The Complexities of Product Assortment Planning
Product assortment depends on location, the target market, and other factors. Many retailers use automated assortment optimization tools to plan their assortment, while some still follow the “trust your gut” path. The last may be based upon years of experience and familiarity with your customers, but it is not always reliable. Even with automated planning tools, only a few are capable of accurate predictions of future demand.
It is Not Just About What Sells
Even experienced large retail chains get it wrong sometimes, like Walmart. In the year 2008, based on a survey that showed customers wanted less clutter in the stores, Walmart eliminated 15% of their SKUs, and decluttered their shops. Only, customers were not rushing in, instead they were going away. This was because the customers wanted those eliminated SKUs even if they didn’t purchase them frequently, and when they weren’t available, they to other stores. Walmart soon brought back most of those eliminated SKUs to cater to customer’s one off requirements.
Merchandise assortment is not just about stocking up on the store’s top selling items. As the Walmart example showed, the slow movers might play a vital role in retaining customer loyalty.
A More Reliable Approach to Assortment Planning
In an article published in Harvard Business Review a few years back, the authors discussed the result of a research they conducted, in making assortment decisions more accurate.
They pointed out that most people think of specific attributes of the product they want to buy rather than a specific brand when they go shopping. So, they came up with a solution that factors in the product attributes as the main criteria.
The Assortment in a Tires Store
For their study, the researchers focused on a tire dealership. The manager informed them that the attributes that mattered most to their customers were the tire brand, mileage warranty, and tire size. The retailer offered many brands that were popular across the country and three house brands, in different quality levels. Within each, many mileage warranty and size options were available. The research team then decided to classify the mileages into Low, Medium, and High. The countrywide brands were regrouped under one title, National. The house brands were classified as House 1, House 2 and House 3 in descending quality levels.
Finally, the entire tire assortment was divided into 6 classifications in descending order of quality/milege – National High, National Medium, House 1 High, House 2 High, House 2 Medium, and House 3 Low.
Studying the sales data, they found that a few SKUs were not selling well, and they wanted to determine what the impact would be if changes were made. They wanted to see what percentage of customers would be likely to substitute one tire quality for another, if their first choice were not available.
For their study, the researchers focused on size F which was offered in four of their six categories. They determined that the total share of sales that these four classifications commanded was 87%. That meant that 13% of potential sales for Size F was untapped, as this was not offered in the other two categories.
The total demand for F was estimated by dividing the total sales of tire size F by the percentage of demand captured – 1,204/87%, which gave 1,384. Now, to determine the demand for size F in any classification, they just had to determine the share it commanded of total sales and apply that to total sales of size F. So, the forecast for size F in the store’s top seller, House 2 Medium, was 1384 x 57.5%, that is 796 units. Actual sales are 763 units. The difference is because the percentage of sales for House 2 Medium got distributed across a wider choice – House 2 Medium was offered in many different sizes, as was House 2 High. The other classifications had fewer size choices.
To adjust for the discrepancy, the researchers used an optimization tool which fed in trial values for sales shares for different SKUs, and used the previously mentioned formulas to estimate demand. This was done in an iterative process until the sum of the discrepancies for all the SKUs was at the lowest possible value.
This process provided the optimal forecast for the shares of the six classifications. Once this best-fit share forecast has been determined, then this can be used to forecast the demand for each SKU.
Predicting the Shift
Now came another challenge, finding the percentage of customers who would shift between the classifications. So, the researchers added three more parameters – percentage of customers who would be willing to shift one step up, those would shift one step down, and those who would shift to House 2 Medium from House 3 Low (The two top sellers). There are now nine parametres, three more added to the existing six brand-warranty combos.
Using the tool again to plug in trial values iteratively, they arrived at a few more conclusions that provided substitution percentages:
- 35% of customers would trade up to House 2 Medium if the item they want was not available in House 3 Low
- For the rest of the classifications, 2% would step up one level for a substitution, and 1% would trade down
Taking all this further the researchers found that House 3 Low sales was just 11% of total sales, but their estimates predicted a 69.6% share. The discrepancy occurred because the dealer offered only a few choices on this cheap category, believing that their customers would be willing to move up to the House 2 Medium for the choices not offered in House 3 Low. This meant that they were losing a potential 45% in revenue (Only 35% of regular House 3 Low buyers would switch up, the remaining 65% wouldn’t make a purchase if their choice was not available). The researchers also established that in locations where the average income was lower, the share of the cheapest category would be higher. The data presented these hard facts and it was clear that a decision had to be made based on this, as against intuition or gut feeling.
So, to determine the optimal assortment, the researchers decided to focus on increasing revenue rather than raising the profit margin. This is an important call for any business when it comes to assortment optimization. Going by the observation that prices of similar styles went down consistently for each lower level, they used this criteria to determine prices for new SKUs.
They then calculated the potential revenue of each SKU by multiplying its forecast sales in units by the retail price. Based on all this, the store could then decide on the assortment by first choosing the SKU that would generate the highest revenue then moving one down and so on until they hit the optimal number of SKUs for the entire chain or for each store.
This process concluded that 47 of the 105 SKUs could be replaced, most of these with SKUs in House 3 Low.
Assortment Optimization – The Trial
The retailer decided to try out these recommendations. They replaced 10 SKUs. After some time, the researchers studied the results and found that even this small change had resulted in an increase of revenue by 5.8% and the profit margin had also improved by 4.2%. The store had also made a small price adjustment based on these recommendations. They had slightly increased the price of their cheapest tire and decreased the price of the next more expensive choice, to prompt more customers to move up.
This research proved that a more scientific approach helps retailers see how each single change would affect the entire assortment and revenue. Thus, they can make more accurate predictions and plan their merchandise based on these for the best assortment optimization.