Measuring the customers’ selection/ grabbing behaviour at Jumbo Supermarkten

November 14, 2019 ECR, Dublin

Rob Broekmeulen, Karel van Donselaar Gijs Bastiaanssen, Frits Jonker Quiz: Potential

What can be the size of prize of knowing / having a very educated estimate of the % of customers buying the freshest item on the shelf?

• 0-2% waste reduction • 2-5% waste reduction • 5%-10% waste reduction • 10%-20% waste reduction • >20% waste reduction

+ additional benefits

Slide 1 Contents 2

• Short intro • Researchers and Jumbo Supermarkten • Customers’ grabbing behaviour • Method to measure grabbing behaviour • Results from tests at Jumbo Supermarkten • Roundtable discussion Intro: The research team

• Rob Broekmeulen and Karel van Donselaar, Eindhoven University of Technology, The Netherlands • Active in research on retail operations (>15 years) • Cooperation with chains via research and student projects • Examples: - Quantifying the improvement potential for fresh departments in 27 stores at 3 large retailers in EU (ECR Sell More, Waste Less project) - Implementation of Fresh Case Cover concept at Albert CZ (savings € 1 mln/year) - Internships at Jumbo, Jan Linders, , , , Delhaize, Makro a.o. on Demand forecasting during promotions, Higher OSA in bread departments, Added value of humans in ordering, etc.

[email protected] andSlide [email protected] 3 Intro: The research team

• Gijs Bastiaanssen • Student Master program ‘Operations Management and Logistics’, TU/e • Intern at Jumbo Supermarkten

• Frits Jonker • Supply Chain Developer at Jumbo

Slide 4 Intro: Jumbo Supermarkten

• 2nd largest supermarket chain in NL: more than 21% market share (2019) and more than 660 stores in The Netherlands, expanding in Belgium • Fast growing: total revenue of almost €8 billion (2018) versus €400 million (2002). • Multiple channels: Jumbo Supermarkets, restaurant chain La Place, Jumbo Foodmarket (a combination between a traditional supermarket and the possibility to buy freshly cooked food made by professional cooks), Jumbo City stores (small stores in urban areas (e.g. train stations, city centers)), Jumbo.com (online store) • Foundation of its success: the importance of the customers in all its core activities (‘7 certainties’) and the ‘Every Day Low Costs’ formula.

Slide 5 Intro: Store replenishment for fresh products at Jumbo

• Distribution structure fresh products: stores are delivered via • inventory in DC’s • cross-docking at DC (no inventory) • Direct delivery

• Currently expiration dates (ED’s) are known in DC at case pack level, and not known in store information systems

• Store replenishment proces: • Automatic order advices (AOA) are generated for each SKU/store combination. • Stores can accept the AOA or adjust the quantity. • Since ED’s of items in the stores are unknown, the replenishment logic cannot include the forecasted outdating in the order advice.

Slide 6 Intro: Customers’ selection/grabbing behaviour

• Retailers tend to put freshest items in the back of the shelf.

• Part of the customers buy the product at the front of the shelf, e.g. since they trust the retailer, consume the item in the near future, prefer more ripe products (banana’s) or they are in a hurry (First In First Out, FIFO) • Part of the customers buy the freshest product on the shelf, e.g. since they only shop once a week or they have ample time (elderly people) (Last In First Out, LIFO)

• Retailers currently do not know the customers’ selection behaviour at the SKU-store level

Slide 7 Method to measure grabbing behaviour

• How to find the fraction of FIFO customers ƒ̂ ?

• Assume the following data are available per SKU (store-item combination) per day: total inventory (so no info on quantity per expiration date available on the shelf), sales, deliveries, waste and the product shelf life.

• The fraction of FIFO customers ƒ̂ is estimated by • starting with an initial estimate (ƒ̂ =0.5), • estimating each day the quantities per batch (i.e. per expiration date) available on the shelf • updating the estimate for ƒ̂ on days when the FIFO and/or LIFO withdrawal can be identified uniquely.

Slide 8 How to estimate the quantities per batch

• The ages of potential batches on the shelf are known: the delivery date and the product shelf life determine the expiration date. • The quantities per batch are based on dead reckoning (best guess based on estimated fraction ƒ̂ ) Hereto, every day an update of the batch quantity is made: • In case a new delivery is made, the batch quantity is known from the delivery quantity. • The actual sales is divided in FIFO-sales and LIFO-sales (using ƒ̂ ), and then the FIFO sales are subtracted from the ‘oldest batch’-quantity and the LIFO sales from the ‘newest batch’-quantity. • In case the expiration date passed, the batch quantity is set to zero • If in any period actual waste differs from estimated waste then quantities of newer batches are updated (calibration procedure).

Slide 9 When to uniquely identify FIFO and/or LIFO withdrawal

• To find the conditions when it is possible to uniquely identify the fraction of FIFO customers on a given day, let’s consider some situations and decide per situation whether it is possible to determine the fraction of FIFO customers

• E.g. is it possible to determine %FIFO on a day where sales for an item- store-combination are zero?

Slide 10 When to uniquely identify FIFO and/or LIFO withdrawal

• To find the conditions when it is possible to uniquely identify the fraction of FIFO customers on a given day, let’s consider some situations and decide per situation whether it is possible to determine the fraction of FIFO customers

• E.g. is it possible to determine %FIFO on a day where sales for an item- store-combination are zero?

• Unknown, since demand was zero • -> to estimate %FIFO, only consider days with positive (>0) demand

Slide 11 When to uniquely identify FIFO and/or LIFO withdrawal

• Is it possible to determine %FIFO if the inventory for an item-store- combination is equal to zero at the end of the day?

• Example: if inventory at start of day was 5 old units and 3 fresh units, and sales were 8 units; %FIFO=?

Slide 12 When to uniquely identify FIFO and/or LIFO withdrawal

• Is it possible to determine %FIFO if the inventory for an item-store- combination is equal to zero at the end of the day?

• Unknown, since demand may have been larger than sales; • If inventory at start of day was 5 old units and 3 fresh units, and sales were 8 units, demand may have been from • 5 FIFO customers + 3 LIFO customers • OR 7 FIFO customers + 3 LIFO customers • OR 3 FIFO customers + 7 LIFO customers • OR…. (many options possible) -> to estimate %FIFO, only consider days with no stock-outs

Slide 13 Example 1: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: • 8 units with expiration date Nov 17 • Consumers bought 3 units on Nov 14.

• Is it possible to determine %FIFO on Nov 14=? • If so, what is %FIFO? • If not, why not?

Slide 14 Example 1: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: • 8 units with expiration date Nov 17 • Consumers bought 3 units on Nov 14.

• %FIFO on Nov 14=? • Unknown, since consumers had no choice between a fresh batch and an older batch (only 1 expiration date on the shelf) • -> to estimate %FIFO, there should be multiple batches on the shelf

Slide 15 Example 2: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: 6 units with expiration date Nov 17 and 2 units with expiration date Nov 19 • Consumers bought 3 units on Nov 14.

• Is it possible to determine %FIFO on Nov 14=? • If so, what is %FIFO? • If not, why not?

Slide 16 Example 2: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: 6 units with expiration date Nov 17 and 2 units with expiration date Nov 19 • Consumers bought 3 units on Nov 14.

• %FIFO on Nov 14=?

• Unknown from which batch(es) the 3 units were bought. • To know this, the quantity sold from oldest batch should be known. Therefore we only consider days when a batch expires.

Slide 17 When to uniquely identify FIFO and/or LIFO withdrawal

• To observe FIFO / LIFO withdrawal at least these 4 conditions should be satisfied: 1. there was positive demand during a day, 2. there was enough inventory (no stock-outs). 3. the customers had a choice between different batches, 4. the oldest batch expired at end of day.

Slide 18 Five potential scenario’s when 4 conditions are met

When the 4 conditions are met, 5 scenario’s are possible (see Figure). 1st scenario: there is no waste and demand > oldest inventory 2nd scenario: there is no waste and demand = oldest inventory 3rd scenario: there is waste and waste+demand = oldest inventory 4th scenario: there is waste (< oldest inventory) and waste+demand > oldest inventory 5th scenario: there is waste (= oldest inventory) and waste+demand > oldest inventory

Slide 19 Example 3: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: 6 units with expiration date Nov 17 and 2 units with expiration date Nov 14 • Consumers bought 3 units on Nov 14. • No waste at end of day (oldest batch expired)

• %FIFO on Nov 14=?

Slide 20 Example 3: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: 6 units with expiration date Nov 17 and 2 units with expiration date Nov 14 • Consumers bought 3 units on Nov 14. • No waste at end of day (oldest batch expired)

• %FIFO on Nov 14=? LIFO sales<= 3units (=total sales); since 6 fresh units were available all LIFO demand has been delivered from fresh units only. ->2 old units sold imply that these 2 old units are sold to FIFO customers. the 3rd unit sold is a fresh unit and it is unknown whether its demand was FIFO or LIFO

Slide 21 Example 4: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: 6 units with expiration date Nov 17 and 2 units with expiration date Nov 14

• 1. Consumers bought 2 units on Nov 14, No waste at end of day • 2. Consumers bought 1 unit on Nov 14, waste at end of day = 1 unit • 3. Consumers bought 5 units on Nov 14, waste at end of day = 1 unit • 4. Consumers bought 5 units on Nov 14, waste at end of day = 2 units

• %FIFO on Nov 14=?

Slide 22 Example 4: %FIFO=?

• At start of day Nov 14, the estimated quantity on the shelf is: 6 units with expiration date Nov 17 and 2 units with expiration date Nov 14

• 1. Consumers bought 2 units on Nov 14, No waste at end of day • 2. Consumers bought 1 unit on Nov 14, waste at end of day = 1 unit • 3. Consumers bought 5 units on Nov 14, waste at end of day = 1 unit • 4. Consumers bought 5 units on Nov 14, waste at end of day = 2 units

• %FIFO on Nov 14=? 1. and 2. 100% FIFO. 3. 1/5=20% FIFO 4. 0% FIFO

Slide 23 Five potential scenario’s when 4 conditions are met

When the 4 conditions are met, 5 scenario’s are possible (see Figure). Out of these 5 scenario’s only in 1st scenario %FIFO is unknown (1st scenario is: there is no waste and demand > oldest inventory).

In all other 4 scenario’s fraction FIFO can be estimated as: % FIFO on day t = sales from the oldest batch / total sales on day t.

Note: sales from the oldest batch = the ‘oldest-batch’-quantity at start of day t minus the waste at end of day t

Slide 24 Recap: Method to measure grabbing behaviour

How to find the fraction of FIFO customers ƒ̂ ?

• Assume the following data are available per SKU (store-item combination) per day: total inventory (so no info on quantity per expiration date available on the shelf), sales, deliveries, waste and the product shelf life.

• The fraction of FIFO customers ƒ̂ is estimated by • starting with an initial estimate (ƒ̂ =0.5), • estimating each day the quantities per batch (i.e. per expiration date) available on the shelf • updating the estimate for ƒ̂ on days when the FIFO and/or LIFO withdrawal can be identified uniquely.

For details, see ‘A new replenishment policy for perishable items when expiration date visibility is limited, including a procedure to estimate customer withdrawal behaviour’, Broekmeulen & Van Donselaar, 2019. Slide 25 Project at Jumbo

Four questions: - Which options are available and which changes are needed in operations, procedures and systems to implement expiry date (ED) visibility in stores?

- For which product category is ED Visibility most beneficial and how large is the waste reduction from ED visibility?

- How many observations are needed to get a good estimate for the grabbing behavior at SKU level? This estimate is needed when the new replenishment logic at Jumbo will be based on the easiest method to implement ED visibility.

- Which factors explain the grabbing behavior?

Slide 26 Which options for ED Visibility?

Which options are available and which changes are needed in operations, procedures and systems to implement expiry date (ED) visibility in stores?

Item ED Visibility (via extended barcode or RFID per consumer unit) and Batch Visibility (via DC sharing ED info with store per batch)

Batch Visibility is easier to implement since Item Visibility requires additional changes at supplier (e.g. attaching an extended barcode)

Slide 27 Where largest benefits?

For which product categories is ED Visibility most beneficial ?

and how large is the waste reduction from ED visibility for this category? ?

Slide 28 Where largest benefits?

For which product category is ED Visibility most beneficial Potato, Vegetables and Fruit (PVF) and how large is the waste reduction from ED visibility for P.V.F.? Depends on the %FIFO customers (f) and Item or Batch ED Visibility: f=40% : 9.3% less waste with Item Visibility and 8.8% with Batch Visibility f=80% : 4.6% less waste with Item Visibility and 4.3% with Batch Visibility

Conclusions: 1. PVF products ideal for pilot. 2. Benefits from Batch Visibility are very close to benefits from Item Visibility and Batch Visibility is easier to implement.

Slide 29 Number observations needed to know % FIFO customers?

How many observations are needed to get a good estimate for the grabbing behavior at SKU level?

Method: simulation with random demand using Jumbo’s product and demand characteristics

Result for one SKU, as an example:

For all SKU’s, roughly 20 observations are needed to be within 10% relative error (i.e. error within 4% if f=40%)

Slide 30 Number of days needed for 20 observations

• Fraction of days when %FIFO can be observed decreases if shelf life increases or if On Shelf Availability (=OSA) target decreases

Fraction of days when %FIFO can be observed OSA=95% OSA=99% Shelf life = 3 days 10,0% 22,7% Shelf life = 6 days 2,2% 11,2%

• Approx. 200 days needed for 20 observations if shelf life=3 days and OSA=95%

• Less time needed if %FIFO is identical within category or among stores

Slide 31 Which factors explain the grabbing behavior?

Differences in %FIFO per product category (in green) are larger than differences in % FIFO per store (in orange)

Product category FIFO % n store 1 store 2 store 3 store 4 store 5 Potato processed 0,41 214 0,36 0,55 0,3 0,34 0,45 Cooking Vegetables 0,36 182 0,44 0,28 0,35 0,39 0,31 Herbs/germ 0,61 119 0,73 0,15 0,62 0,78 0,69 Raw food 0,49 143 0,16 0,85 0,35 0,52 0,5 Stir-fry 0,42 418 0,5 0,34 0,33 0,41 0,44 Lettuce basic 0,27 132 0,2 0,03 0,25 0,22 0,4 Lettuce premium 0,36 145 0,55 0,35 0,41 0,39 0,26 Burgers 0,54 108 0,71 0,61 0,55 0,62 0,28 Filet-pate-raw 0,45 116 0,31 0,53 0,28 0,73 0,36 AVERAGE all categories 0,43 0,44 0,41 0,38 0,49 0,41

n = nbr observations in five selected stores All reported categories have >100 observations in total

In addition, average demand is negatively correlated with % FIFO Linear regression revealed: Average demand and ‘category=Herbs’ explain 32% of the variation in %FIFO Slide 33 Roundtable discussion

How can your company (or your customer) benefit from knowing at item- store level the % customers selecting products on First In First Out-basis?

Potential areas: • Inventory replenishment • Dynamic pricing • Assortment selection • Type of Packaging • Presentation in store • Planograms (number of facings) • Store operations, e.g. shelf refilling (procedures or execution) • …

Slide 34 Roundtable discussion

Which opportunities do you see for future research related to the customer selection behavior at item-store level?

• Finding more evidence for the drivers of %FIFO, e.g. how is %FIFO dependent on Shelf life, Product category, Average demand, Type of retailer, Type of customers, Rural/Urban area, Country, Display type, Shelf filling execution, etc….? • How to deal with data inaccuracy? • Empirical validation • Added value for dynamic pricing • Other opportunities, being …. ?

Slide 35 Key Takeaways

• A method is available to measure consumer grabbing behavior (%FIFO) for each item-store combination using retailer’s current databases. • Circa 20 observations are needed to get a good estimate for %FIFO (~ 200 days sales data if shelf life=3 days). Smaller period is possible when behavior is identical within a group of similar products or similar stores. • Application at Jumbo Supermarkten reveals first drivers of this behavior: - differences between categories are larger than differences among stores. - SKU’s with higher average demand show lower %FIFO • Using %FIFO-estimate in replenishment logic yields ca 8% lower waste. • Ideal pilot category is PVF. • Batch ED visibility performs close to Item ED Visibility, easier to implement.

Slide 36 Discussion