Atlanta – The old adage that we should learn from our mistakes is uttered by many of us, but John Buccola, CIO of Wash Multifamily Laundry Systems, actually has quantifiable data to prove that he’s really doing it.
In business since 1947 in the U.S. and expanding to Canada in 2013 with the acquisition of Coinamatic, the country’s largest laundry service provider, Wash has been experimenting for a long time to answer what seems like a simple question: What’s the right price to charge for a load of laundry? It turns out there’s no one answer, but each location has its own right price.
![](https://i.itworldcanada.com/wp-content/uploads/2015/03/image1-e1426700373565.jpg)
At first, Wash was trying what Buccola calls a “shotgun” approach, guessing based on human intuition. Then it started collecting data on how those price changes would affect the locations. Was revenue going up? That set the stage for the big data project that Buccola is seeing grow today, using Microsoft Azure machine learning to help determine pricing.
“Let’s just call it what it is, we were making mistakes,” he says. “But that set up a whole trove of data to take and use as informative for our regression model data.”
Azure Machine Learning allows users to create data models using visual diagrams, similar to other logical schematic programs. It then can be put into use on the Azure cloud infrastructure, being fed from data sources via the Internet. A Machine Learning Marketplace is even available, currently with 36 plugins offering different data models that can be used for tasks such as sentiment analysis, recommend products that are often bought together to ecommerce customers, or rate a company’s “green score.”
For Buccola, using Azure means no more scouring through government spreadsheets to update his economic regression models. Wash takes into consideration census track level data, surveys laundry mats, and measures the data in its own rooms to determine pricing.
“It’s a hyper-local business,” he says. “If we overprice by a quarter, sometimes there’s a protest factor where a resident in the community says ‘screw this I’m taking my stuff elsewhere.'”
Based in El Segundo, Calif. Wash started its big data initiative in 2013, the same year that it made its expansion to Canada. Now after testing out its econometric regression models with Azure Machine Learning in markets closer to home, the results have been good enough that it will move the practices to Canada. Wash has boosted revenue to some of its stores by as much as 7.5 per cent – often by lowering the price for residents.
Since 70 per cent of its machines in Canada are paid for with an electronic card system, it’s possible price tweaking could be done for denominations even more exact than a quarter. The majority of Wash’s business is with apartment buildings. Wash installs the machines and splits the revenue with the building operator either 50-50 or 60-40 depending on the deal.
Wash still relies on human observation to help inform its pricing models, Buccola says. There are times when data just doesn’t tell the whole story.
“Sometimes you see something on a map and you say ‘there’s a laundry mat here, there’s a laundry mat there, this looks like it might be the right price,'” he says. “But you send someone out there’s and there’s a freeway overpass and people just don’t walk across it. It serves as a moat to that area.”
But that old tactic is now being combined with the new big data approach. Buccola is always expanding and honing his data models to predict the effects of pricing changes even more accurately. He’s identified the most important variable factors that determine pricing, which can differ for location, and fed that back into machine learning. He started with 60 different models to price the west coast portion of the 70,000 Wash locations across North America, and he’s already boosted that to 2,600 models.
Next, Buccola wants to move from adjusting basic pricing to a tiered approach for pricing. He’s acquiring new machines that support tiered pricing based on the time of day, day of the week, and cycles used. What might that look like?
“Saturday mornings are kind of crazy,” he says. “I don’t want to Uber surge this type of thing, but we can shift usage from peak periods by perhaps making Tuesday morning more attractive.”
He’s kidding, but if Buccola did try surge pricing and it turned out to be a mistake, at least we know that he’d learn from it.