Decision modeling with data: real world cases Stephane Collignon, Ph.D. 10/24/2017 1
Agenda Business Data Analytics program at WVU 2 cases Delivery quality problem Business Analytics is a process Data and different phases of decision making Tools to help support decision making Truck transportation behavioral capacity assignment issue Quantifying the impact of behavioral deviance Combining modeling tools 10/24/2017 2
Decision-making process Simon, H (1977) The new science of Management Decision. Englewood Cliffs, NJ: Prentice Hall. Simon Intelligence/ Monitoring Design Choice Implementation Picture source: Sharda et al. (2015) Business Intelligence and Analytics: Systems for Decision Support. 10 th ed., Pearson 10/24/2017 3
Delivery quality problem - context 10/24/2017 4
Symptoms Major client complains about late deliveries Threatens to change distributor $45 million/ year revenue loss (5%) All deliveries in southern area HQ HQ 10/24/2017 5
Problem identification: Data to the rescue Only 1 client s issue? NO! Data shows all types of clients are late Southern warehouse issue? NO! Tasks on time & in correct sequence Northern warehouse prepares incorrectly HQ HQ Northern warehouse issue? NO! Records show otherwise 10/24/2017 6
Problem identification: Multiple logistic regression analysis Independent Variable Correlation to being late 1 st shuttle transfer High Delivery time Medium Shipping time Medium Homogeneous pallets Low Tour number - Driver ID - Shipper ID - Week day - 10/24/2017 7
True problem(s) Homogeneous pallet Two types of product preparation Analysis - Logistic regression: Homogeneous * shuttle 1 = late Heterogeneous pallet Homogeneous not in sync with heterogeneous new issue = cannot prepare on time for shuttle 10/24/2017 8
Decision Making: Scenarios/ gathering data Have Mgt prepare Mgt manages 1 shift 2 shifts 1 shift 2 shifts 1 big shuttle A B C D 2 small shuttles E F G H Volumes to be prepared Productivity with/without Mgt Crowdedness for 1 or 2 shifts Dock capacity Time on dock (chain of cold) Arrival of products in southern warehouse MONTE CARLO SIMULATION (300 iterations)
Decision Making: solution choice/ implementation Have Mgt prepare Mgt manages 1 shift 2 shifts 1 shift 2 shifts 1 big shuttle A B C D 2 small shuttles E F G H Reduced night time hours Reduced over time On time shuttle deliveries Client s business secured 3% logistic costs in this area ($400k/ year) 10/24/2017 10
Decision modeling with data case 1 Combination of techniques Analytics at different stages of decision making process Importance of data over guts Importance of critical thinking (tools do not give answer) Importance of involvement of managers 10/24/2017 11
Transportation management: behavior correction 10/24/2017 12
State of situation and Behaviors Based on past data Linear programming model was run: Estimated cost savings of 11% compare to a dedicated fleet True cost savings about 7% 10/24/2017 13
Problem identification expensive carriers assigned to some routes Short time windows Fear of not finding capacity Lack of information on capacity available Lack of information on reasonable pricing Can we modify routers behaviors? Live change is problematic Long-term experiment Possibility of losses Possibility of creating a bad reputation Can we anticipate the consequences of behavioral changes on the overall system? 10/24/2017 14
Model design: tools to avoid Traditional simulation models seem inadequate: DES: Stochastic in nature. Process oriented: Simulate passive individual entities moving through a series of activities at discrete points in time within a fully defined system behavior. (Tako & Robinson, 2012) SD: Generally deterministic in nature, model at a system and not agent level (Tako & Robinson, 2012) SD and DES tend to simulate at macroscopic level and assume the behavior of the system (Bonabeau, 2002) [based on past data] 10/24/2017 15
Model design: ABM View of a system from the agent level (Bonabeau, 2002) agents can be companies or people No assumption on the behavior of the system The system outcomes are dictated by agents behaviors / policies ABM finds its roots in Complex Adaptive Systems (CAS) as researchers studied adaptation and emergence in biological systems (Samuelson and Macal 2006). 10/24/2017 16
Model design: Agents in ABM Agents are: Self-contained Autonomous Agents can be: Adaptive Goal directed In different states Heterogeneous social Picture and table from Macal and North (2010) 10/24/2017 17
Model design: ABM Shipper initial behaviors Taking what comes first Waiting for better price (different levels) Shipper evolved behaviors Vengeance on past over-charge (different levels) Interrupt waiting for favored carrier Adapt waiting period Carrier initial behaviors Fair pricing Over-charging (different levels) Carrier evolved behaviors Lost bids influence toward fair pricing Lost fair bids influence toward abandon of spot Won bids influence toward higher prices Won after waiting influence toward over-charging New carriers arrive 10/24/2017 18
Decision Making: preliminary results No spatial animation sorry! With non-evolving waiting period and non-evolving carriers Estimated additional gain of 1.5% on transportation cost Limited gains due to absence of market behaviors 10/24/2017 19
Decision modeling with data Case 2 Prescriptive tools in different phases of decision making Combination of techniques and tools/ software Behaviors can be the object of analytics (Look at latest Nobel prize in finance: Richard Thaler) 10/24/2017 20
THANK YOU 10/24/2017 21