Putting predictive analytics on the menu at EAT.
- Summary:
- Better forecasting through predictive analytics cuts waste by potentially 14% at UK food-to-go retail chain.
I know what success looks like, but I have absolutely no idea what [analytics provider] Blue Yonder is doing and I don’t want to.
Ignorance about the complexities of predictive analytics is definitely bliss for Strahan Wilson, chief finance officer at UK food-to-go chain EAT.
And so are the results. It’s still in the pilot phase, but applying Blue Yonder’s software-as-a-Service (SaaS) Forward Demand to predict demand for food and drink could potentially knock 14% off EAT.’s food wastage, as well help to manage staffing requirements more effectively.
Wilson just needs to set the right KPIs, and then the Blue Yonder team, many of whom are ex-CERN data specialists, do the hard number crunching.
Wilson has a half-hour weekly meeting with Blue Yonder, but apart from that it is a relatively “light touch” from his point of view and from an IT standpoint:
At this stage we’re very happy – all the hard work is being done on their side of the fence.
EAT. is a familiar high-street food brand in the UK, particularly in central London where most of its 115 stores and 1,500 staff are based.
Predicting demand for food and drink is a difficult art to perfect, as it is affected by external factors such as public holidays and the weather. And while the UK may not have the extremes of weather endured by other parts of the world, it can run through all the facets of the four seasons in one day.
It’s early days, but this new relationship with Blue Yonder is already eclipsing EAT.’s experiences of its previous forecasting tool. Wilson explains:
We did have an in-house forecasting tool built for us three years ago, but that became redundant about 18 months ago. We needed something more sophisticated and able to move with our business. As the business moves and competitive environment changes you need to change your models.
The old system had been built by consultants and as everyone knows, there’s “not inconsiderable costs” involved in hiring consultants. While the system was fit-for-purpose when it was built, the models used were cast in stone and EAT. lacked the IT skills to update the models internally. Although EAT. has a great IT resource, Wilson points out:
To maintain a model you need data scientists and that’s a very specific skill set requiring a degree of sophistication that is beyond us.
Despite having a background in statistics, Wilson felt he didn’t have the depth of experience to tell whether a data scientist was doing a good job or not and wouldn’t know how to motivate and manage them. And, he says:
It’s a very specific and specialist role and not core to our company.
The company had three choices. It could go down the traditional on-premise route, involving a hefty lump of capital upfront, or it could go for a cloud-based system that had a lower capex, but would still leave them with a model that needed maintenance. The third and preferred option was to outsource the system and the service to Blue Yonder.
Testing times
To prove Blue Yonder could deliver on its promises, EAT. set the company a test. It provided all the data Blue Yonder required on product sales, stores and other factors up to July 14th 2014. Blue Yonder’s task was to forecast demand for July to December – it was effectively a live forecast based on historical data. EAT. then compared Blue Yonder’s prediction s with their own forecast.
According to Wilson:
When we compared metrics, their worst-case scenario was still 14% better than our actual forecast. That meant payback of investment in about three-and-a-half months.
It was “a pretty robust test”, notes Wilson, and one that in his mind Blue Yonder comprehensively passed, so getting the go-ahead from the board was an easy sell.
Predictive analytics is helping the company in two main areas: reducing food waste and staff scheduling. Cutting food waste by 14% represents a colossal saving, but there’s a balance to be struck between waste and availability – zero waste may mean the company loses out on sales, because of lack of availability of products. Wilson notes:
It’s a two-dimensional matrix. There’s a trade-off between waste and availability. So we need to drive 14% improvement of waste without upsetting availability levels.
At this early stage, while the service is still being rolled-out, the emphasis is on ensuring the organization gets its investment back, but that will change over time, says Wilson, and the company will play around with the waste and availability ratio to find the optimal balance:
When we’re up and running, we’re happy to invest 14% in ways to improve availability because that’s an indicator of customer satisfaction.
Forecasting is complicated by the fact that hot food and cold food have different resource requirements. Cold food can be prepared beforehand and put in the fridge, making it “light on labour”. Both the production, preparation and servicing of hot food is “much more resource intensive” requiring more staff in-store. Wilson expands:
£1,000 of hot food takes more resources that £1,000 of cold food, so we have to schedule those requirements.
Armed with more accurate knowledge about anticipated buying habits, EAT. will be able to put extra staff in their stores when there is a high demand for hot food.
Being able to look at sales on a product level rather than by total sales will give EAT. a far richer insight into its own product lines, including identifying the best-selling products. Store managers used to spend 45 minutes to an hour a day making their orders, but admin can be reduced to just five minutes. This three week forecasting will also help the supply chain team immensely to order food more efficiently.
So beyond reducing waste, optimizing staffing levels and keeping customers happy, Wilson believes:
The ripple effects of this are significant.