A few years ago, I stepped into a leadership role at a restaurant technology company and inherited a team that had been building a proprietary digital ordering platform from scratch. This was right around the time when best-in-class third-party solutions were emerging that did this exact thing, at scale, for a fraction of the cost.
The internal product was not working the way it was supposed to. There were interesting experiments happening. There were engineers doing creative things. But reviews from store operators were bad, the customer experience was inconsistent, and the P&L was painful to look at.
We were spending over a million dollars a year. The platform was generating about three thousand dollars in monthly profit at the corporate level.
The question I kept asking was simple: what is the actual return on all of this, on the business and for our customers?
The decision to move to proven external solutions was easy once I framed it that way. Better margins. Better franchisee profitability. A better experience for customers. The technology already existed. We just needed to stop trying to reinvent it and start measuring what actually mattered.
I'm watching the same pattern play out today across the restaurant & retail industry with AI.
Less about whether it's in house or outsourced.
More so about the bottom line impact of AI and how to measure it at the unit level.
The AI Problem No One Is Talking About
Most multi-unit brands I speak with are testing some form of AI right now. Drive thru menu boards, ordering assistants, labor scheduling tools, customer sentiment analysis. The experimentation is increasing, but based on operator sentiment and industry data, the ROI is not completely there.
There are a few reasons for this.
The first is that a significant amount of what gets sold as AI today is a wrapper on top of an existing tool. It adds a conversational interface or a prediction layer to something you already use. One operator I spoke with recently had deployed exactly this kind of solution and was paying a meaningful amount for it monthly. When I asked what it had changed in their business, they struggled to answer. The tool worked, technically. But it was not connected to anything that moved their P&L.
The second issue is that most operators are adding AI on top of existing workflows rather than using it to solve a specific, named problem. These are different things. If your line during a line rush is slow because of a specific bottleneck in how orders move from the point of sale to the kitchen, AI can help with that directly. But adding AI on top of your current process without identifying the constraint first will give you a marginal improvement at best. The bottleneck wins every time.
The third is the most common: operators are not tying AI investment to a clear bottom line result. They are measuring adoption. They are measuring usage. They are not measuring what it costs a franchisee or what it adds to same-store sales.
How to Think About It Instead
Before any technology decision, including AI, I think the most useful thing an operator can do is get clear on the fundamental levers of their business. Everything in your P&L traces back to four things.
Frequency
How often guests come back.
Check size
How much they spend when they do.
Throughput
How many guests you can serve in a given period.
Customer experience
The quality and consistency of the interaction that determines whether any of the above improve or erode over time.
At the brand level, this is the equation. Every problem worth solving, and every technology worth buying, maps back to one or more of those four levers.
The question is not whether AI can help your business. The question is which specific lever it moves, by how much, and how quickly.
Once you have identified where your biggest opportunities sit against those four levers, you can start evaluating solutions with a clear framework. Here is how I would approach it as a CEO / COO / CFO.
Quantify the opportunity before you start. Identify the specific metric you are trying to move. Labor cost as a percentage of sales. Upsell attach rate. Guest return frequency. Put a dollar figure on what a meaningful improvement would mean at the unit level, and then multiply it across your brand. This is the number you are solving for. If a technology cannot be traced back to it, it is probably not the right investment.
Set clear criteria with any AI partner before the pilot begins. What does success look like, specifically, at the end of 30 or 60 days? What data will you use to measure it? Who owns the decision on whether to proceed? These questions sound obvious, but most pilots end without clear answers to any of them. That is how you end up spending another quarter evaluating something that should have been a clean yes or no.
Tie it directly to the income statement. Not to usage metrics. Not to adoption scores. To same-store sales growth and unit-level profitability. If a solution improves upsell attach rate by 30% but it does not show up anywhere in the four-wall P&L, you have a data problem or a deployment problem, and both are worth understanding before you scale.
Build the rollout plan before the pilot ends. This is the part that gets skipped most often. A successful pilot followed by a six-month procurement process is not a successful pilot. Before you begin, map out what a brand-wide rollout looks like operationally, what franchisee adoption looks like, and what it takes to get from three locations to fifty. That plan should exist on day one of the pilot, not day thirty.
What Good Looks Like
The best AI implementations I have seen in this industry share a few things in common. They start with a clearly defined problem, not a technology. They are measured against unit-level outcomes from day one. And the people running the business, not just the people buying the technology, can articulate exactly what changed and why.
A regional fast casual chain recently told me that upselling had always been inconsistent across their locations. They knew it was a problem. They had tried training programs. The numbers would improve for a few weeks and then drift back. When they started using real-time coaching at the point of interaction, coaching that happens during the shift rather than after it, their upsell rate saw a dramatic jump. That is a result you can put on a P&L.
The common thread is not the technology. It is the decision to start with a specific, quantifiable problem and demand a specific, quantifiable answer.
A Note on Where This Is Going
We are still early in what AI can do for physical businesses. Most of what exists today helps operators understand what happened. The more interesting question is what changes when a business can understand what is happening right now, in a specific location, during a specific shift, and act on it before the moment passes.
That shift from retrospective to real-time is where the meaningful P&L impact lives. Dashboards tell you what went wrong on Tuesday. Real-time intelligence tells you what to do about it before Tuesday is over. And real time actions allows you to capitalize on those opportunities with bottom line impact.
The brands that figure out how to connect those two things, insight and action, at scale and across locations, will have a durable operational advantage. The ones that continue adding AI experiments without a clear bottom-line mandate will have something far more common: a lot of interesting data and a P&L that does not reflect it.
The bar for AI in this industry should not be 'does it work.' It should be 'does it show up on the income statement.'
That was the right question for digital ordering five years ago. It is the right question for AI today.
Tanvir Bhangoo
Tanvir Bhangoo is the Chief Revenue Officer at Meadow AI, the intelligence and action layer for multi-unit restaurant and retail brands. He previously led enterprise growth at Toast and held tech leadership roles at Freshii and RBI.
See what an AI co-manager would find in your stores.
We'll walk your operation through where Meadow drives measurable revenue, margin, and execution gains.