1. Carrier Strategies in the Spot Trucking Market
Authors: Paul Jeffrey R. Leopando
Kyle A.C. Rocca
Advisor: Dr. Chris Caplice
Sponsor: Coyote Logistics
MIT SCM ResearchFest
May 22, 2014

2. Given a set of available loads, which, if any, should a carrier take?
Problem Statement
Given a set of available loads, which, if any, should a carrier take?
Omaha, NE
Chicago, IL 1 2 3
Milwaukee, WI
Minneapolis, MN
Loaded
Empty
Set up problem statement
Explain deadheading, repositioning
Mention decision with uncertainty of P(Onward Load)
SIMPLIFIED IMAGE: more than 3 loads, multitude of loads at each destination
Decision parameters
Load 1
Load 2
Load 3
Total revenue
$700
$1000
$850
Revenue per mile (RPM)
$1.84
$1.95
$1.80
Onward load likely?
Yes No
N/A 2

3. Key Findings Can disaggregate load decision
Complex strategies > simple strategies
Deadhead distance = most important factor
Distinct results by geographic location 3

4. Overview Problem Statement Key Findings Research Context Project Goals
Research Design
Data Profiling
Initial Simulation Testing
Strategy Testing 4

5. Research Context Spot Market Owner-Operators No more than 5 vehicles
30,000 carriers
Operating strategy:
Maximize utilization
Take long hauls
Project proposal from a freight brokerage company, tasked with matching carrier capacity and shipper freight in the spot market.
Spot market loads are negotiated as one-offs
Owner operators are the largest class of carrier by number of businesses servicing spot
30,000 carriers could theoretically have 30,000 different strategies
We know some things about their strategies, but it hasn’t been widely studied 5

6. Project Goals Carriers Understand how load selections affect success
5/8/2018
Project Goals
Carriers
Understand how load selections affect success
Brokers
Improve matching of capacity and freight
Explain success metrics – revenues earned and utilization
Key up data slide
All well and good, but where do we begin to answer our research question? Masked shipment data from a proprietary load board, wealth of information, but difficult to divine strategies.
Scholars
Fill the gap in transportation literature 6

7. LoadID 2312209 2315311 2311913 2312708 LoadDate 6/1/13 Equipment V,R V
Miles 20 66 78
1230
RPM
19.25
9.09
5.13
1.71
OrigCity
Stockton
Atlanta
Downey
Paw Paw
OrigState CA GA MI
DestCity
Tracy
Duluth
Oceanside
Arcadia
DestState FL
TotalCost
385
600
400
2,100
LoadBuild
5/31/13 7:46
5/31/13 14:29
5/30/13 19:51
5/31/13 9:16
LoadBooked
5/31/13 9:48
6/1/13 8:48
5/31/13 12:23
5/31/13 11:21
TruckArrival
6/1/13 2:30
6/1/13 13:05
6/1/13 8:30
6/1/13 16:27
PickupAppt
6/1/13 4:00
6/1/13 15:00
6/1/13 9:00
6/1/13 13:00
TruckDelivery
6/1/13 9:45
6/1/13 17:00
6/1/13 13:45
6/3/13 2:56
CarrierID
26126
125042
176944
55981
CarrierHQCity
Mableton
Fontana
Harrison
CarrierHQState AR
CarrierUnits 13 1 9
Project Goals: “This is what we want to achieve”
[data]: “But this is what we have.”
[Research Design]: so this is how we thought about attacking the question 7

8. Research Design Data Profiling Initial Testing Strategy Testing
Simulations 8

9. Data Profiling Masked shipment data Scrubbed data: Objectives:
5/8/2018
Data Profiling
Masked shipment data
May to mid-October, 2013
Scrubbed data:
207,592 loads
13,117 carriers
Objectives:
Summarize data
Define relevant features of loads
First thing – what is the data? What features does it have?
Includes fields for:
Carrier ID
Equipment
Origin and Destination coordinates
Time stamps:
Entered into system
Assigned a carrier
Scheduled pick-up
Actual pick-up
Second – what analysis can we do
----- Meeting Notes (5/14/14 13:58) -----
more here, characteristics 9

10. Data Profiling Objective: Summarize data Total Owner-Operators
Generally aligned with industry trends
Total
Owner-Operators
Number of Carriers
13,117
7,590 (58%)
Avg. Length of Haul (miles)
619
593
Avg. Revenue per Loaded Mile ($/mi)
2.33
2.51
Discuss revenue per loaded mile 10

11. Utility Function for Scoring Attractiveness
Data Profiling
Objective: Define relevant features of loads
Utility Function for Scoring Attractiveness
α, β, γ, δ: Relative weights
Score = ( α
[Length of Haul] + β
[RPM]
Defined relevant features of loads and created utility function to score loads
Can use in a simulation
Can apply a linear additive function
3 features of loads, one of carrier
----- Meeting Notes (5/14/14 13:58) -----
change alignment + γ
[P(Onward Load)] + δ
[Deadhead Distance] ) 11

12. Boston, MA Chicago, IL Los Angeles, CA Atlanta, GA Dallas, TX Start
Review Test Load
Discard Test Load
Validate Test Load Feasibility
Start
All Loads During Wait Time Evaluated?
Feasible? No
animate

13. Initial Simulation Testing
Objective: Define relevant operating procedure parameters
Model parameter
Initial case
Alternative cases
Haul length preference
Long
Short
Waiting time (hours) 1
2, 3, 5, 7, 9
Speed (mph) 55
40, 45, 50
Deadhead ceiling, c (%) 10
15, 17, 20, 25
α: Length of Haul
0.25
0, 0.3, 0.5, 1
β: RPM
0, 0.3, 1
γ: P(Onward Load)
δ: Deadhead distance
Model parameter
Initial case
Alternative cases
Haul length preference
Long
Waiting time (hours) 1
Speed (mph) 55
Deadhead ceiling, c (%) 10
α: Length of Haul
0.25
β: RPM
γ: P(Onward Load)
δ: Deadhead distance
Deadhead ceiling vs. deadhead weight
relevant: length of haul, truck speed
Fix variables: long, 2, 55, 20. 13

14. Fixed parameters: Long-haul, 2 hr wait, 55 mph, 20% deadhead
Strategy Testing
Strategies
Simple
Complex L R P D LD
LRD
LPD
LRPD
α: Length of Haul 1
0.5
0.3
0.25
β: RPM
γ: P(Onward Load)
δ: Deadhead distance
Fixed parameters: Long-haul, 2 hr wait, 55 mph, 20% deadhead
arbitrary 14

15. Strategy Testing Results
Simple
Complex L R P D LD
LRD
LPD
LRPD
Total Revenue ($)
Atlanta
87,692
83,402
77,615
139,709
181,450
178,043
168,220
165,460
Boston
176,819
88,132
94,132
146,726
168,525
182,363
168,255
179,913
Chicago
79,329
84,034
84,944
151,960
186,729
198,256
162,535
184,545
Dallas
125,100
97,611
85,907
151,002
172,672
186,108
167,083
162,693
Los Angeles
86,148
81,264
80,887
155,093
175,649
191,923
172,018
189,170
Utilization (% loaded) 51 48 47 87 95 91 93 90 77 27 92 89 50 46 64 38 88 94 54 35 52 86
Quickly cover main points:
-Complex strategies superior to simple
-Deadhead single most important factor
***relative to each other
----- Meeting Notes (5/14/14 13:58) -----
explain cities
decision with uncertainty
center
use an example case
callouts
% of best
(months)
----- Meeting Notes (5/14/14 14:01) -----
center numbers
39% of maximum 15

16. Strategy Testing Results
Home city: Chicago, IL R
LRD
Strategy LRD
Strategy R
Strategies R
LRD
Total revenue ($)
16,050
42,644
Utilization (% loaded) 50 90
Total distance (mi)
10,045
24,704
Number of loads 25 16
Loaded
Empty
Month: September 16

17. Strategy Testing Results
Strategy LRD
All Owner-Operators
Avg. length of haul (mi)
1,317
593
Avg. RPM ($/mi)
1.89
2.51
Long-Haul Owner-Operators
1,280
1.63
>
Strategy LRD produces 16% improvement in average RPM. 17

18. Key Findings Can disaggregate load decision
Complex strategies > simple strategies
Deadhead distance = most important factor
Distinct results by geographic location
Less text
----- Meeting Notes (5/14/14 13:58) -----
data mining - ALOH, etc.
problem statement:
explain revenues and utilization
explain all 5 cities 18

19. Future Directions Expand scope of data and inputs:
Different carrier types, equipment, data
Longer season
Improve simulation model:
Hours of Service (HOS) restrictions
Personalize:
Profit function
Return-home pressure
Excluded states 19

20. Acknowledgments Partners at Coyote Logistics: Advisors at MIT:
Bill Driegert
David Ossim
Amit Prasad
Ella Revzin
Advisors at MIT:
Chris Caplice
Lenore Myka
Thea Singer 20

21. Questions? 21