Using statistical models to predict future sales

BrainTech, LLC: Services and Products


Currently offering one free analysis and summary report per company


Ongoing services includes tracking changing trends and estimating day-to-day sales


Sales Summary Report
 Home Online Tools Resources Log In 
Weather, Sales Trend demand forecasting
Return to the Home Page View a behavioral model built using regression

The following is an example summary report we produced for a client in May, 2006
To help protect the identity of this client, the product they sell has been replaced by *****

Analysis of sales figures you provided (for January 12th through May 15th, 2006) revealed several important factors that may have influenced daily sales in Seattle and Tacoma stores:

Daily high temperatures in relationship to sales 1a. Day-of-Week by Temperature (Interaction)
  • Weekend sales tended to increase as daytime high temperatures increased.
  • In contrast, mid-week sales tended to increase as low temperatures became less extreme.
  • The difference is subtle but important: Extremely cold morning temperatures on the weekend were not associated with fewer sales (as long as it eventually turned out to be a warm day). However, when daily low temperatures dropped during the week, mid-week sales remained low even if the daily high temperatures were relatively warm.
  • In summary, warm weather may have encouraged weekend shoppers to buy *****, whereas mid-week shoppers appeared to be more concerned about avoiding extremely low temperatures.  Further investigation confirmed that mid-week shoppers bought far fewer ***** when low temperatures approached or dropped below freezing.
  • This likely indicates market segmentation; your weekend consumer audience likely differs from your weekday consumer audience in both preference and behavior (see footnote *1).

1b. Practical Application of this Knowledge

  • You may already be aware of this market segmentation. If not, you may be able to use this information to increase sales by:
  • Planning different promotions on weekends and weekdays that specifically focus on the preferences of each consumer audience.
  • Modifying retail displays based on the day of week, [material omitted to protect client].
  • varying content of advertisements based on whether the content is likely to be viewed during the week or over the weekend. [further discussion omitted].

2a. Seasonal trend (progression of time)

  • The number of weekday sales increased steadily with each passing week.
  • In contrast, weekend sales also increased over time, but the amount of increase in weekend sales grew with each passing weekend.
  • We found no indication that the rates of increase in weekday or weekend sales were beginning to slow by May 15th, despite your indication that demand for your product usually peaks near the end of June.  However, we noticed a large effect around May 14th, likely due to Mother's Day.
  • Even after investigating several models excluding sales after May 10th, no indication of slowing sales were noted.

2b. Addressing your question and comment
  • You requested that we address the question of when your sales season really begins.
  • Using consistent, predictable demand for your product to define the beginning of the sales season, your suspicion was confirmed; buying patterns were fairly inconsistent throughout January and became more predictable during the second week of February despite the overall low number of sales that month (see footnote *2).

3. Precipitation / Day-of-Week / Temperature (Interaction)
  • Another important interaction involved precipitation and can explained as follows:
  • Regardless of the day-of-week, sales tended to be lower when it rained.
  • Sales tended to be dramatically lower on rainy weekends.
  • Warmer average daily temperatures tended to offset this negative effect on sales when it rained on weekdays.
  • However, warmer temperatures did not help offset the negative effect of rain on sales when it rained on a Saturday or Sunday.
  • This interaction accounted for the apparently paradoxical reversals in Saturday/Sunday buying patterns in February and March, including the extremely low sales on a rainy Saturday, March 18th followed by higher sales on a dry Sunday the 19th. The following weekend, sales resumed a more common pattern where Saturday sales exceeded Sunday sales.
  • This is another clear indication that you have two different consumer audiences: weekday and weekend shoppers.

4. Daily low temperature.
  • In general, cold temperatures were associated with low sales.

Wind speed in relationship to sales5. Average wind-speed.
  • Higher wind-speeds were strongly associated with fewer sales.
  • This could be an indication of general consumer behavior, the necessity to move your product indoors on windy days, or a combination of the two. In addition, wind speed is often a general indication of poor weather.

6. Stability of the temperature throughout the day.
  • Greater fluctuation of temperature throughout the day was associated with fewer sales.
  • Note that stability of temperature could be an indicator of general weather conditions, such as the motion of storm fronts, or unpredictable and constantly-changing weather.

In summary, using these four simple effects and two interactions, we were able to develop models from a limited set of sales data that predicted both future and past sales in the SeaTac area with impressive accuracy (see online figures 1, 2, and 3). Additional information, such as the accuracy of currently-used sales predictions, is needed in order to assess the potential benefit of predictive modeling services. It is also important to note that our source of recent historic weather data only included daily ground-related weather data from SeaTac airport.  We were unable to use SeaTac weather variables to predict sales in your retail location in Mount Vernon. When we attempted a brief analysis combining sales from geographically distant regions, more general weather indicators such as barometric pressure and wind direction began to dominate the model, indicating that the weather differs significantly between the regions.

(*1) We were unable to more thoroughly explore weekend vs. weekday shopper behavior because of limited data; only 18 weekends occurred between January 12th and May 15th. Using historic sales figures (or this year's complete sales figures after the season ends), we would be able to provide a more thorough analysis and comparison between weekend and weekday shopper preferences and behaviors.

(*2) Stating that your sales season begins in early February would be supported by your sales figures. However, to accurately estimate the beginning of next year's sales season would require analysis of sales figures during the same months from previous years.


Home Page 
Try it Now
Account
Services
Weather
Resources
Contact Us
Free business advertising Home | Privacy Policy | Contact Us  
Copyright © 2006 BrainTech, LLC. All Rights Reserved