How to turn weather chaos into cash

Remember when the weather was dull? Nowadays it’s borderline psychotic. This year we’ve shivered through snow, been sunburned by sunshine, and been gripped by a combination of fear and excitement as a rare hurricane headed for the UK.

When it comes to retail those emotions are magnified, because chaotic weather can have a big impact on the bottom line - and it doesn’t have to be a negative one. As a result, retailers are using increasingly sophisticated software to turn the weather into cash.

The logic is simple, if a retailer can all but guarantee a heatwave is coming, it can order 10 times as many barbecue lines as its nearby rival. Simple in theory, but in practice how easy is it for retailers to get an accurate steer? What forecasting services are on offer? And how far ahead can they look with reasonable accuracy?

Former BBC weatherman Michael Fish - who infamously reassured the UK that the 1987 hurricane, which destroyed great swathes of the country and claimed 22 lives, wasn’t going to happen - recently praised the increasing accuracy of modern-day forecasting: “A five day weather forecast today is as accurate as a one-day forecast was in my day. We were only able to do weather forecasts for one or two days ahead.”

And retailers can call on a rapidly expanding industry of number crunching analysts to look even further forward, using increasing levels of data to make weather predictions up to 15 days ahead.

“It couldn’t be done in my day. We were only able to do weather forecasts one or two days ahead”

Michael Fish, former BBC weatherman

“We analyse more than 8,500 categories derived from over 10 trillion sales transactions,” says Scott Bernhardt, president of Planalytics, which claims accuracy of “about 80%-85%” for “most product categories”. “We help retailers with everything from near-term decisions like replenishment, staffing and advertising, to longer-term forecasts that identify weeks and markets where demand is higher or lower.”

Longer-term forecasts can go up to 11 months into the future, he adds, but the predictions are basic and limited to whether demand will be “higher or lower”.

Andy Mason, business development manager for Metraweather, which provides information to “four of the top five” supermarkets, says “a few short years ago”, retailers were happy with an email saying “hot down south, cold up north”.

“With regards to snow, that can only be discussed in our five day forecasts, so it’s too early to say. However, the current outlook says the chance of the coldest scenario happening is between 20 and 25% and the chances that the period will fall into the warmest scenario is between 10 and 15%. So uncertainty is quite large, but below average temperatures are more likely than above (average max temp winter are about 6.6C and average min temps are about 0.9C).

However, “the most advanced retailers are demanding individual forecasts for several hundred store locations, in a streaming data format, to automatically drive sales prediction and supply chain systems.” Metraweather “consistently” exceeds or meets its targets to get within two degrees of the weather at any of those stores, he claims. And while predicting further into the future is still being worked on “our sweet spot of accuracy and confidence is still 14 days.”

The Met Office has been working with Sainsbury’s since April, delivering “up to 15-day” forecasts to help with “stock control and demand planning”.

Helen Smith, who heads the Met’s retail business, is keen to stress that it doesn’t just provide Sainsbury’s with temperature or precipitation data. “We describe the feel of the weather, how people will react to it, and how it might change over the course of the day,” she says. “It’s not just about maximum temperature. For instance, looking at the data might just tell you it’s going to be a hot day, but if it’s going to be overcast during the early afternoon, people may not want a BBQ.”

Sainsbury’s has spent a lot of time and energy understanding its weather sensitivities, she adds. “It’s become a strategic focus for them because we live in a culture that expects everything any day of the year.” And it is delighted with the results. A spokesman says the Met helped it record its “best-ever week of seasonal barbecue sales” with a “600% increase year-on-year”, including “200,000 bags of charcoal” that it had ready and waiting to flood in to its stores and buy.

Waitrose uses a different forecaster and believes it is “pretty reliable up to five days out and, at times, longer. They predicted snow in December 2010 before the Met Office did.”

Operations pore over the weather every morning to give a “general indication of challenges coming up,” says a spokeswoman, as well as “how we ensure we have enough of the right stock out there”. She adds: “The volumes going through our supply chain at times this summer were massive - as much as the Christmas before on some key weekends,” and Waitrose did “pretty well.”

How much does it cost?

The potential cost of using sophisticated analytics to predict the weather is dwarfed by what you can make, or save. “Forecast too high and excess inventory sits idle, forecast too low and sales are lost,” says Ron Menich, chief scientist at Predictix, which has just teamed up with Direct Wines, the largest online wine retailer in the UK, to predict demand. “Both are errors, but for categories with significant weather-induced demand, forecasting too low is by far the worst error. Empty freezers are an ugly sight.”

“Before weather intelligence, wastage averaged 10%. Post weather intelligence has slashed it to 1%, so the savings are enormous”

Andy Mason, Metraweather

But a “multimillion overstock that doesn’t sell and has to be written off,” is not to be sniffed at, adds Mason. “Before weather intelligence, wastage averaged 10%. Post-weather intelligence has slashed it to 1%, so the savings are enormous.”

The cost is also coming down, according to SAS UK & Ireland retail specialist Alex Fovargue. “I worked at Sainsbury’s in 2000 and we spent over £10m on hardware and software for our forecasting once or twice a week. It took us hours to run. Today I’ve been running forecasts on three years’ worth of supermarket data on software and hardware that cost less than £100,000.”

The latest techniques also mean micro adjustments can be made, with a fast-developing trend to “predict and change an order three or four times a day”. “What used to take overnight can now run in seconds. You can forecast on an afternoon and change a delivery the following morning.”

Fovargue also says supermarket store managers are routinely buzzing around carrying iPads (or a Hudl) and entering weather data (among other things) at a local store level. The data feeds straight back to a buyer. “That portability means we are now seeing de-centralisation of forecasting for the first time,” says Fovargue. “It’s a level of flexibility that didn’t exist before.”

And thank heavens for that variety. Say what you like, the weather is never dull. And for all the advances, it’s never entirely predictable.