On the impact of inventory accuracy improvements on sales Christoph - - PowerPoint PPT Presentation

• 1 -

ECR, Paris, France – February 08, 2018

Christoph Glock, Yacine Rekik, Aris A. Syntetos

On the impact of inventory accuracy improvements on sales

• 2 -

Background and objectives Inventory inaccuracies: major issue in retailing and apparel industry. Physical stock is (typically) less than what we think it is.

 Most reasonable assumption in retailing. Generally, stores are negative in terms of

stock and distribution centres are positive.

 Thus, reconciling inventories may only lead to an increase in sales.  (We will see later that positive stock is also possible, still leading though to reduced

sales!)

The problem has been established; we are not here to argue for its existence.

• 3 -

Background and objectives

But rather:

 Assess the implications of the problem, or rather the implications of fixing the problem (phase 1);  Assess alternative ways of fixing the problem itself (phase 2).

Phase 1: What is the impact on (increased) sales if inventory accuracy is increased by x%?

 How does inventory accuracy develop over time after a stock take?  Is there an optimal number of stock takes?  What exactly constitutes this problem of inventory discrepancies?

Phase 2 (upon convincing everybody of the implications): what are the strategies to be employed (algorithmic driven, new identification technologies, counting, etc.) to fight the route causes of the problem?

• 4 -

Error free (r,Q) inventory policy

Safety Stock L1 L2 t Reorder Point: r Order Quantity Q Q Lead Time

• 5 -

Safety Stock L1 L2 t Order Quantity Q Q Lead Time Expected error free stock level

Impact of errors on the (r,Q) inventory policy

Reorder Point: r This is the visible stock behavior: POS (Real Demand) + Replenishment This is the invisible stock behavior: POS (Real Demand) + Replenishment + Skrink (Ghost Demand) Actual stock level subject to errors

• 6 -

Safety Stock L1 L2 t Order Quantity Q Q Lead Time Wrong

• rdering

timing and quantity

Impact of errors on the (r,Q) inventory policy

Lost Sales are more frequent Reorder Point: r And more importantly, they are not detectable

• 7 -

Safety Stock L1 L2 t Order Quantity Q Q Lead Time Wrong

• rdering

timing and quantity

Impact of errors on the (r,Q) inventory policy

Lost Sales are more frequent Reorder Point: r And more importantly, they are not detectable

Errors act as “ghost” demand decreasing the stock level without generating a revenue: The Inventory is controlled based on some visible wrong information, whereas the sales are satisfied based on some correct but invisible information

• 8 -

Matched store experiment

Sales (e.g.) 12 weeks before count Sales (e.g.) over 12 weeks “Test” “Control”

Stock Count

Sales 12 weeks after the inventory records have been “trued up”

No Stock Count

Sales 12 weeks after

• 9 -

Empirical analysis

Currently working with 8 Retailers across Europe:

 NDAs have been signed and we are in various phases with regards to data transfer and analysis;  4 Grocery retailers (supermarkets), 2 Apparel retailers and 2 other;  Customised reports to be produced for all participating retailers.

Initial results:

 2 Grocery retailers: ALPHA and BETA

• 10 -

Aim of the study

There is some published evidence suggesting that as much as 65% of inventory records are wrong (Nicole DeHoratius, 2012). This is True for retailers ALPHA and BETA: Our objective is not to prove the existence of Discrepancies, but to help answering the questions:

 Does more accurate inventory grow sales, if so by how much?  What investment is required to improve it?

Week 1 2 3 4 5 Average Negative Discrepancy 26.46% 28.39% 27.35% 28.33% 35.42% 29.19% Positive Discrepancy 11.34% 12.99% 12.67% 13.33% 11.44% 12.36% Zero Discrepancy 62.20% 58.62% 59.97% 58.34% 53.14% 58.45% Week 12 Negative Discrepancy 29.93% Positive Discrepancy 24.44% Zero Discrepancy 45.63%

Retailer ALPHA Retailer BETA

• 11 -

Experiment at Retailer ALPHA

12 weeks Stock Audit Sales in the Test and Control stores are compared and the impact of the stock audit on sales of the test store is investigated 12 weeks CONTROL Store TEST

Store

• 12 -

SKUs Clustering: ABC classification based on Turnover

Turnover: 69.87% Turnover: 25.06% Turnover: 5.07%

0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00%

Turnover Contribution

Fast Movers 21.26% of SKUs Middle Movers 40.91% of SKUs Slow Movers 37.81% of SKUs

• 13 -

SKUs Clustering: ABC classification based on Turnover

Discrepancy: 57.93% Discrepancy: 28.23% Discrepancy; 13.83%

Turnover: 69.87% Turnover: 25.06% Turnover: 5.07%

0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00%

Discrepancy Contribution Turnover Contribution

Fast Movers (17.78% of SKUs) are generating 69.87% of turnover. They also represent 57.93% of Discrepancies

Fast Movers 21.26% of SKUs Middle Movers 40.91% of SKUs Slow Movers 37.81% of SKUs

A more accurate inventory system will highly benefit the Fast Movers!

• 14 -

SKUs Clustering: ABC classification based on Discrepancy

Less than 2% of SKUs are generating 70% of total discrepancy But 45.6% of High Discrepancy Class also belong to the Fast Mover Class The Fast/High SKUs (Fast Mover class for Sales and High Discrepancy class for Discrepancy) needs a careful analysis of inaccuracy sources and

• perations improvements

Discrepancy Discrepancy Class % of SKUs Contribution Mean € Min € Max € High Discrepancy 1.83% 69.92% 94.30

• 14485.75

14163.01 Middle Discrepancy 17.29% 25.08%

• 1.07
• 365.21

357.12 Low Discrepancy 80.88% 5.00%

• 0.41
• 19.87

19.87

• 15 -

SKUs Clustering based on the Discrepancy sign

Discrepancy is not always negative. It is not all about Shrinkage Positive discrepancy is not negligible and generally is caused by Information Systems manipulations and errors Positive or Negative discrepancy: the impact on sales is the same

Discrepancy Sign SKUs % Discrepancy Mean € Negative 29.93%

• 80.79

Positive 24.44% 103.88 Zero 45.63%

Discrepancy Sign Class SKUs % Discrepancy Mean € Discrepancy Min € Discrepancy Max € 1 4.03% 555.59 66.9 14163.01 2 4.46% 39.97 23.07 66.34 3 4.91% 14.98 9.59 23.03 4 6.22% 6.09 3.48 9.57 5 51.88% 0.16

• 1.28

3.47 6 8.49%

• 3.25
• 5.52
• 1.3

7 6.09%

• 8.31
• 11.96
• 5.53

8 5.21%

• 17.39
• 25.06
• 11.97

9 4.66%

• 39.38
• 61.27
• 25.07

10 4.05%

• 510.22
• 14485.75
• 61.44

• 16 -

Comparison Test vs. Control

The Turnover in the Test store is 5.22% higher than in the Control Store

6.11% 5.90%

• 8.21%
• 10.00%
• 8.00%
• 6.00%
• 4.00%
• 2.00%

0.00% 2.00% 4.00% 6.00% 8.00%

Fast Mover Middle Mover Slow Mover

Turnover comparison Test vs Control Store

Fast and Middle Mover SKUs which account for more than 86% of discrepancies seem to benefit from the stock audit taking place in the Test Store

• 17 -

Sales increase in the Test store, after the Audit, compared to the Control Store

4.91% 3.14% 2.52%

0.00% 1.00% 2.00% 3.00% 4.00% 5.00% 6.00%

High Discrepancy Middle Discrepancy Low Discrepancy

Both positive and negative discrepancy classes benefit from stock audit

3.61% 8.01% 2.56% 0.43% -0.22% 1.91% 1.71% 4.80% 10.42%

• 0.54%
• 2.00%

0.00% 2.00% 4.00% 6.00% 8.00% 10.00% 12.00%

Sign Class 1 Sign Class 2 Sign Class 3 Sign Class 4 Sign Class 5 Sign Class 6 Sign Class 7 Sign Class 8 Sign Class 9 Sign Class 10

All Discrepancy classes (High, Middle and Low) benefit from the stock audit with a higher benefit for the “High Discrepancy” Class

• 18 -

Experiment at Retailer BETA

4 weeks 1 week 1 week 1 week 1 week 1 week 1 week 1 week 1 week 1 week 1 week Stock Audit Computer and Counted Physical stock during the audit are contrasted with what they should be (last counted physical stock + stock input – stock

• utput)

Computer Discrepancy: what was found vs what is shown in the computer before the audit Physical Discrepancy: what was found vs what should be in the stock given last week audit and stock movements

• 19 -

SKUs Clustering: ABC classification based on Turnover

Fast Movers (14% of SKUs) are generating 69.97% of turnover. They also contribute to 48% of Computer Discrepancies and 46% of Physical Discrepancies

46.14% 24.80% 29.05%

69.97% 25.02% 5.00% 48.65% 29.23% 22.12%

0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00%

Fast Mover Middle Mover Slow Mover

Turnover vs Discrepancy Contribution Computer Discrepancy Contribution Turnover Contribution Physical Discrepancy Contribution

54% of Fast Mover class belongs to “High Physical Discrepancy” class. 33% of Fast Mover class belongs to “High Computer Discrepancy” class

• 20 -

Correlation stock movement-discrepancy

A Strong correlation between stock movements (Input and Output) and the Discrepancies (Computer and Physical)

y = 0.4037x + 6.0082 R² = 0.7622 5 10 15 20 25 30 35 40 ( 10) 10 20 30 40 50 60

Average Physical Discrepancy as a function of the Stock Output

y = 0.4332x + 3.647 R² = 0.8274 10 20 30 40 50 60 ( 10) 10 20 30 40 50 60 70 80 90

Average Physical Discrepancy as a function of Stock Input

• 21 -

Exploration of shortage situations

Over the 5 weeks of the experiment, 40% of SKUs experienced at least once a zero stock count during the weekly stock audit; These 40% of SKUs experienced possibly at least once a shortage situation over the 5 weeks; To evaluate whether a shortage occurred or not, we compare the sales for these SKUs when the stock count is zero and when the latter is positive; When the counted stock is zero during the stock audit, the EPOS are decreased by 2.86% in average. In term of sales £ turnover, this represents a loss of 3,69%; 12% of these SKUS belong to the Fast Mover class; These 40% of SKUs which experienced at least once a shortage are subject to a probability of 81% of Physical Discrepancy and a probability of 76% of Computer Discrepancy.

• 22 -

Exploration of “Frozen SKUs” cases

Some SKUs faced a situation during some weeks where the computer stock record was positive whereas the counted physical record showed a stock level equal to

• zero. With a positive computer stock level, the replenishment process does not

trigger an order from suppliers leading to a physical stock level equal to zero in the forthcoming week and consequently leading to lost sales. Each week, approximatively 10% of SKUs faced this situation and it could be verified that the EPOS for these SKUs is approximately equal to zero in the forthcoming week. 6% of SKUs faced this situation each week and a big majority of them end up as "frozen SKUs" without any EPOS signal during the 5 weeks experiment.

• 23 -

Exploration of the implication of hand adjustment of the stock

y = -0.5716x - 2.4757 R² = 0.5254 y = -0.5368x - 0.9836 R² = 0.6936

• 12
• 10
• 8
• 6
• 4
• 2

2 4 6 8 10

• 12
• 10
• 8
• 6
• 4
• 2

2 4 6 8

Discrepancy on Computer and Physical as a function of Hand Adjustment of the Stock

Average Variance Store PI vs Count Average Variance Expected PI vs Count Linear (Average Variance Store PI vs Count) Linear (Average Variance Expected PI vs Count)

There is a strong correlation between the hand-made adjustment quantity and the later-on discovered discrepancies

• 24 -

Next steps

Extending and finalizing the analysis for the rest of stores for Retailers Alpha and Beta; Completing the analysis for the rest of retailers; Producing customized reports for each of the participating retailers; Synthesizing the results in the form of a white paper to be hopefully useful for the retailing industry (and sectors within it).