RFID-enabled Visibility and Inventory Accuracy: A Field Experiment Bill Hardgrave John Aloysius Sandeep Goyal University of Arkansas Note: Please do not distribute or cite without explicit permission.
RFID-enabled Visibility and Inventory Accuracy: A Field Experiment John Aloysius Sandeep Goyal Bill Hardgrave University of Arkansas Note: Please do not distribute or cite without explicit permission.
RFID-enabled Visibility and Inventory Accuracy: A Field Experiment Sandeep Goyal University of Arkansas Note: Please do not distribute or cite without explicit permission.
Premise Does RFID improve inventory accuracy? Huge problem –Forecasting, ordering, replenishment based on PI –PI is wrong on 65% of items –Estimated 3% reduction in profit due to inaccuracy What can be done? –Increase frequency (and accuracy) of physical counts –Identify and eliminate source of errors
Causes of Inventory Inaccuracy PI inaccuracy causes Results in overstated PI? Results in understated PI? Can case-level RFID reduce the error? Incorrect manual adjustment Yes Improper returnsYes No Mis-shipment from DC Yes Cashier errorYes No
Examples – Manual adjustment PI = 12 Actual = 12 Casepack size = 12 Associate cannot locate case in backroom; resets inventory count to 0 PI = 0, Actual = 12 (PI < Actual) Unnecessary case ordered
Examples – Cashier error Product AProduct B PI10 Actual10 Sell 3 of A and 3 of B, but Cashier scans as 6 of A PI = 4 Actual = 7 (PI < Actual) PI = 10 Actual = 7 (PI > Actual)
Proposition RFID-enabled visibility will improve inventory accuracy RFIDVisibility Inventory accuracy Out of stocks Excess inventory
Read points - Generic DC Receiving Door Readers Shipping Door Readers Conveyor Readers Distribution Center
Read points - Generic Store Backroom Storage Sales Floor Door Readers Backroom Readers Box Crusher Reader Receiving Door Readers
RFID Data LocationEPCDate/timeReader DC :15inbound DC :54conveyor DC :23outbound ST :31inbound ST :14backroom ST :54sales floor ST :45sales floor ST :49box crusher
The Study All products in air freshener category tagged at case level Data collection: 23 weeks 13 stores: 8 test stores, 5 control stores –Mixture of Supercenter and Neighborhood Markets Determined each day: PI – actual 10 weeks to determine baseline Same time, same path each day
The Study Looked at understated PI only –i.e., where PI < actual Treatment: –Control stores: RFID-enabled, business as usual –Test stores: business as usual, PLUS used RFID reads (from inbound door, sales floor door, box crusher) to determine count of items in backroom Auto-PI: adjustment made by system For example: if PI = 0, but RFID indicates case (=12) in backroom, then PI adjusted – NO HUMAN INTERVENTION
Results - Descriptives 12% -1% 12% - (-1%) = 13%Numbers are for illustration only; not actual
Results - Descriptives
Random Coefficient Modeling Three levels –Store –SKU –Repeated measures Discontinuous growth model Covariates (sales velocity, cost, SKU variety)
Factors Influencing PI Accuracy (DeHoratius and Raman 2008) Cost Sales velocity SKU variety Audit frequency (experimentally controlled) Distribution structure (experimentally controlled) Inventory density (experimentally controlled)
Results: Test vs. Control Stores Linear Mixed Model of Test versus Control Stores VariablesEffects (Intercept)6.05*** Velocity1.92*** Variety-0.03 Item cost-0.01 Test-1.28** Period0.54*** Test: Dummy variable coded as 1 - stores in the test group; 0 - stores in the control group Period: Time variable with day 1 starting on the day RFID-based autoPI was made available in test stores * p < 0.05 ** p < 0.01 *** p < 0.001
Variable Coding For discontinuity and slope differences: Add additional vectors to the level-1 model –To determine if the post slope varies from the pre slope –To determine if there is difference in intercept between pre and post
Results: Pre and Post AutoPI Results of Linear Mixed Effects VariablesEffects (Intercept)7.871*** Velocity-0.925*** Variety Item cost-0.001* PRE0.131** TRANS-2.038*** POST-0.298*** Pre: Variable coding to represent the baseline period Trans: Variable coding to represent the transitions period—intercept Post: Variable coding to represent the treatment period p < 0.05 ** p < 0.01 *** p < 0.001
Results: Discontinuous Growth Model Model of Understated PI Accuracy over Time Intervention
Results: Known Causes Influence of RFID-enabled Visibility on Known Predictors of Inventory Inaccuracy VariablesModel 1Model 2Model 3Model 4 Intercept11.50***7.40***7.55***7.77*** Treatment-1.83***-1.28***-1.27***-1.82*** Cost Per Item (in dollars)-0.55** Velocity-0.83***-0.87***-0.90***-0.89** Variety Treatment X Cost Per item0.38*** Treatment X Velocity-0.06 Treatment X Variety-0.15** * p < 0.05 ** p < 0.01 *** p < 0.001
Results: Interaction Effects
Implications What does it mean? –Inventory accuracy can be improved (with tagging at the case level) –Is RFID needed? Could do physical counts – but at what cost? –Improving understated means less inventory; less uncertainty Value to Wal-Mart and suppliers? In the millions! –When used to improve overstated PI: reduce out of stocks even further –Imagine inventory accuracy with item-level tagging …