Vending

Why 2 veteran vending operators support AI

A pair of vending industry veterans describe their cumbersome data gathering practices for trying to determine what product changes to make in their machines. Both believe artificial intelligence will make a difference in their efficiency and productivity.

Image courtesy of Hivery.

August 12, 2020 by Elliot Maras — Editor, Kiosk Marketplace & Vending Times

How well does modern vending technology allow operators to maximize their assets? Can artificial intelligence, which experts claim will allow organizations to make use of the increasing amounts of data, significantly improve a vending operation?

According to a pair of veterans — Jared Detwiler, vice president of operations, One Source Office Refreshments, Pottstown, Pennsylvania, and Marc Whitener, CEO, Refreshment Solutions, Norco, Louisiana — the status quo is insufficient.

While most U.S. vending operators would envy Detwiler's and Whitener's technology savvy, the two did not hesitate to describe their operational shortcomings during a webinar, "Putting data to work — using artificial intelligence to merchandise your vending machines," sponsored by the National Automatic Merchandising Association.

Sharyn Kolstad, business development director, Hivery, an Australia based artificial intelligence provider, moderated the session, along with Shantanu Pasari, a product director at Hivery.

The operators rate themselves

Kolstad asked the veteran operators to rate their proficiency on a scale of one to give, with five being the best.

Jetwiler said he'd give his company a two.

Whitener said he would give himself a one despite a lot of effort invested.

"We know that if we do this (achieve level five status), the opportunity is double our profitability," Whitener said.

The status quo

Both operators described their existing data gathering processes, which they agreed is very cumbersome.

"We have more data than we know what to do with," Whitener said. The problem is how to use the data they have.

"Sorting and parsing the data is very complex," Whitener said. "That's the part that we're weakest at."

Much of the discussion focused on metrics used to determine when to add new products to the machines.

Detwiler said it is usually necessary to look at a lot of reports to find even a small amount of helpful information. His company regularly reviews contribution margins, profitability levels, service visits per machine, item performance reports, spoilage reports and average service visits and collections to decide what products should go in. They consider the customer demographics — white or blue collar.

"It takes way too long to do all this," he said. One of the hardest things is to train someone to do it.

He said he believes sales per service could improve by 60% to 70%.

"We can't get that productivity through scheduling," he said. "We've taken the scheduling module as far as we can get… It really comes back to merchandising; resetting the machines."

Trying to report on the results by asset level is even harder than getting the information they need to make the changes, Detwiler said.

For Whitener, the number one factor is "out of stocks." The top sellers usually have out of stocks. They also look at average collections per service, gross profits per service, and individual turns per item.

He said he tracks his contribution margin per service compared to the cost per service on a weekly basis.

Staffing requirements

Asked about how their companies are staffed for machine merchandising, Whitener said his company has two dedicated two-man teams to merchandise machines and micro markets. There is also a back-end person that remerchandizes using the data at the back end. However, he said they are not very good at it, the data is very complex and very hard to manage to identify the right merchandising opportunity.

Detwiler said his company has a full-time scheduler and merchandiser who manages par levels and spoils. There is also a micro market manager who does the same thing. Detwiler also contributes to the merchandising mix for high profile machines.

Asked how they make sure what they recommend is executed, Whitener said he has a chief customer officer who has employees who manage groups of customers. The operations people who do the resets report to the operations team.

Detwiler said his company uses photo audits to verify planogramming is followed. They also monitor price discrepancy alerts to make sure pricing has changed.

Why they want AI

Both operators agreed AI can make a difference.

"With the help of artificial intelligence, we want to see this merchandising happen at least to a minimum of every 30 days or seven to 10 services of the machine if you're visiting quite often to make sure we are keeping up with the latest consumer trends," Detwiler said.

Kolstad said AI augments human decision making. She said it doesn't eliminate jobs, but helps people make the best possible decisions in a proactive way.

"AI allows us to manage large amounts of data easily," she said. "It allows us to identify an optimum solution." It also eliminates human bias.

The time spent analyzing can let humans apply logic, intuition and nuance — things the data may never know, she said, like the owner's favorite flavor that has to be in the machine.

Detwiler said a data driven approach would make life easier. Working with Hivery has already enabled the company to identify product restrictions for healthy-only locations. He said AI algorithms will continue to improve.

Whitener agreed, saying AI will make him far more productive. Merchandising decisions will be automated and driven by AI.

A decade ago he was at a NAMA event and a Mars Inc. rep was showing Mars data that was telemetry driven from machines nationwide. Out of that group of machines, the number one selling item was for the past 90 days was Cheez It, but in half the machines it didn't sell one item, even though it was in all machines.

"Half of the machines should not have had Cheez It in it at all because it didn't even sell one item," he said.

Also, at last year's NAMA, he saw a presentation by a beverage bottlers who used Hivery AI. One account, a girl's dormitory, had no no diet lemonade. After they put that product in the machine based on Hivery's suggestion, it became top selling item.

"They were simply using data form other machines around the U.S. or around the world," he said. "There is no way we would ever be able to figure that out."