CarMax Put AI on Its Inbound Sales Calls. The Way It Did It Is the Real Story.

The largest used-car retailer just told everyone what its phones have been doing since May

On August 6, CarMax announced it has been working with Sierra - the AI agent company co-founded and led by Bret Taylor - on a new inbound call experience for its customers. The scale of the operator is the point: CarMax runs more than 255 stores with roughly 28,000 associates, and retailed about 780,000 used vehicles in the fiscal year that ended February 28, 2026 (CarMax, August 2026).

The detail worth reading twice is the timeline. The AI voice agents were deployed in May. CarMax ran them for roughly three months, watched the numbers, and only then went public - with something concrete to report: an increase in call resolution and a decline in the rate of unresolved calls. No launch-day hype, no promises about reinventing retail. A quiet rollout, then a results-first announcement.

For a dealership principal, GM, or BDC manager, that sequencing is the story - not because CarMax's exact stack is something a 60-car-a-month store can or should replicate, but because the way the company scoped, staged, and measured the deployment matches the discipline that has separated working dealership AI from shelfware all year.

The myth: AI answering the sales line is experimental technology no serious retailer would put in front of real customers. The data: CarMax has had AI voice agents on its inbound sales calls since May 2026 - across a business that retailed about 780,000 used vehicles last fiscal year - and reports call resolution went up while unresolved calls went down. Source: CarMax press release, August 6, 2026.

What CarMax turned on - and what it deliberately did not

Read the announcement closely and the scope is strikingly narrow. Sierra's voice agents greet inbound sales callers, ask clarifying questions to understand what the caller actually needs, answer the most frequent questions directly - store hours, whether a vehicle is available - and get everyone else to the right associate faster, in CarMax's words, "regardless of call volume, time or time zone." Sierra CEO Bret Taylor described the customer-side goal the same way: no waiting on hold, no repeating yourself as you get transferred from one place to another.

Notice what is not on the list. The agent is not negotiating price, structuring financing, or closing deals. Shamim Mohammad, CarMax's chief information and technology officer, framed the division of labor plainly: answering the most frequent questions "consistently and on demand" is the machine's job, so associates can "provide knowledgeable guidance through the bigger decisions in the car shopping journey."

The next phase is just as deliberate. CarMax says it is working to add appointment management - scheduling appraisals, browsing visits, and test drives. That is the move from answering to booking, the same execution tier the rest of the industry has been converging on all year.

One more piece of context: this is not CarMax's first AI system. The company already runs Skye, a 24/7 virtual assistant on carmax.com, and was the first U.S. auto retailer with a car-shopping app in the ChatGPT app store. The phone is the newest layer of a stack, not a lark.

Source: CarMax announcement, August 6, 2026.

Read the sequencing like an operator

Strip away the brand names and CarMax made four choices any store can copy.

1. It started at the routing-and-FAQ tier. The highest-volume, most repetitive calls - hours, availability, "am I in the right place" - are where hold times and misroutes do the most damage and where automation risk is lowest. That is the tier CarMax automated first. Not coincidentally, it is the same first job we argued for in scoping an AI agent's task list.
2. It measured resolution, not activity. The two results CarMax chose to disclose - call resolution up, unresolved-call rate down - are operational outcomes. Not "calls handled," not "conversations had." If the AI answers instantly but callers still do not get what they need, those numbers do not move.
3. It ran quietly before it talked. Three months in production before a press release. The announcement is a consequence of the results, not a substitute for them.
4. It is expanding one adjacent tier at a time. Appointments come next only after the first tier held up - and the handoff to a human stays central to the design, a pattern we examined in how the best deployments handle the human handoff.

The same week, the money said the quiet part out loud

The CarMax news did not land in a vacuum. Within roughly five days of the announcement, the companies that sell and carry voice AI posted a run of numbers that describe where the category is headed.

Twilio, the communications infrastructure many phone tools are built on, reported its first $1.5 billion quarter and raised full-year guidance, citing demand from AI-native companies - while cautioning that AI revenue remains in "very early innings." Five9 landed a roughly $100 million, five-year contract the same quarter it launched its own voice AI agents, and raised its full-year AI growth outlook to at least 60%. And three separate voice-agent startups announced funding rounds within five days of one another.

For a dealer, the practical meaning is about vendor risk. A year ago, a fair question about any voice-AI vendor was whether the category would still exist at renewal time. That question is fading: this is now infrastructure with revenue, contracts, and balance sheets behind it. The question that remains is the harder, store-level one - which tool fits your operation, and can it prove it - the evaluation lens we laid out in judging dealership AI in the shakeout.

The honest counterweight: deploying is easy now. ROI still is not.

The same week produced the most useful caution in the pile. Auto Finance News ran a piece titled "Voice AI Is Easier to Deploy. Long-Term ROI Is Harder" (August 4) - authored, it should be said, by Prodigal, itself an AI servicing vendor, so read it as a practitioner's warning wrapped in marketing. Its description of the plateau rings true regardless: the predictable calls automate quickly, the team celebrates, and months later operating costs have barely moved. The edge cases pile up exactly where the value was supposed to be - background noise garbling a verification, a caller switching languages mid-authentication, the unpredictable minority of calls that still need a person and now reach one with less context.

The lesson is not that voice AI does not pay. It is that the demo is no longer the differentiator - integration and measurement are. Which makes CarMax's disclosure choice the tell worth copying: resolution rate and unresolved-call rate are numbers a single store can baseline before a pilot and audit every week after. A pilot without a baseline cannot visibly plateau - which sounds like a blessing and is actually the trap.

What this resets for a franchised dealer's phone room

Two things changed for stores that compete with CarMax for used-car customers - which, at roughly 780,000 retail units a year, is most of them.

The reference experience moved. When the largest used-car retailer answers on the first ring, sorts callers conversationally, and hands off without repeats, that becomes the standard a used-car shopper carries into the next call. The next call is often a franchised store, where independent benchmarks still show the average caller waiting about 73 seconds on hold before reaching a human, and 42.7% of qualified sales leads being mishandled (Foureyes 2026 benchmarks). We have put numbers on that gap in what missed calls really cost and why speed to lead decides deals.

The backdrop keeps rewarding capture over spend. Cox Automotive's Manheim Used Vehicle Value Index closed July at 210 - down 1.4% from June on a seasonally adjusted basis and about 2.5% below its March peak - with Cox describing genuine seasonal depreciation doing more of the work as the wholesale market normalizes. New-vehicle sales are still tracking toward roughly 15.8 million units for the year, down about 2.9% from 2025. Flat demand plus normalizing used values is a market where the margin is made on execution: every inbound call is demand the store has already paid for.

The copyable playbook, compressed: automate the routing-and-FAQ tier first; instrument answer rate, resolution rate, and unresolved calls before the pilot starts; add appointment booking second; keep people on trades, financing, and closes; and insist every AI-handled interaction writes back to the CRM.

The measured takeaway

CarMax has advantages most stores do not - a large in-house technology organization, one brand, one process, and years of AI deployments behind it. "CarMax did it" is not, by itself, a reason to sign anything. But the direction is hard to argue with: the biggest, most data-driven retailer in used cars looked at its inbound sales line, decided hold times and misroutes were a solvable problem, solved the narrow version first, measured it for a quarter, and is only now expanding the scope. The technology to run the same play at a single store is better funded and more accessible than it has ever been. What is scarce is the discipline - scope, baseline, measure, then expand.

If you are working out what the first narrow, measurable job for an AI voice agent should be at your store - and how you would know within 90 days whether it worked - that is the conversation the team at Carbuki has with U.S. dealerships every day.

Sources

• CarMax, "CarMax Teams with Sierra to Enhance Inbound Sales Call Experience," August 6, 2026 (GlobeNewswire): globenewswire.com
• CarMax Investor Relations, same release: investors.carmax.com
• Investing.com, "Twilio Q2 2026 slides: revenue growth accelerates, guidance raised," August 2026: investing.com
• Pulse 2.0, "Five9 Lands Approximately $100 Million Contract As Voice AI Agents Drive Contact Center Push," August 2026: pulse2.com
• CX Today, "Five9 $100M CCaaS Deal: Fortune 100 Financial Services Win," August 2026: cxtoday.com
• Tech Times, "HappyRobot Raises $150M" (enterprise AI agents, $1.2B valuation), August 4, 2026: techtimes.com
• TechCrunch, "Omilia raises $67M to scale its customer support platform," August 6, 2026: techcrunch.com
• AI Insider, "AI voice agent startup Encore AI closes $30M Series A led by Team8," August 7, 2026: theaiinsider.tech
• Auto Finance News (authored by Prodigal), "Voice AI Is Easier to Deploy. Long-Term ROI Is Harder," August 4, 2026: autofinancenews.net
• CBT News, "Manheim used-vehicle prices fall 1.4% in July as wholesale market normalizes," August 10, 2026: cbtnews.com
• Cox Automotive, Manheim Used Vehicle Value Index, July 2026: coxautoinc.com
• Foureyes, "2026 Automotive Dealer Benchmarks Report": foureyes.io