Numa CEO Tasso Roumeliotis: The Dealership AI Operating System Is No Longer Optional

Numa's AI Operating System already runs on the exact principle its CEO, Tasso Roumeliotis, laid out in a recent Digital Dealer feature: the dealership tools that win from here forward won't just store what happened, they'll coordinate what happens next, across 1,300+ dealerships and more than 1 billion calls handled. Roumeliotis's argument, published July 20, 2026, is that the DMS, CRM, and scheduler were all built as systems of record, filing cabinets for a transaction after it's over, while the customer standing in front of an advisor today may have already priced the repair, checked the market value, and rehearsed the negotiation with ChatGPT before they got out of the car. This piece walks through that argument in depth, with the research and internal data behind each claim.

The Core Argument: Systems of Record vs. Systems of Action

Roumeliotis's piece for Digital Dealer opens with a scene every GM recognizes without needing it explained: a customer walks in already knowing the numbers, because the information that used to live behind the service desk now lives in their hand. His argument is that this shift exposes something structural about how dealership software has always been built. The DMS records the transaction, the CRM records the lead, the scheduler records the appointment, and the CSI system records the survey result once the experience is already over. Every one of those tools is a system of record: built to store what happened, not to act on what's happening.

What he argues comes next is a different category entirely: an AI operating system that watches the store in real time, understands what needs to happen next, and coordinates the work across people, systems, and customers, rather than waiting to be queried like a dashboard or a chatbot bolted onto the side of the workflow. The distinction matters because most of what breaks in a dealership's day-to-day communication, a call going to voicemail, a text answered without context, a repair order sitting disconnected from the conversation about it, isn't a tooling gap. It's a coordination gap, and no system of record was ever built to close it.

This is the same distinction Numa's own comparison of communication AI and workflow AI covers from the buyer's side: one half of the category interprets unscripted conversation, the other half executes deterministic, rule-based triggers, and a dealership evaluating either half in isolation misses why the two have to run on the same customer record to actually close the coordination gap Roumeliotis is describing.

Why Customers Have a "New Edge" Right Now

The premise that a customer might already know the repair estimate or the market price before speaking to anyone at the store isn't a rhetorical flourish. It's measurable, and the numbers are moving fast enough that they were stale within months of being published.

Ekho's 2026 AI Vehicle Research Study, based on a survey of verified in-market shoppers, found 30% of vehicle buyers now use an AI tool during their research, more than double the share using online marketplaces, and among those buyers, 68.4% used ChatGPT specifically, far ahead of every other assistant combined. The same research found buyers moving through a predictable pattern: open-ended exploration, then model comparisons, then price and reliability validation, then a local-purchase question, arriving at the dealership already validated and ready to transact rather than still deciding.

Separately, a late-2025 Cars.com survey cited in that same research found 44% of car shoppers had already used AI tools during their buying process, and 97% said AI would influence their purchase decision going forward.

McKinsey's 2026 Mobility Consumer Pulse survey, which drew responses from more than 20,000 mobility users across five countries, found AI-informed research rising fastest among Gen Z and millennial buyers specifically, the customers dealerships will be serving for the next two decades. The same survey found 32% of respondents planning to postpone their next vehicle purchase over affordability concerns, and 45% considering smaller vehicle categories than originally planned, meaning today's buyer isn't just more informed. They're more price-sensitive and more willing to walk, a combination that makes the negotiation-ready customer Roumeliotis describes the norm rather than the exception.

Real-Time CSI Replaces Post-Mortem CSI

The Digital Dealer piece makes its sharpest point through a specific example: a customer who's been waiting four hours for an update, whose technician notes are unclear, and whose advisor hasn't responded yet. Roumeliotis's framing is blunt about what that customer actually represents: "That customer is a CSI event waiting to happen."

That reframing, treating CSI as something to catch in the moment rather than measure after the fact, is the exact argument Numa's own research has made about the survey model's core weakness: J.D. Power's 2026 U.S. Customer Service Index Study found that when overall satisfaction hits 950 or higher, 86% of mass-market customers say they'll definitely return for paid service, but the survey producing that score arrives weeks after the visit, when the angriest customers are also the most likely to respond and the store has no ability to change what already happened.

Real-time CSI monitoring exists specifically to close that gap: reading sentiment from the live call or text, not the survey that shows up a month later, so a manager can intervene while the customer is still reachable rather than reading about the problem in a report.

The Digital Dealer piece's example resolves the way this actually works in practice: the system flags the delay, drafts a clear update, notifies the advisor, and escalates to a manager if the issue isn't moving, all before the customer has decided the store doesn't care. That's a fundamentally different operating discipline than a monthly CSI scorecard review, and it's the same discipline behind how Numa's missed-call recovery works: catching the gap in the moment it opens, not reconstructing it after the fact.

Where the DMS Actually Stops Keeping Up

Roumeliotis draws a clear line in the piece between what a DMS should keep doing and what it never should have been asked to do in the first place. The accounting ledger, manufacturer and warranty submission, regulatory and compliance records, payroll, and parts inventory accounting all genuinely need an authoritative system of record, and none of that is going anywhere. What he argues doesn't belong there is the workflow scaffolding built up around it: finance managers retyping the same data across screens, service writers translating a technician's shorthand into structured repair order fields by hand, and canned reports that answer a question days after a manager needed the answer.

That distinction matches what shows up in the data on dealership technology sprawl directly. Cox Automotive's AI Readiness in Auto Retail study, based on a survey of 537 franchise dealership leaders, found the average dealership already runs more than 40 different software systems, and the study's own leadership warned that additional tools only help when they connect into that existing infrastructure rather than becoming one more disconnected system fighting for the same customer record. Most BDC-DMS integrations underdeliver for exactly the reason Roumeliotis describes: they rely on overnight syncs or shallow data pulls rather than the live, bidirectional connection an AI operating system actually needs to act on what's happening instead of reporting on what already happened.

The research on enterprise AI more broadly backs up why this distinction, systems of record versus systems of action, isn't unique to dealerships. McKinsey's research on scaling agentic AI makes the same case for any organization: the tools that only store and report information are hitting a ceiling, while the ones that can act on live data inside defined boundaries are where the actual advantage is showing up.

What Moving First Actually Looks Like

Roumeliotis's closing argument is that every major technology shift splits operators into two groups: those who use new tools to speed up the old process, and those who redesign the process around what the new tools can actually do. He describes watching the same negotiation pattern play out repeatedly: a customer agrees to a price, checks it against ChatGPT or Claude overnight, and calls back the next morning with a different number in hand. The dealerships holding those deals, in his account, aren't the ones with the best counter-argument. They're the ones who can see the whole relationship in real time and respond before the customer has a reason to doubt the store at all.

This isn't a hypothetical distinction. Numa's own conversation with dealership leaders on navigating exactly this kind of change covers the leadership side of this same argument: AI adoption succeeds or stalls based on whether a GM redesigns the operation around the technology or just bolts a tool onto an unchanged process, the same split Roumeliotis is describing at the industry level. One CDJR store's Fixed Ops Director reported exactly this kind of shift in a single month after redesigning around real-time coordination rather than adding a tool on top of the old process: CSI moved from 820 to 981 between October and November, with fewer customer complaints and customers reporting faster callbacks on status updates. The technology didn't get better in that window. The process around it did.

Where This Leaves Your Dealership: The Edge Belongs to Whoever Coordinates Fastest

The uncomfortable part of Roumeliotis's argument for most GMs isn't that AI is coming. It's that the DMS, the system most stores have organized their entire operation around for decades, was never built to do the one thing that now matters most: coordinating a live conversation with a customer who already knows more than the store assumes they do. Numa was built around that exact gap, running voice, text, workflow triggers, and real-time CSI monitoring on one connected record specifically because none of those functions do their job in isolation. The dealerships that redesign their operation around a system that acts, rather than one that only records, are the ones who'll still be setting the pace when the next shift arrives. The ones waiting for the DMS to catch up will be negotiating from behind a customer who already made up their mind in the car.

Frequently Asked Questions

What does "AI operating system" mean for a dealership?

An AI operating system is software that actively watches a dealership's operations in real time and coordinates work across people, systems, and customers, rather than simply storing records for later reporting. Numa's CEO, Tasso Roumeliotis, has described the distinction as systems of record, like a DMS or CRM, versus systems of action that detect what needs to happen next and make it happen.

Why do customers have a "new edge" when they visit a dealership now?

Research from Ekho's 2026 AI Vehicle Research Study found 30% of vehicle buyers now use an AI tool during their research, with ChatGPT used by 68.4% of those buyers. McKinsey's 2026 Mobility Consumer Pulse survey found similar trends accelerating fastest among Gen Z and millennial buyers. Customers increasingly arrive at the dealership already knowing repair estimates, market pricing, and negotiation strategy before speaking to anyone at the store.

Is real-time CSI actually different from traditional CSI monitoring?

Yes. Traditional CSI relies on a post-visit survey that typically arrives weeks after the experience, at which point nothing can be done about what already happened, and the most frustrated customers are also the most likely to respond. Real-time CSI monitoring reads sentiment from live calls and texts as they happen, so a manager can intervene while the customer is still reachable rather than discovering the problem in a report.

Will an AI operating system replace the DMS entirely?

No. The core functions a DMS handles well, the accounting ledger, warranty and manufacturer submissions, regulatory compliance, payroll, and parts inventory accounting, genuinely require an authoritative system of record and aren't going away. What's shifting is the workflow scaffolding built up around the DMS: manual data entry, disconnected reporting, and the coordination work between departments that a DMS was never built to run.

How is Numa connected to the ideas in the Digital Dealer article?

Numa CEO Tasso Roumeliotis wrote the Digital Dealer piece this post is based on, and Numa's AI Operating System is built around the same systems-of-action principle described in it: voice, text, workflow triggers, and real-time CSI monitoring running on one connected customer record across 1,300+ dealerships, rather than a dashboard layered on top of an unchanged process.

See how Numa's AI Operating System runs the coordination a DMS was never built to handle. Talk to Numa.