Call the front desk of any decent-sized hotel and you’ll notice something: the person who answers rarely asks you to “press 1 for reservations, press 2 for housekeeping.” They just ask what you need. You say “the shower in 412 won’t drain,” and within a sentence or two they’ve already routed it, no menu tree, no repeating yourself for a transfer, no guessing which department “shower” falls under. It works because the person on the other end is trained to hear intent first and category second.
Now put yourself on hold with a big-box retailer’s store line, the automated one, the one that makes you say “existing order” out loud to a robot before it lets you say anything else. You already know which experience you’d rather have. Home Depot’s bet, as of 2026, is that a voice AI system can behave more like the hotel desk than the phone tree, and the company has enough of a live pilot running to say whether that bet is paying off.
What did Home Depot actually build?
In April 2026, Home Depot launched an AI-powered phone agent for store customer support, built on Google Cloud’s Gemini Enterprise for Customer Experience, and piloted it across 50 stores nationwide. The pitch, per EVP of Customer Experience Jordan Broggi, was a shift in posture: “We’re moving away from ‘please listen to these options’ and toward ‘how can I help?’”
Concretely, the system listens to the caller’s own words, determines intent (checking an order, confirming whether a product is in stock at that specific store, kicking off a service request) and then acts on it directly: sending a pre-filled shopping cart link by text, confirming inventory in real time, or initiating a Pro service request without a transfer. It also handles real-time translation for non-English speakers, which matters more than it might sound like it does given how much of Home Depot’s Pro and DIY customer base is multilingual. Critically, and this is worth sitting with for a second: every call still has an explicit path to a human associate. Home Depot didn’t build a wall between the caller and a person, it built a faster front door.
How well did the pilot actually work?
The pilot’s headline number, per Home Depot’s own release and corroborated by trade press coverage (Chain Store Age, CX Dive, StockTitan), is that the AI agent identifies why a customer is calling within about 10 seconds and resolves it roughly four times faster than the traditional phone-menu path. That’s a self-reported figure from the company and its vendor, so treat “4x” as directional rather than an audited benchmark, but the underlying claim (menu-free intent recognition beats “press 1, press 2, press 3”) is plausible on its face and consistent with how conversational voice AI has performed in other customer-service deployments over the past two years.
A more interesting, harder-to-fake data point: Home Depot’s 2026 American Customer Satisfaction Index (ACSI) score rose to 81, up a full point year over year, and the company specifically cited improvements in staffed pickup and AI call center support as contributing factors. ACSI is a third-party benchmark, not a Home Depot press release metric, so this is closer to independent corroboration than the “4x” figure is. Store associates in the pilot locations also reported higher job satisfaction, plausibly because fewer routine “is this in stock” calls means more time actually helping people on the floor.
The AI didn’t replace the phone tree with a smarter phone tree. It replaced the phone tree with a listener.
Where does this sit inside Home Depot’s broader AI push?
The phone agent isn’t a standalone experiment, it’s one piece of a Google Cloud partnership that dates back to 2015 and has deepened considerably. The same January 2026 announcement that detailed the phone pilot also expanded Magic Apron, Home Depot’s generative AI shopping assistant (originally launched in March 2025), giving it multimodal reasoning so it can process text, voice, and images to build a shopping cart autonomously, and tying it into real-time local store inventory and in-store wayfinding, not just a general product catalog.
Here’s the honest part, and it’s a useful comparison for the point about deployment discipline: Magic Apron didn’t roll out cleanly everywhere. Home Depot originally launched a professional-contractor version of Magic Apron alongside the consumer one, and pulled it offline after Pro customers found its questioning oversimplified, apparently tuned for a DIY homeowner’s mental model rather than a contractor’s. That’s a real, acknowledged misfire, not a hypothetical risk. It’s also a useful tell: the 50-store phone pilot, gated and measured before wider rollout, looks like Home Depot applying the lesson from the Magic Apron Pro stumble rather than repeating it.
| Traditional IVR phone menu | Home Depot’s Gemini Enterprise phone agent | |
|---|---|---|
| How it starts | ”Press 1 for orders, press 2 for…” | Open-ended: “How can I help?” |
| Intent recognition | Caller self-sorts into a category | AI infers intent from natural speech, ~10 seconds |
| Resolution speed | Baseline | ~4x faster (Home Depot / Google Cloud claim) |
| Action taken | Routes to a queue | Can directly build a cart, confirm inventory, start a service request |
| Multilingual support | Limited or none | Real-time translation |
| Human fallback | Eventually, after menu navigation | Always available, one step away |
| Verified so far | N/A | 50-store pilot; ACSI score (third-party) up 1 point YoY |
What does this cost Home Depot, and where does it still fall short?
Home Depot doesn’t disclose the phone agent’s cost or its Gemini Enterprise licensing terms, which is standard for a Google Cloud enterprise deal but does mean outside observers can’t independently size the investment against the “4x faster” payoff. Voice AI also has known, unglamorous failure modes that trade coverage on this rollout flagged directly: poor cell signal on a jobsite, heavy accents, and background noise all still trip up voice recognition in ways a human agent handles more gracefully by just asking the caller to repeat themselves without losing patience. And a phone agent that resolves calls faster doesn’t, by itself, fix the deeper problem CX Dive raised, that merchandising, loyalty, e-commerce, fulfillment, financial services, and Pro operations at Home Depot still largely make decisions independently of each other. A faster phone call is a real win. It is not the same thing as a connected customer experience end to end, and Home Depot hasn’t claimed otherwise.
Broader context worth having: this sits inside a $25 billion e-commerce business (Broggi’s remit, per Fortune’s June 2026 reporting) and a company whose net sales grew 4.8% in a fiscal quarter it itself described as facing economic headwinds. The phone pilot alone isn’t going to move a number that size, and Home Depot hasn’t tried to claim it does. It’s a service-quality and cost-to-serve play, not a revenue play, at least so far.
AI Suite and Open-Source Implementation
Home Depot’s actual stack, as far as it’s been made public, is more layered than “we use Gemini.” The customer-facing voice and shopping agents run on Google Cloud’s Gemini Enterprise for Customer Experience and the Gemini model family, grounded against Home Depot’s own product catalog and real-time per-store inventory data, which is what lets Magic Apron and the phone agent answer “is this in stock at my store” rather than just “is this a real product.” Separately, and less publicized, Home Depot has given office workers access to Microsoft Copilot for general productivity work, and its engineering org uses Anthropic’s Claude to accelerate internal software development, meaning the company is deliberately multi-vendor rather than betting its entire AI posture on one provider. That’s worth noting for any enterprise evaluating its own AI stack: Home Depot’s Google relationship is deep and eight-plus years old, but it isn’t exclusive.
An enterprise without Home Depot’s decade-long cloud partnership can replicate a rough version of the phone agent’s core mechanics with open-source tooling. Speech-to-text can run on OpenAI’s Whisper (fully open-weight, self-hostable), real-time voice orchestration, turn-taking, and the interrupt handling that makes a voice agent feel conversational rather than laggy can be built on Pipecat or Vocode, both open-source voice-agent frameworks designed specifically for this kind of low-latency pipeline. Intent classification and response generation can run on an open-weight model such as Llama 3.1 or Mistral, served locally through vLLM or Ollama, with the “is this actually in stock” grounding handled by a vector store like Qdrant or pgvector indexing the company’s own product and inventory data rather than relying on the model’s training knowledge.
None of that stack replicates Gemini Enterprise’s managed reliability, its enterprise support contract, or eight years of Google-side tuning against Home Depot’s specific traffic patterns, and it shouldn’t be treated as a drop-in substitute. The practical path for a mid-size enterprise: (1) start narrow, one call type or one store cluster, not a company-wide launch, mirroring Home Depot’s own 50-store gate; (2) ground every response in your own live data (inventory, order status, account records) rather than general model knowledge, since that’s what separates a genuinely useful agent from a chatbot that sounds confident and is wrong; (3) keep a one-step human handoff visible and easy at every stage, not buried in a submenu. And before any of that goes live against real customers: deploy in shadow mode first, running the AI agent silently alongside your existing phone tree and comparing its inferred intent against what a human agent actually resolved, and validate accuracy on that shadow data before the AI ever answers a real call. Open-source tooling can get you a working prototype in weeks. It cannot replicate the years of real traffic Home Depot used to tune and gate this rollout, and skipping that validation step is exactly how a voice-AI pilot turns into a customer-facing failure instead of an ACSI point.
Q: What AI does Home Depot use for customer service?
A: Home Depot’s phone support pilot runs on Google Cloud’s Gemini Enterprise for Customer Experience, launched in 50 stores in April 2026, with a nationwide rollout planned over the following year.
Q: How much faster is Home Depot’s AI phone agent than the old system?
A: Home Depot and Google Cloud report the AI agent identifies caller intent in about 10 seconds and resolves calls roughly four times faster than the previous menu-based phone system, a company-reported figure not independently audited.
Q: Is Home Depot replacing human customer service with AI?
A: No. Every AI-handled call retains a direct, always-available option to speak with a human associate; Home Depot has positioned the AI agent as a faster front door to service, not a replacement for it.