What separates real AI from a fancy if-then statement? Quite a lot, it turns out. Most of what the restaurant industry is calling AI today doesn’t make the cut. In this episode, Matt Wampler sits down with Toby Malbec, VP of IT at Church’s Chicken and a 30-year veteran of restaurant, retail, and hospitality technology, to cut through the hype and get practical about what AI actually means for operators, how to evaluate and deploy it, and where it’s already delivering results inside the four walls.
What You’ll Hear in This Episode
Why Most Restaurant “AI” Is Just If-Then Statements Toby draws a clear line between descriptive analytics (what happened), predictive analytics (what will probably happen), and prescriptive analytics (here’s what you should do about it) and explains why most of what vendors are calling AI today barely clears the prescriptive bar. True AI, in his framing, synthesizes multiple data streams simultaneously to surface a single decision. Like knowing to send a manager home early because highway construction, dropping sales velocity, and incoming weather all converge at the same moment. The manager doesn’t need to know why. They just need the recommendation.
The Fragmented Tech Stack Problem and Why Restaurants Became Accidental Systems Integrators Toby traces the root of restaurant technology’s data problem back to the industry’s tight margins. Unlike retail, where a $5 tie sells for $180 and there’s budget to invest in standards, restaurants have always operated in a resource-constrained environment that made vendor interoperability an afterthought. The result: operators end up stitching together point solutions that don’t talk to each other, effectively becoming systems integrators for a business that’s supposed to be in the chicken business. He makes the case for RTN standards as the structural fix the industry still hasn’t fully committed to.
Kiosks: Why the Technology Isn’t the Problem, The Rollout Is Toby is direct about kiosks: the technology has been proven for decades (Southwest Airlines made the case 20 years ago), and the sales lift data is real at 15 to 20% on average. The reason kiosk deployments fail in restaurants has nothing to do with the hardware and everything to do with operators dropping them in without changing the operating model around them. He describes watching a restaurant with four empty, beautiful kiosks while a single cashier managed a line of 20 frustrated customers. A failure of change management, not technology.
Voice AI at the Drive-Through: What Church’s Learned from Testing Church’s has been actively evaluating drive-through voice AI, and Toby shares the most revealing things they found. Not the obvious technical challenges like accents and ambient noise, but the edge cases that only show up in real-world testing. A car full of kids all shouting orders at once. A guest who ends their order by saying “and that’ll do it for me.” A profanity gauge that reroutes the order to a human cashier when it detects a raised voice. He’s not dismissing the technology and he cites White Castle and Bojangles as genuine pioneers, but Church’s made the deliberate decision to let it keep baking before rolling it out at scale.
The Franchisee Data Problem Nobody Talks About One of the sharpest moments in the conversation is Toby’s take on data sharing between franchisors and franchisees. Church’s built a program that collects operational data from franchisees and gives it back to them as anonymized peer benchmarking dashboards, not to police them, but to show them how they stack up and where to improve. His point: the brands that treat data as something they extract from franchisees will lose trust. The brands that treat it as something they give back will earn the kind of relationship that actually drives multi-unit growth.
Predictive Analytics in Practice: Inventory, Labor, and Kitchen Production When Matt pushes Toby on where predictive analytics is actually delivering bottom-line value today, he identifies three areas: inventory ordering (where tribal knowledge and conservative hedging by experienced managers leads to chronic over-ordering and waste), labor scheduling, and kitchen production. That last one is particularly relevant for Church’s, where bone-in chicken takes 20 to 30 minutes to cook and running out is a day-altering operational event. He’s candid that what Church’s is doing today is still closer to predictive than true AI, and that the goal is to layer in weather, traffic, and customer sentiment data to push it further.
How to Actually Deploy Technology Without Becoming a Vendor’s Guinea Pig Toby closes with the most practical advice in the episode: engage your entire organization before selecting any technology, define what success looks like before anyone starts a demo, and never contract with a vendor after a single conversation. His rule of thumb is that a vendor should be scared when an operator buys their product after a one-hour pitch, because neither side actually knows what they’re getting into. That’s a useful gut check for any operator currently fielding sales calls.
Key Topics Covered
- The difference between descriptive, predictive, and prescriptive analytics
- Why restaurant tech fragmentation is a margin problem, not a technology problem
- RTN standards and the case for vendor interoperability
- Kiosk deployment failures and how to actually operationalize the technology
- Voice AI at the drive-through: real-world testing at Church’s Chicken
- Franchisee data sharing and why most franchisors get the dynamic wrong
- Buy vs. build decision-making for enterprise restaurant operators
- Kitchen production forecasting for time-intensive proteins
- How to vet vendors using AI before signing a contract
- Why the GM of the future won’t need to know why – just what to do
Subscribe to the Restaurant AI Podcast on YouTube or Spotify
Follow the Restaurant AI Podcast on LinkedIn
Connect with Matt Wampler and ClearCOGS on LinkedIn
