1. BIA 674 - Supply Chain Analytics 11a. Retail Analytics – Assortment Planning

2. Introduction  Retail’s Assortment -> the set of products carried in each of each its stores  Must take into account strategic issues – whether products align with their brand  Must often choose among thousand of products a small subset  Might decide not to carry all products in all stores  Rhythm of assortment is different for different retailing segments

3. Introduction  At the strategic level  Decide the amount of resources, for example, shelf space and purchasing capital for each product category  Resembles capital budgeting  At the operational level  Drill down to the Stock Keeping Units (SKUs) that you carry in each category  In both cases, you consider whether and how much you localize your assortment: from single assortment for entire chain, to … unique assortment per store

4. Overall Approach  Follow the approach by Fisher & Raman: look at products as sets of attributes, and try to forecast sales based on the trends of the attributes  Define each SKU as a set of attributes  Use prior sales to forecast market share for each attribute value  Use this info to forecast demand for any potential SKU  Assumes that attribute demand patterns are independent of each other  Assumes that consumer preferences for attributes are stable over time  Assumes that there is substitution

5. Convenience Store Example *  The retailer carries 2 brands of ice cream: A and B  Each brand comes in 17 different flavors, and in 2 sizes: Single and family, so there is a total of 68 brand-size-flavor combinations (2 brands x 2 sizes x 17 flavors = 68) at a particular location  Retailer keeps only 39 SKUs  Prices varied slightly across the different flavors and the 2 brand within sizes, but an average price is assumed for single and family sizes per brand  Store-SKU sales over the last 6 months of 2005 are provided  Assume there is substitution within the flavor and the size, between the two brands *From Fisher M and Raman A: The New Science of Retailing, HBP, 2010

6. The Existing Assortment Single SizeFamily Size Price ($)1.293.34 FlavorBrand ABrand BBrand ABrand BRevenue ($) 17,2463,10047245616,445.86 23,1821,4875513859,149.25 31,3981,3553314,656.91 43535703352,309.57 53,1001,4713987,225.91 61,5131,1551844,056.28 72,0341851393,706.02 82,9265632745,415.97 93,0093254,967.11 104,7803807,435.4 112,1699674,045.44 1222460489.36 131,5962,058.84 142,1622,788.98 154,0495,223.21 161,8272,356.83 17 100 334 Totals37,58614,6502,2872,28882,664.94 Sale Shares662644  17 flavors of ice cream  Comes in 2 packages: Single Size and family Size  Total: 68 possible SKUs  Store only 39 SKUs  Total Revenue = $82,665

7. How to find the optimal assortment?  Trial-and-error (guessing) approach:  Iteratively identify poorly performing SKUs (worst selling) and delete them from the assortment, while add randomly new potential offerings.  Over time you may have good chances to identify a productive assortment  A more elaborate and sophisticated approach:  Generate an accurate sales forecast for any assortment a store might offer and choose the assortment that maximize the revenue.

8. 4-Step Approach  Step 1: Estimate the shares of the 4 brand-sizes  Can you use instead the sales shares? NO, because these shares has been influenced by the current assortment, while customers in many cases they were forced to substitute. E.g. take flavor 3: how do you know that all 331 units of Brand B are truly for B and do not include some substitution for A? (note: A is not available in Family size for this flavor)  Select a subset of the sales data where all 4 brand- sizes are offered, and based on them estimate the demand shares. Only flavors 1 and 2 qualify for this purpose

9. Estimate shares of the 4 brand-sizes Single SizeFamily Size FlavorBrand ABrand BBrand ABrand B 17,2463,100472456 23,1821,487551385 Total10,4284,5871,023841 Demand Share 61.78%27.18%6.06%4.98% Observe that the demand shares are different compared to the overall sales shares. Single Size ~ 89%Family Size ~ 11%

10. Step 2: Estimate the Substitution Rates (1)  Step 2: Use the brand-size demand shares estimates to estimate the demand for SKUs that are not offered.  Can you use them directly? For example, to say that since the demand for Brand B, flavor 3, family size is 331, then the demand for Brand A flavor 3 family size is expected at 397. No, because we ignore substitution!  How can we estimate the likelihood of substitution between brands?

11.  Select a subset of the sales data those flavors where both brands are offered in single size, and only one brand is offered for the family size.  E.g. flavors 3, 4, 5, 6  Using the only one size, calculate total demand for all brand-sized, using the results of step 1  Estimate the demand and calculate the substitution rate Step 2: Estimate the Substitution Rate from A to B in Family Size

12.  Observed sales for these four flavors (3, 4, 5, 6): Single SizeFamily Size FlavorBrand ABrand BBrand ABrand B 31,3981,355331 4353570335 53,1001,471398 61,5131,155 184 Total6,3644,5511,248 Aggregate total10,915 Demand Share61.78%27.18%6.06%4.98% Size Share88.96%11.04% Total estimate*12,270 Estimated demand6,3644,551744611 Substitution demand637 Substitution rate85.61% =10,915/0.8896 =1,248-611 =637/744 Step 2: Estimate the Substitution Rate from A to B in Family Size

13.  Look at the sales of flavor 8: S8SA*=2926; S8SB=563; S8FA=274; S8FB=0  Total demand = TD8= (2926+563)/0.8896 = 3922  Estimated demand in family: D8FA=(3922)(6.06%)=238 and D8FB=(3922)(4.98%)=195  Since observed sales S8FA=274, and estimated D8FA=238  Substitution demand for flavor 8, family size, from B to A = 274-238 = 36  This means that 36 out of an estimated 195 demand for B moved to A  Therefore, SR8FBA = 36/195=18.5% Step 2: Estimate the other Substitution Rates (Family: B  A) *SIJK = sales of flavor I, size J, brand K DIJK = estimated demand for flavor I, size J, brand K

14.  Look at flavor 7: S7SA=2034; S7SB=0; S7FA=185; S7FB=139  Total demand = TD7 = (185+139)/0.1104 = 2935  Estimated demand in Single: D7SA=(2935)(61.78%)=1813 and D7SB=(2935)(27.18%)=798  Since observed sales S7SA=2034, and estimated D7SA=1813,  Substitution demand for flavor 7, single size from B to A = 2034- 1813= 221  This means that 221out of an estimated 798 demand for B moved to A  Therefore, SR7SBA = 221/798=27.7% Step 2: Estimate the other Substitution Rates (Single: B  A)

15.  Look at flavor 12: S12SA=0; S12SB=224; S12FA=0; S12FB=60  No full data for Family Size … but:  Sales of Family size B includes 4.98% of its “own” demand PLUS 85.61% “substitution” demand from the 6.06% of FA  60 = [4.98% + (6.06%)(85.61%)]x(Total Demand)  Therefore, Total Demand = 591 units  Estimated demand in Single: D12SA=(591)(61.78%)=365 and D12SB=(591)(27.18%)=161  Since observed sales S12SB=224, and estimated D12SB=161  Substitution demand for A = 224-161= 63  This means that 63 out of 365 demand for A moved to B  Therefore, SR = 63/365=17.48% Step 2: Estimate the last Substitution Rate (Single: A  B)

16. Step 2: All Substitution Rates  Observe that the sorts of willingness to substitute have a big impact on assortment planning!  Customer that buy brand A are not loyal, 86% switched to brand B for family size Single ServeFamily Size To FromBrand ABrand BBrand ABrand B Brand A 17.48% 85.61% Brand B 27.77% 18.57%

17. Step 3: Estimate demand for “missing” SKUs  Given the cross-brand substitution frequencies and the demand sizes for each brand-size, the estimated demand for all missing brands, sizes and flavors can be calculated.

18. 1. Compute the total demand for each flavor (based on the current assortment) 2. Compute the fraction of potential demand captured, both from whom the a brand-size offered is their first choice and from customers for whom it is their second choice but they are willing to substitute. 7% for Flavor 17 comes from the 6,06% of customers who had Brand A family size buttercream their first choice + 1% who wanted Brand B but substitute for Brand A (18,57% of 4,98% =1%) Captured only 7% of demand Step 3: Estimate demand for “missing” SKUs

19.  Estimate the total demand for each flavor (assuming that all flavors are offered) as total sales divided by share captured (e.g. 1429(100/0.07)). Then…  Take the multiple total demand by brand-size shares to obtain demand estimates for each SKU (e.g. 0.62*1429=886 for Brand A single size flavor 17) Step 3: Estimate demand for “missing” SKUs

20. Demand Estimates Share of demand capturedSingle SizeFamily SizeTotal Direct Sub- stitutionOverall Total demandBrand ABrand BBrand ABrand B demand estimate 11,274100.00%0.00%100%11,2747,2463,10047245611,274 5,605100.00%0.00%100%5,6053,1821,4875513855,605 3,08493.94%5.19%99%3,1111,3981,3551891553,097 1,25893.94%5.19%99%1,26935357077631,063 4,96993.94%5.19%99%5,0133,1001,4713042505,125 2,85293.94%5.19%99%2,8771,5131,1551741432,986 2,35872.82%7.55%80%2,9341,8137971851392,934 3,76395.02%0.93%96%3,9222,9265632381953,922 3,33467.84%8.47%76%4,3692,6991,1872652184,369 5,16067.84%8.47%76%6,7624,1771,8384103376,762 3,13688.96%0.00%89%3,5252,1789582141763,525 28432.16%15.99%48%5903641603629590 1,59627.18%10.80%38%4,2032,5971,1422552094,203 2,16227.18%10.80%38%5,6943,5181,5473452845,694 4,04961.78%7.55%69%5,8403,6081,5873542915,840 1,82761.78%7.55%69%2,6351,6287161601312,635 1006.06%0.93%7%1,43188438987711,431 56,811 80%71,05543,18420,0234,3143,53371,055 Step 3: Estimate demand for “missing” SKUs

21.  How we can improve the assortment?  We must find which flavors customers want most.  One approach is to look at the sales rank  If we look at the demand estimates that we just have derived, what do we observe? Look at vanilla/chocolate flavor… Step 4: Improve on the Assortment

22.  Objective: Maximize Revenue  Choose no more than 39 SKUs sequentially  Select the SKU with the maximum demand revenue plus any revenue from substitution demand from other SKUs (always take into account your prior choices)  You continue to pick SKUs that maximize the increase in the revenue until you reach the maximum allowable number of SKUs Step 4: The Greedy Heuristic

23.  From $82,665 (current) to …  $92,115 (proposed) Single SizeFamily Size Brand ABrand BBrand ABrand BRevenue ($) 7,2463,10047245616,446 3,1821,4875513859,149 1,3981,3553,551 0 3,1001,4712506,731 1,5131,1553,442 1,8132,338 2,9261954,427 2,6991,1872185,740 4,1771,83841033710,253 2,1789584,045 0 4,2031,1422097,595 3,5181,5472847,481 3,6081,5872917,674 1,6282,100 8841,141 44,07316,8271,4332,62592,115 Step 4: Implementing the Greedy Heuristic

24. Remarks  Choosing attributes, attributes values and possible substitution paths.  No substitution across functional attributes (e.g. shoe size)  Price and value: the closer the price of a product to consumer’s price/value point, the more likely to accept a substitution  Taste attributes may include such things as flavor, color and fabric type

25. Remarks  Limited data undermines accurate estimation  Choose carefully the level of aggregation  Group attribute values (e.g. size of packaging) that differ slightly and may be indistinguishable to the customers  The assortment currently offered, affects the sales, thus estimation of the demand for new products can be really hard!  Use sales of data of existing products that have at least one common attribute with the new product  Conjunction analysis – ask panel of potential customers and make pairwise comparisons with various product concepts

26. Remarks  Take in account high levels of stockouts (if any) when analyzing sales data  Besides revenue maximization, different objectives can also considered, such as maximizing gross margin, unit sales, percentage margin, or any metric of strategic importance  Decisions on different product facings can also be considered.