Ebonyi, Nigeria: Direct Delivery Information Capture:
Transportation Optimization Analysis, July 2015JULY 2015
This publication was produced for review by the U.S. Agency for
International Development. It was prepared by the USAID | DELIVER
PROJECT, Task Order 4.
Ebonyi, Nigeria: Direct Delivery Information Capture Transportation
Optimization Analysis
The authors' views expressed in this publication do not necessarily
reflect the views of the U.S. Agency for International Development
or the United States Government.
USAID | DELIVER PROJECT, Task Order 4 The USAID | DELIVER PROJECT,
Task Order 4, is funded by the U.S. Agency for International
Development (USAID) under contract number GPO-I-00-06-00007-00,
order number AID-OAA-TO-10- 00064, beginning September 30, 2010.
Task Order 4 is implemented by John Snow, Inc., in collaboration
with PATH; Crown Agents Consultancy, Inc.; Eastern and Southern
African Management Institute; FHI 360; Futures Institute for
Development, LLC; LLamasoft, Inc.; The Manoff Group, Inc.;
Pharmaceutical Healthcare Distributers (PHD); PRISMA; and
VillageReach. The project improves essential health commodity
supply chains by strengthening logistics management information
systems, streamlining distribution systems, identifying financial
resources for procurement and supply chain operation, and enhancing
forecasting and procurement planning. The project encourages
policymakers and donors to support logistics as a critical factor
in the overall success of their healthcare mandates.
Recommended Citation Purcell, Ryan. 2015. Ebonyi, Nigeria: Direct
Delivery Information Capture—Transportation Optimization Analysis.
Arlington, Va.: USAID | DELIVER PROJECT, Task Order 4.
Abstract From November 2014–February 2015, the USAID | DELIVER
PROJECT, Task Order 4, analyzed the public health transportation
network and routes in Ebonyi state, Nigeria.
Focused on the two-year-old Direct Delivery and Information Capture
(DDIC) last mile delivery system to service delivery points
throughout Ebonyi, the goal of the analysis was to understand the
current efficiency of the distribution network, as well as to
anticipate future capacity and other bottlenecks as the DDIC
program expands.
Photo note: December 2013, net distribution campaign in Sokoto,
Nigeria. Hamisu Hassan, USAID | DELIVER PROJECT
USAID | DELIVER PROJECT John Snow, Inc. 1616 Fort Myer Drive, 16th
Floor Arlington, VA 22209 USA Phone: 703-528-7474 Fax: 703-528-7480
Email:
[email protected] Internet: deliver.jsi.com
Transportation Optimization Models
.........................................................................................................................11
Round 12 Historical Baseline and Optimized Round 12 Historical
Baseline ...............................................11 Future
Baseline............................................................................................................................................................11
Future Volume
Scenarios..........................................................................................................................................13
Service Time Scenarios
.............................................................................................................................................13
Truck Capacity Scenarios
.........................................................................................................................................13
Changing Shift Length Scenarios
.............................................................................................................................14
Example Routes from Model
........................................................................................................................................21
5. SDP Master Name Continuity
Mapping............................................................................................................10
6. Future Baseline for All
Routes............................................................................................................................12
iv
Acronyms
GIS geographic information system
GPS global positioning system
MOH Ministry of Health
SCG Supply Chain Guru™
SDP service delivery point
VRP vehicle routing problem
v
vi
Acknowledgments
The authors would like to thank the USAID | DELIVER PROJECT staff
in Nigeria for their support of this analysis; they provided data,
feedback, and comments, which proved invaluable. In addition,
special thanks to Allison Ebrahimi Gold in Washington, DC, for her
data-cleansing efforts during the initial phase of the project and
her general support throughout the project.
vii
viii
Executive Summary
Building off two-plus years of Direct Delivery Information Capture
(DDIC) program execution and data collection, as well as previous
analytical work by the USAID | DELIVER PROJECT early in 2014; this
analysis examined, in detail, the current performance of the
transport network; as well as ways to improve it, going
forward.
By using rigorous data collection, validation, and optimization
modeling analysis techniques, the team separated and isolated the
impact of several different input variables on network
efficiency.
Results indicate that volume assumptions and cubic capacity
constraints on vehicles are less likely to significantly impact the
optimum route plan for each round. Changes to time-based factors,
such as dwelling times during deliveries and total working hours
per day; however, should be expected to have significantly more
impact on route efficiency and overall performance on metrics,
including the total number of routes and total kilometers
driven.
These results highlight areas of robustness in the network route
plan, such as the ability to handle significant volume growth
without major changes. At the same time, the results highlight
areas of sensitivity—and opportunity—with changes to dwelling times
having a direct positive or negative impact on network
performance.
ix
x
Analysis Overview
Background The Direct Delivery Information Capture (DDIC) informed
push system is based on a model of vendor-managed inventory (VMI)
that has been used in Zimbabwe since 2008. In March 2012, the USAID
| DELIVER PROJECT (the project) received core funding to pilot the
VMI model in Nigeria; Bauchi and Ebonyi were selected as pilot
states.
Since early 2013, DDIC has been in place for direct delivery from
the state central warehouse to service delivery points (SDPs), such
as hospitals and clinics throughout Ebonyi. The SDPs are visited
bimonthly to check inventory levels and to top up stock to reach
preset levels for each commodity in the program. In this way, the
delivery vehicle acts as a mobile warehouse for the SDPs throughout
Ebonyi, with the goal of minimizing stockouts, expiries, and total
inventory costs.
As the DDIC program grew from the piloting stage to now serving
250+ SDPs in Ebonyi state, it became clear that delivery route
planning and execution would eventually become performance-
limiting constraints for these objectives. It was determined that
an analysis of the DDIC transportation system would provide
valuable insight to current performance, areas for improvement, and
planning for the future of the program.
Objectives The project had the following objectives:
• Collect and analyze detailed current data for the DDIC
program:
− delivery volumes
− global information system (GIS) locations of the SDPs
− global positioning system (GPS) tracking information from
delivery vehicles.
• Build an up-to-date data layer for the road network (with true
travel distances and times).
• Create analytical models for the historical transportation
network.
• Create analytical models of the best-guess future network, based
on available projections.
• Build and analyze several future model scenarios to test key
input factors and their impact on network performance.
• Draw conclusions and suggest recommendations for the future of
the DDIC program, based on these models and an analysis of their
output.
1
Methodology The agreed-upon project approach uses best practices
supply chain modeling techniques and methodology; it uses
LLamasoft’s Supply Chain Guru™ software to build and analyze the
current state and potential future transportation network
configurations for the DDIC program in Ebonyi state. The
methodology is outlined below:
1. Capture applicable local data and historical DDIC data:
a. SDP and warehouse locations, incorporating GPS data
b. delivery volumes for the history of the program (quarter 1,
2013– quarter 4, 2014)
c. GPS tracking data from delivery vehicles
d. road network data (distances and travel times)
e. central warehouse and SDP operating hours
f. truck capacities
g. physical volumes for all commodities.
2. Build a baseline model that replicates the historical reality of
DDIC round 12 deliveries (November–December 2014).
3. Build an optimized baseline model to analyze improvement
potential from round 12.
4. Build various alternative scenarios and compare results across a
variety of transportation routing metrics:
a. total delivery volumes
b. total routes required
c. total delivery time
d. total delivery distance.
5. Analyze all this information to understand future DDIC planning
and operations.
Modeling Technology The analysis team used LLamasoft’s Supply Chain
Guru™ (SCG) software package throughout the project. This tool
allows real-world supply chains to be built into mathematical
models, which can then be altered, stress-tested, optimized, and
compared against one another. Key features used on this project
include—
• LLamasoft proprietary vehicle routing problem (VRP) optimization
engine
• GPS location data for all SDPs
• road network data for calculating real-world transportation
distances and travel times
• scenario-building tools to build and compare the various
options
• geographic maps for visualizing all scenarios
2
Direct Delivery Information Capture Transportation Analysis
Delivery Volume Analysis The first completed step was collecting
historical delivery volume information, which is used in aggregate
to validate total network flows, and also to understand the
relative spread of delivery volumes throughout Ebonyi. The Nigeria
field office staff helped collect the product unit volume data (see
table 1).
Table 1. Health Commodity Volumes
This data was then merged with historical delivery quantities from
all the DDIC rounds to calculate the total volume delivered to each
SDP during the course of the program. See figure 1 for the
volumetric data. Of particular note are the relatively low volumes
from round 12, which local staff characterized as a temporary
reduction; they expect the volumes to grow significantly in the
near future.
5
Figure 1. Ebonyi State Total Commodity Volumes by Delivery
Round
0
5
10
15
20
Vo lu
m e
in c
ub ic
m et
er s
Delivery Round
Ebonyi State Total Commodity Volumes by Delivery Round
Dwell Time Analysis Next, the project team analyzed the delivery
dwelling times at all SDPs, which is the time required to review
stock levels, top up products that are below reorder points, and
complete necessary paperwork. This data was obtained from
historical delivery reports, which had good data. It should also be
noted, however, that the data were buried in many files and with a
format that required significant time to piece together.
The objective of this step was to better understand the
relationship between delivery volume and the time required to
complete the visit. Figure 2 depicts the line of best fit for this
data, with a fixed dwell time (time required regardless of volume)
of 48.455 minutes and a variable dwell time (volume-dependent) of
72.246 minutes per m^3 delivered. The average delivery volumes, per
spot, at a fraction of an m^3, indicates that most of the dwelling
time for each stop is fixed; this suggests that, with operational
efficiency improvements during stops, dwell times can be
reduced.
6
Figure 2. Fixed and Variable Dwell Times for SDP Deliveries
y = 72.246x + 48.455
GIS Data and Road Network Analysis To understand where all the
deliveries were physically made, the next step was to collect GIS
data for all SDPs. Figure 3 shows this information for all 250+
SDPs in Ebonyi. At this time, the project team has validated 95
percent of the coordinates; the remaining questionable locations
are flagged in the master GIS data file.
7
Figure 3. Service Delivery Point Locations in Ebonyi State
After completing this step, to map all the connections between the
SDPs, the team built a digital version of the road network in
Ebonyi. They used truck GIS tracker data points to plot the roads,
distances, and travel times (see figure 4). This data were then
used to build a matrix of distances and drive times between all SDP
pairs—more than 70,000!
8
Figure 4. Ebonyi Road Network Data Layer
Data Continuity Mapping While working through all the data analysis
and preparation steps, the team discovered that the raw data had
several different naming conventions; this made data merging and
comparisons very difficult. Because of this, the team also
completed a Master Data Continuity Mapping exercise, which can be
used going forward for any future analysis with the same data
sources. Figure 5 shows some of the raw data; it has been cleaned,
validated, aggregated, and given a standardized name under the
“SDP_NameSCG” heading.
9
10
Transportation Optimization Models
Round 12 Historical Baseline and Optimized Round 12 Historical
Baseline In these scenarios, round 12 historical delivery data
represented volumes delivered to each SDP. In the baseline
scenario, all route groupings and sequences were replicated; while
in the optimized baseline, these constraints were relaxed to allow
the transportation optimization solver to search for improved
routes. See table 2 for a comparison of the two scenarios.
Table 2. Health Commodity Volumes
Scenario Total Routes Total Distance (km) Total Time (min)
Baseline* 65 7,766 23,739
Differential -38% -37% -17%
The key finding from the scenarios in table 2 is that, during round
12, significant routing efficiency improvement potential was seen,
with reductions of more than 35 percent possible for both the total
number of routes, as well as the distance traveled.
Future Baseline In this scenario, and all forward-looking scenarios
that follow, future delivery volumes were projected off historical
data from rounds 4–12, with rounds 1–3 removed from the dataset
because the project was still in the initial ramp-up stages. To
provide a safety buffer for the new routes, after determining the
average delivery volume for each visit to each SDP, a +1 standard
deviation was added to each delivery volume estimate.
For routes and sequences, all options were allowed, subject to the
following assumptions: (1) available working hours at all
facilities remained constant, (2) individual truck cubic capacities
were honored, (3) all SDPs were visited once during the delivery
round, and (4) all delivery routes had to be completed in a single
work day. The future baseline scenario made no other changes to
input assumptions; therefore, it is intended to be the basis for
comparison with all other future scenarios.
11
• 5,641 total kilometers
Figure 6. Future Baseline for All Routes
12
Future Volume Scenarios In this subset of analysis scenarios, the
projected future volume assumptions were tested for their
sensitivity. By increasing and decreasing the assumed volumes by
+/- 15 percent, we can understand the impact of delivery volume
changes on the network. Table 3 lists the results for a variety of
metrics for each of these six future volume scenarios and compares
them against the future baseline.
Table 3. Future Volume Scenario Output Comparisons
With minimal differences between important outputs metrics like the
number of routes and total kilometers driven, the key finding from
this set of scenarios is that the DDIC program looks as if it can
handle healthy growth in delivery volumes without significant
changes to their transport operations.
Service Time Scenarios For this subset of scenarios, the
assumptions around fixed and variable dwelling times were tested
for their impact on network performance. Table 4 compares four
scenarios where both service time components were adjusted by +/-
20 percent, and then they were compared against the future baseline
scenario.
Table 4. Service Time Scenario Output Comparisons
As shown by the key metrics of total kilometers driven and the
calculated 8-hour working days to complete all routes, adjusting
dwelling time has only a moderate impact on network performance.
From this, we can predict that dwelling time efficiency
improvements (i.e., faster times for each stop) will have a
definite positive impact on measures that include working days,
total kilometers, and total routes required.
Truck Capacity Scenarios For this set of four scenarios, the team
tested the impact of capacity changes to the delivery vehicles. The
objective was to understand if investing in different types of
vehicles would be likely to impact
13
the transportation network performance. Table 5 indicates no change
to output metrics within a cubic capacity range of +/- 20%,
indicating that space on the delivery vehicles in not a limiting
constraint for the network. In practice, this is an incentive to
use more flexible, reliable, and fuel- efficient vehicles where
possible, because capacity is not a significant factor.
Table 5. Truck Capacity Scenario Output Comparisons
Changing Shift Length Scenarios After noting in the truck capacity
scenarios that space on trucks was not a significant concern for
route planning, the team decided to explore the time impacts on
route performance. Table 6 shows results for +/- 60 minutes of
working time in a single day, effectively expanding/reducing the
number of SDPs that could be added to a single route.
Table 6. Changing Shift Length Scenario Output Comparisons
As shown in table 6, changes to the length of the work day
significantly impact the efficiency of the network. In general,
shorter work days have more of a negative impact on routing
efficiency than any positive impact on routing efficiency when time
is added to the day. However, improvements are possible; and, it
should be noted that adding time to a single work day has very
little impact on the total working time required to make all
deliveries.
14
Conclusions
Based on the total analysis, the key points are—
• Generally, time-related inputs have more of an impact on routes
than volume/space-related factors.
• Because routes are not space constrained, minor to moderate
changes in future delivery volumes should not be expected to
significantly impact routes.
• Time efficiency (dwell times) and availability (shift length) are
much more likely to impact routes and strain the system.
Data Collection and Organization The following recommendations are
shared with respect to the DDIC team data collection and
organizational practices:
1. Adopt standard SDP naming convention across all files and
systems to facilitate tracking and analysis.
2. Collect and track dwell time data in a single sheet, which is
built on with each successive round.
− Consider tracking the processing and loading times at the
warehouse, because this will directly impact the number of SDPs
that can be reached in a given day.
3. Use round 13 (or 14) to collect or validate the geo data for the
15 SDPs with current latitude/longitude issues.
Transportation Route Planning and Delivery Operations The following
recommendations are shared for future DDIC transport route planning
and delivery operations:
1. Efforts to reduce processing and loading/unloading times at the
warehouse and SDPs should be considered because they will probably
have a positive impact on route efficiency.
− Consistency improvements for these activity times would also
increase the reliability of the new route plans.
2. Given the volatile nature of dwell times, a planned methodology
for drivers to add stops when sufficient time and inventory is
available could significantly improve network efficiency.
15
16
Example Volume Data
Source: “Ebonyi Data Dump – Dec 19 2014” supplied by JSI Nigeria
staff
17
Example Dwell Time Data
Source: “Delivery Team SDP dwelling time – Dec12_2014” supplied by
JSI Nigeria staff
18
Service Delivery Point GIS Data
Source: “Master SDP List” sheet in the ‘EBONYI MASTER – v2.xls’
file constructed by the project team
19
20
Example Routes from Model
Source: Supply Chain Guru model outputs from the “FUTURE
Unconstrained” scenario
21
22
Arlington, VA 22209 USA