Every time there’s a temptation to turn a new technology loose on real work, I ask two questions. Does it pencil? Does it solve real customer problems?
Those two questions have saved me over and over again, from production floor to C-suite.
Throughout my career and I’m guessing yours, technological change has been a constant. Right now, of course, the headlines are all about AI. The two questions still work, though, and the reason they matter more than ever is the blast radius of AI. By blast radius, I mean this. When you turn technology loose and you’re wrong, how far does the damage spread, and who does it land on? The two questions are how I measure things before I act. The first sizes the cost. The second tells me who pays if we get it wrong.
Let’s start with a look back at the dawn of the new century. In the early 2000s, the biggest retailers badly wanted an RFID tag on every product they carried. And by badly, I mean an RFID tag to track not just pallets but each unit on the shelf. Walmart stood up a giant task force. Walmart and P&G built a giant lab. I sat in meetings where I was told it was a foregone conclusion. We flew to Bentonville to savor the shared vision.
Sweeping change didn’t happen overnight, which allowed time to run the math. The tags were silicon, and the whole pitch rode on eventually getting RFID tags down to about half of one cent. But at first, based on the cost of raw materials alone, they were more like five cents each. That meant I’d be adding a 5-cent tag to a 75-cent bag of chips that had about a dime of margin. Pencil, meet eraser. No matter how creatively I tried, RFIDs in my world of CPG presented a cost substantially greater than even the most optimistic estimate of benefits.
Data processing was even worse. One of the math wizards on my team dug into what it would take to track and store the data for every unit moving through every Walmart. His calculations revealed that you’d need more computing capacity than the largest scientific computer of the day, the Stanford Linear Accelerator. So basically, we’d be adding costs and reducing margin without a viable tech stack to support operations.
RFID wasn’t all wrong. It later found a home in apparel and other product categories where margins can absorb the costs of a tag. So it found its spot. Just not for snacks, groceries and beverages at the start.
Years later, at SpartanNash, automated micro-fulfillment had its day. The pitch, as always, was efficiency. The idea was labor savings, but the full calculation could not ignore labor RATE. Before the micro-fulfillment center (MFC), online orders were picked by front-end clerks earning about $15/hour. At the MFC, orders were picked by warehouse pickers making about $22/hour. Those figures were set. Thus moving orders the MFC might run 40% more before any of the technology enhancements.
OK, OK … the automation was supposed to earn that back. So how much should we expect in new efficiencies? Maybe 25%. Wait…what? Spend 40% more to recover 25%? … no.
And that’s not the worst of it. The warehouse couldn’t pick all the deli and fresh foods a lot of people came to the store for. That meant an important part of the order was something MFC could not supply. So we’d need to drive a truck from the MFC to a store and top it off by hand.
The longer we looked at it, the more we realized it just wasn’t viable for us. Notably, this past January, Kroger paid Ocado $350 million to walk away from automated capacity it had committed to place in the Kroger fulfillment centers. Costly learnings, yes. Definitely a blast radius you can measure in cash. But now they know.
Which brings me to the present. There’s no disputing that AI is fast with the potential to scale both good and bad. McKinsey’s 2026 grocery report describes agents that can launch thousands of other agents to update critical data within seconds, fast enough, the report notes, to overload the systems underneath them. The same report sees agentic AI continually optimizing pricing, promotions and replenishment. More speed, more reach and more … whoa … what-just-happened.
That’s the part that should keep an operator up at night. Not whether the system will ever be wrong, because nearly all systems fail sometimes. It’s that the damage can compound slowly, and then all at once. The real question is, are we swift enough to tell wrong from right and stop it in time?
I’ve deliberately used the metaphor of pencils and simple math because noodling with numbers takes time. Not a massive amount, because easy addition, subtraction and percentages are often all you need. But that’s enough to test some what-ifs and also check your work.
At Frito-Lay, we once had a prototype automated packaging line that a task force had studied and tinkered with for years. The goals were familiar, efficiency and cost-savings, but in seven or eight years all the tech re-tooling for automation didn’t work…and thus hadn’t saved anything. Then a savvy operator named Dave got fed up and built his own skunkworks fix-it lab. Instead of trying to remake the whole line at once, he took it a piece at a time, worked with front line associates and made sure each fix paid for itself. Then he identified and moved onto the next piece. Then the next. By the time he finished, Dave had produced a viable new packaging process that ended up saving significant costs.
His plan worked for two reasons. He did it with the people doing the work, not to them. And he never made the customer pay for it. Our buyers and retail shoppers never noticed and never needed to. The product quality was not compromised, and it didn’t cost more. Dave’s fix earned an A+ because he aced the test on the two points that matter.
Here’s what you learn about the blast radius. Initially, you can measure it in dollars, the writedown or a check you have to write. But that’s not really where it lands. In the end, the cost to the company and the cost to the employee both end up as a cost to the customer. Those are the people who get hurt.
So that’s all there is to it. Make it work for the customer, and make it work for the people who do the work. Do those two things in tandem, and you can do amazing things. That’s how you go fast without breaking things that matter. Blast radius stopped, before it started.