EAT builds appetite for predictive analytics

Food retailer EAT uses predictive analytics from Blue Yonder to cut food waste by predicting future demand better

Soups, salads, hotpots, sandwiches, baked goods and coffee, all freshly prepared each day. That is EAT’s promise to its customers. But the question for the ‘on-the-go’ food retailer is how much to prepare. Too much risks discarding good food, while too little can mean running low on stock and disappointing customers.

Ramesh Bukka, head of IT at EAT, says his team has developed a bespoke “retail operating system” to help managers to control stock and cash using data direct from the electronic point-of-sale system. On top of this, they built an SQL-based system that employed linear algorithms to forecast the replenishment stock necessary for the next four or five days.

But there was an enormous number of variables governing sales, including weather and customer behaviour, and the system was unable to account for all of these, says Bukka. Meanwhile, in the four years since the system was rolled out, the business had changed, offering hot food and creating weekly, daily and seasonal specials. Although some store managers were good at using the system to forecast demand, staff turnover is high at EAT, and they might quickly move on.

EAT was founded in 1996, with its first shop on London's Villiers Street. It now operates 114 stores across the UK in the £3bn specialist food and coffee ‘to go’ market. To improve forecasting for short-shelf life products, in 2014 EAT teamed up with Blue Yonder, a provider of cloud-based analytics software and services, for a pilot project to integrated Blue Yonder's Forward Demand product into its back-office systems.

Using neural network technology, Blue Yonder finds associative patterns of historic internal data, including sales, promotions, product type, store location, store size and other variables, together with records of external events, including weather, holidays, days of the week, special cultural and sporting events. By building up a picture of the correlations between these factors, the system can create more accurate and detailed forecasts of future demand.

For example, while some stores are affected by the local weather, stores in airports are not. But they are affected by passenger volumes and the cultural make-up of people passing through the terminal, says Bukka.  

“With Blue Yonder, we knew we were talking to people who had ‘been there and done that’,” he says. “They had very strong case study history. We didn’t want to work with a company that learned from us rather than us from them.

“When we did the proof of concept, they worked with ‘German’ accuracy and never compromised on data integrity, so at times it was tough for us. We all know that the more accurate the data is, the more accurate the output. It was a hard journey.” 

Hitting the right spot

After sending XML-based data describing more than 30 variables across 200 different products through a structured API over a secure channel, EAT found that Blue Yonder was “hitting the right spot”, says Bukka.

“All complexity was abstracted. We sent tons of data to Blue Yonder without worrying about what they do behind the screen. From an IT perspective, we did not want that complexity.”

But preparing the data took some time, he says. Where the business thought the data was consistent, analysis would reveal inconsistencies. Something as simple as a broken down fridge could cause anomalies in the data, says Bukka.

“It was worth spending the time on that. Once you have defined data and the APIs, everything is automated,” he adds.

After a successful pilot, EAT has rolled out Blue Yonder across all of its business. Success is being measured across a range of key performance indicators, one of which, food waste, has seen a 14% reduction.

Although the system has been well accepted by its employees, EAT has had to emphasise the importance of data quality. It identified that some data from outlets was not accurate, because shop-floor staff can identify the wrong product passing through the till. 

Accurate picture of stock

It might not seem important to mix up two different items with the same price, but the difference can be significant when it comes to building up an accurate picture of the stock that needs to be ordered. “That is the journey we have to go on,” says Bukka. “It is end users you cannot automate – you can only try to persuade them.”

Bukka says cloud computing is creating new scope for smaller businesses to exploit advanced analytics, as a computing model offers access to high-calibre data science and tools without major recruitment or capital investment. 

“This would be difficult for me to adopt in-house,” he says. “The cost of investment, maintenance, infrastructure and hardware would be too high. With SaaS, you pay less up front and get a hundred times more than you could afford on-premises. They are disrupting that industry and we get to leverage new technologies.” 

By investing time in data-cleansing and user education, EAT has been able to benefit from predictive analytics tools to help forecast future demand. In a highly competitive market, that could be a vital differentiator.

Read more about predictive analytics in retail

 

This was last published in August 2015

CW+

Features

Enjoy the benefits of CW+ membership, learn more and join.

Read more on Data quality management and governance

Join the conversation

2 comments

Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.

The is similar to the way films are produced. Predictive analysis determines audience interest, the varying response to different stars, even the best time of year to open the show.

It's gotten quite complex for making movies - a film on THIS subject, made for SOME amount of money, featuring certain STARS can be expected to make $$$ profit. The predictions have proven to be surprisingly accurate.
Cancel
This is a great way to do it. Look at historical use, model it, see if it is a bell curve, go to the 65% percentile or so for stocking, adjust. Nice. No need for "big data" when SQL will do.
Cancel

-ADS BY GOOGLE

SearchCIO

SearchSecurity

SearchNetworking

SearchDataCenter

SearchDataManagement

Close