensive experience, they're adding even more because they're not forgetting items," she said. DoorDash co-founder Andy Fang told the Groceryshop conference that Ask DoorDash users build bigger baskets and check out faster. His more useful point for anyone designing customer-facing AI was about format. "Oftentimes when people think of using AI to order, they think of a chat window popping up," he said, as reported by Modern Retail. DoorDash found that people shown a blank chat box often do not know what to ask it, so the company puts intelligence into specific moments instead: a prompt to reorder your usuals, a substitute when an item is out of stock. Kroger's new shopping assistant accepts a photo of a handwritten list or a family recipe and turns it into a basket. These are figures the companies chose to share about their own products, so they deserve the usual squint. They still point one direction. For AI consumer products like shopping assistants, the companies measuring results are expanding them, and the ones that do it well are building AI into specific steps of the shopping trip. That same instinct, a defined step instead of an open-ended chat, is how a team should design the handoff to a human. Our [consumer goods and retail industry page](https://www.agentpmt.com/industries/consumer-goods-retail-services) covers more of where e-commerce and AI are meeting in production. ## Building the exit before January A handoff that counts is a short sequence of steps: recognize the request, package the context, route to a person or book a time, and record what happened. Built as a deliberate part of the customer services workflow, each of those steps leaves a timestamp, and timestamps are what a business can measure, tune, and show a prosecutor who asks. **Recognize the ask in plain language.** The statute's own phrasing, commonly understood commands, prompts, menu selections, or spoken requests, is a useful test suite. Run the phrases customers actually type and say ("agent," "representative," "talk to a person," "human please," the frustrated zero on a phone keypad) through every channel, including the new shopping assistant if counsel decides it is in scope. **Hand over the conversation along with the customer.** The transcript, the order, the account, and what the bot already tried should arrive on the agent's screen before the agent says hello. This is the direct fix for the repetition complaint, and it shortens the human conversation, which helps with the hold limits too. **Ask better questions before the handoff.** A preprint posted to arXiv on Sept. 29 by Ido Levy, Asaf Yehudai, Segev Shlomov, Asaf Adi, and Leshem Choshen trained an agent to pick questions based on which ones actually retrieve the information a task needs. Dropped into an interactive customer service setting with a simulated customer, it completed more tasks while asking fewer questions, and in a retail scenario it beat a much larger model that had only been prompted for the role, while needing fewer follow-up turns from the customer. The paper carries two lessons. Smarter information gathering means less repetition at the handoff, and a smaller model trained for the job can outperform a far bigger one. That second lesson is a cost question you can answer with data. On AgentPMT, a team can run the same customer-service workflow with two different models on the run, price, and compare test bench, read the itemized per-token and per-tool-call cost of each run, and pick the cheapest model that clears the quality bar. **Book the fallback automatically.** When a person cannot pick up within 15 minutes, the law's alternative is a specific appointment within one business day. That is a scheduling job an AI agent handles well. On AgentPMT, a team can hand that job to the [Booking & Scheduling Agent](https://www.agentpmt.com/agents/booking-scheduling-agent), which checks Google Calendar availability and books the slot. From there, the team's AI agent can text the customer a confirmation through the Twilio Messaging agent tool and place the callback through the [Twilio Voice](https://www.agentpmt.com/marketplace/twilio-voice) agent tool when the time comes. For decisions that should not run on autopilot, like a refund above a set amount, a team can drop a human-in-the-loop checkpoint into the same workflow so the agent pauses until a staff member approves. We have written about [designing approval checkpoints that stay meaningful](https://www.agentpmt.com/articles/the-approval-workflow-nobody-wants-to-design-and-why-it-s-the-most-important-thing-you-ll-ship-this-quarter). **Keep the receipts.** Good faith is the legal standard, and a record is what shows it. Timestamp when the customer asked for a person, when they were connected, and when an appointment was offered and kept. [AgentPMT's audit logs](https://www.agentpmt.com/marketplace/agentpmt-audit-logs) record every action an agent takes through the platform, with the full request and response, so the moment the agent booked the appointment or sent the confirmation is on the record with a time attached. That record does not make a business compliant on its own. It gives the people answerable for compliance something concrete to review. Voice agents carry one more item for the checklist. CX Today reports proposed class actions alleging that Walmart's AI-powered customer service phone system and Lowe's customer service lines collected callers' voiceprints without the notice and written consent Illinois' Biometric Information Privacy Act requires. These are allegations, and neither company has been found liable. The practical step for any team running automated customer service tools on the phone is to know whether the system builds speaker templates or authenticates callers by voice, and to make sure the disclosures match what the system actually does. ## Holiday traffic first, then the clock starts The next three months are the busiest stretch of the shopping year, and retailers are heading into it with new assistants switched on across their automated online services. On Jan. 1, every customer service channel a covered business runs needs its exit in place. The work between now and then is mostly design: decide which assistants count, wire the request phrases, carry the context, automate the booking, and keep a record. The customer data and the operator data agree on one thing. Shoppers will use a brand's AI when it helps them finish faster, and brands that build it well are seeing bigger baskets. A handoff designed into the agent, with a person, an appointment, and a record ready the moment a customer asks, satisfies California and protects the trust that makes those assistants worth building. * * * ## Sources - California Law Gives Consumers the Right to Reach a Human Agent Instead of AI, Gadget Review - California AB 1609 Requires Large Businesses to Disclose AI Customer Service Bots and Provide Human Support, Fox Rothschild Privacy Compliance & Data Security - California Just Made It Easier to Get Past AI Customer Service Chatbots and Reach a Real Person, Law Commentary - AI Legislative Update: Oct. 2, 2026, Transparency Coalition - Customers say AI has made their service experience worse, CX Dive - Consumers favor AI shopping assistants from brands they already use and trust, CX Dive - Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents, arXiv - OpenAI Helps Albertsons Expand Its AI-Powered Grocery Offerings, PYMNTS - Groceryshop 2026 Takeaways: From AI Experimentation to Everyday Value, Clarkston Consulting - How Kroger, Giant Eagle and DoorDash are building their agentic AI assistants, Modern Retail - Voice AI Lawsuits Put Contact Centers on BIPA Watch, CX Today - Gap unveils AI-assisted shopping, three decades after its first website, Retail Dive