How to predict your sales with weather forecast?

• Are consumers more inclined to shop under clear skies or when it’s raining?
  
• Is there a positive correlation between weather and sales?
  
• Can you cut costs or avoid losing sales using weather forecasts to better plan your staffing needs?
  
  

This is the type of question we recently answered for a Canadian retail business. For those seeking the quick answer, it depends. If you want to know more, be our guest and read on! As this is the first blog of a series about weather analytics, we’ll be diving more into details in the upcoming weeks.

Weather… just another data source?

Weather data cannot be linked to other data like any other information. It can and must be linked using time, but any linkage other than that is not a hard fact; the best we can get is a correlation or similar distribution between two measures. Moreover, weather data comes in many different units; typically, weather data tells us about temperature and precipitation, but personally I would say that wind speed, humidity and air quality also have an effect, at least when we talk about extreme values.

Also, correlations might not be of the simple, straightforward type. Let’s take the correlation between temperature and sales: if it’s really hot, people might enjoy the A/C of a large shopping center but avoid small shops on the streets in the neighbourhood, and it is not unlikely that we might observe the same behaviour if it’s freezing cold. Intuition (and most of the literature) says that if it’s just comfortable outside, we’re most inclined to shop, but will it be more or less than in those extreme scenarios? Will there be a huge difference between 21 and 23 degrees Celsius? Do temperature and weather influence sales of specific product categories? Does swimwear sell in rainy weather? How would a rebate on those items during a "rainy day" affect sales?

By linking your sales data to weather data, you create a lot of questions before answering them…

So what’s the answer?

We performed a thorough data profiling to get an idea of the data we’re dealing with, and we could confirm most of the known patterns in retail. But it’s actually what doesn’t conform to common knowledge that adds value. Here are some of the most striking conclusions:

• First and foremost, weather is a local phenomenon. What seems to be a no-brainer at first sight is more important than it seems as it forces you to break down the weather data to the lowest geographic granularity available. Not only does the weather differ between St. John’s and Victoria, but -5 degrees Celsius is one thing in Vancouver and something completely different in Quebec City. As tempting as it is, you’ll hardly succeed in linking general weather data to the local way of dealing with that weather.
  
• Then, shopping is a local phenomenon. Judgment calls like: “People shop more on the weekend”, can hold true for one store but not for another one. Yes, generally speaking, good weather has a positive influence on a buyer’s mood, as do poor conditions, but it depends on the type and location of the store.
  
• That being said, the effect of weather is predictable, but only on a local scale, taking into consideration as many other influencing factors as possible.
  
• Many phenomena can superimpose, compensate or even reverse the effect of weather. Never forget to account for events such as holidays, major sport events, construction and seasonal events such as Christmas or Mother’s Day, depending on the industry you’re operating in. What’s most likely going to show up as a peak in your sales timeline will most likely weaken your correlations as people will shop no matter what. One more tip: take a look at your own promotions to make sure that the higher sales in October are not mistakenly attributed to global warming.
  
• Don’t forget about online sales! Selling less in stores might very well be overcompensated by the online channel, even more so with a targeted promotion.

3 steps to prediction

For the technician, weather has become another data source. For the decision maker, it has become a manageable input that should no longer be ignored. On the road towards true integration of weather into your BI data, we see the following three steps:

1. Analytics, proving the existence (or absence) of actionable correlations;
   
2. Integration / Interrogation, making data a well-integrated component in your BI, and;
   
3. Prediction, which finally turns your weather-enriched BI into a true decision support system.
   
   

Stay tuned on this blog to read about these steps and other analytics soon!