BRdata Analytics: Trend Decomposition

Grocers need to ensure that their shelves are stocked with the items their customers want. Maintaining a high price for an item where market demand is falling uses valuable inventory space, as well as shelf space, in which customers note that their needs aren’t being served. Maintaining a low price for an item where market demand is rising, or simply failing to offer that merchandise, risks lost profit. Both high prices and out-of-stocks nudge customers to look elsewhere. This presents an interesting, and difficult, problem: for which items is market demand rising or falling and by how much?

NGA Independent Insights: Discovery

BRdata offers subscribers of NGA Independent Insights the Trends report in partnership with the National Grocers Association to assist grocers in answering exactly this question. For example, imagine a grocer has noticed that the days are growing shorter and colder and they are wondering if they have everything they want in their seasonal candy section. By clicking “Trends” in the upper navigation bar, they would be presented with a screen showing the top trending Parent Categories (1234xxxx) within over 1,300 independent grocery stores across the nation.

NGA Independent Insights Landing Screenshot

Our grocer can see right in the top row that Parent Category 01780000 is described as “Candy Seasonal-Christmas”, is “Strongly Trending Up” in terms of units sold, has a “Neutral Trend” in terms of its price, has weekly sales in the last month that more than doubled its weekly sales over the last quarter (137% change), the number of stores supplying this data, and the most recent update. Now that our grocer has found the Parent Category that they were looking for, they simply click on the blue button labeled 01780000 in order to travel down to the Categories that make up that Parent Category.

NGA Independent Insights Categories Screenshot

Although the results grid is sorted by trend strength, our grocer may be interested in the Category where per-store sales are highest in the past month–so they click on the blue button labeled 01782010 in order to see the UPCs that make up that Category. After a few seconds, results are ready.

NGA Independent Insights UPCs Screenshot

Our grocer can now inspect the underlying UPCs that were making this Category trend up and see if they are missing any hot sellers. And look, there are some Lindt UPCs that this grocer’s store group does not currently carry (for example, rows 4 and 6). It may be the case that their warehouse doesn’t carry Lindt, but if they did, there could be some items worth trying out. Nonetheless, they have many Hershey’s brand items (selling at between 1 and 12 of their stores) but they are missing row 5, Hershey’s Miniatures Assortment. This is what we call “Discovery”–where NGA Independent Insights helps you to discover hot sellers you don’t currently carry.

Although this example worked within a Category whose sales trend was expected, I actually hid from you that the top trending Parent Category was “BatteryElectronics.” Once we see it, the story is obvious–we need batteries for all the holiday decorations, gadgets, and toys. Now, even if that didn’t surprise you, we also see “Water Softener Salt & Pellets” as well as “Deli Meat – Retail Ready” on the upswing. While the end-of-year trends are dominated by the holidays, every week there is a mix of expected and unexpected Categories and Items trending up. Some of our customers have even capitalized on the expected Categories by preparing earlier than ever for seasonal sales and highlighting trends with Department Managers, procurement, and others.

NGA Independent Insights: Benchmarking

Another way that you might be able to use the “Trends” report is to understand how you’re stacking up against peers, noting where you are far ahead or far behind as ways to think about your store group’s performance, identity, as well as market. One way to do this is by adjusting the filters used on the landing page of the “Trends” report. A key filter is the dropdown box labeled “Filter Type” that can be set to:

  • State
  • Census Division
  • Census Region
  • Population Density
  • Place Size

That is, there are three geographic filters and two demographic filters. Imagine that our grocer is in the Pacific Census Division–they simply choose “By Census Division” in the “Filter Type” dropdown and then choose “Pacific” in the “Secondary Filter” dropdown.

NGA Independent Insights Parent Categories Pacific Screenshot

Although “Candy Seasonal-Christmas” is still among the top trending Parent Categories, this definitely looks different from our national picture above. Here “Skin care” and “Hardware” are far outside their typical sales ranges–note that percentage change does not define a trend and that a small percentage change is unusual for some Categories while a much larger percentage change is required for others to be considered “Strongly Trending.”

NGA Independent Insights: Opportunities

Lastly, by presenting the trend in units sold alongside the trend in price, we think that NGA Independent Insights enables our grocer to identify opportunities to adjust prices, clear inventory, make new orders, etc. that accompany a shift in market demand. For example, if units sold are going up while price is neutral, this suggests a shift in demand (seasonal or otherwise) that may be worth responding to with changes in price, location, or other marketing in one’s own stores.

Non-seasonal Trends: Decomposition of Item Movement

One solution that BRdata’s Data Science team is playing around with, imagining it brings new value to our customers, is letting this grocer ask a slightly different question: which are the most powerful trends among independents, ignoring all the seasonal highs and lows that we expected? We’re also going to add in a few other twists. Instead of covering millions of UPCs, we’re going to focus on the top 5,000 UPCs in each of nine Census Divisions. This lets us reduce the computational resources required while still focusing on UPCs most likely to impact the bottom line.

Now, how can we ignore the seasonal highs and lows? We will decompose each UPC’s movement history into three parts, where raw movement is the sum of:

  1. Non-seasonal Trend
  2. Seasonal trend
  3. Other

Decomposing Movement to Separate Seasonal and Non-Seasonal Changes

Decomposing Movement to Separate Seasonal and Non-Seasonal Changes

For example, imagine that you’re making winter orders for ice cream and you’re trying to decide how much freezer space to offer two different brands of vanilla: “Bryrs Natural Vanilla” (BNV) and “Haagen Dazs Vanilla 28oz” (HDV). They seem extremely similar to you as ice creams (although you note that the first is actually a 48 ounce container) but which should you purchase more of? You turn to the bird’s-eye view of your Census Division, let’s say “Middle Atlantic” to find out. By taking raw movement in the left-hand figure and decomposing it into the right-hand figure–and a little more magic from BRdata to get you closer to decision-critical data–you learn a few things (among “Middle Atlantic” independents):

  1. BNV has a non-seasonal trend in November roughly equal to HDV (although HDV is actually a bit higher)
  2. BNV is expected to sell fewer units in December [BNV: 1,628; HDV: 3,867]
  3. BNV is expected to sell 8% more units in December than it did in November (while HDV is expected to sell 20% fewer) [BNV: 1,761; HDV: 3,099]
  4. BNV is expected to sell volume (units * size) at a ratio of 1:1 with HDV (while it sold 0.7:1.0 in November)

Given that BNV is forecasted to increase it’s ratio to HDV, you go ahead and order more of BNV than usual, giving it closer to 1:1 shelf space. Since this is historical data, we can answer the question: how did this decision pan out? While we have of course chosen our specific example for illustrative purposes, we’ll highlight that the average forecast reduces benchmark prediction error by ~15%. In this particular case, BNV sold 1,657 units (+2% over November) while HDV sold 3,232 (-16% over November). If you had used November’s numbers (BNV: 1,628; HDV: 3,867), you would have given BNV 0.7x the shelf space of HDV. Since the actual December numbers were 0.9x by volume, your decision to approach 1:1 shelf space may have paid off in your Store. Of course, we don’t yet know how this decision would fare in January and February.

This type of bird’s-eye view is also available at the level of Categories (12345678) and Parent Categories (1234xxxx), allowing you to quickly spot market trends that have been broken down into their component parts: non-seasonal changes in movement, seasonal changes, and other changes. Of course, if you’re interested in learning more (about this bird’s-eye view, or even applying this technique to your own Stores), don’t hesitate to reach out to us.