Data-driven optimization of OFDA’s
disaster response capacity
Final report
© 2020 MIT Center for Transportation & Logistics | Page 2
Agenda
• Introduction
• Inventory optimization
• Recommendations
• Discussion
© 2020 MIT Center for Transportation & Logistics | Page 4
Introduction
Disaster relief
items in storage
Shipping items into
disaster regions
All pictures from USAID News Photo and other US government sides
Distributing items to
affected population
Inventory network
Potential Inventory Locations Research Questions
➢ How well does the current network perform against a portfolio of disasters?
➢ Should OFDA hold more or less inventory?
➢ Should OFDA redistribute inventory to improve performance, and if so, where should the inventory be located?
➢ How efficient is OFDA’s current prepositioning network based on existing costs and capacity? ➢ What alternative prepositioning strategies should
OFDA consider considering cost and response time?
OFDA holds a strategic stockpile of key disaster response commodities to support people world
wide in crises situations.
© 2020 MIT Center for Transportation & Logistics | Page 6
Model (i)
• Risk portfolio
• Inventory portfolio
• Carrier portfolio
• Item/usage characteristics
Inputs
Stochastic linear
program (SLP)
Model
How well does the
current setting perform?
Where to position
inventory?
What are recommended
procurement strategies to
reduce stockout risk?
How much inventory
to hold?
How much budget
to spent?
System Assessment
System Optimization
Outputs
© 2020 MIT Center for Transportation & Logistics | Page 8
Model
min 𝑦 𝑘 𝑝𝑘 𝜏𝑖,𝑗𝑚𝑘 ∙ 𝑦𝑖,𝑗𝑚𝑘 𝑗 𝑦𝑖,𝑗𝑚𝑘 ≤ 𝑋𝑖 ∀𝑖 𝑖 𝑦𝑖,𝑗𝑚𝑘 = 𝑑𝑗𝑘 ∀𝑗 𝑦𝑖,𝑗𝑚𝑘 ≥ 0 s.t. Model assumptions➢ Minimize time to respond to a portfolio of disaster scenarios.
➢ One disaster scenario k is a random combination of resupply lead time and disasters.
➢ OFDA can only supply as much as they have in storage.
➢ OFDA must satisfy entire demand. ➢ OFDA can only spent their budget.
➢ OFDA can use different modes of transportation.
𝑖
𝑦𝑖,𝑗𝑚𝑘 ∙ 𝑐𝑖,𝑗𝑚𝑘 + 𝑋𝑖 ∙ 𝑠𝑖 ≤ 𝐵
(minimize time)
(use available inventory)
(satisfy demand)
(do not exceed budget) Simplified representation of the model
(1)
(2)
(3)
(4)
Disaster portfolio
Disaster portfolio
➢ We use past disasters that OFDA responded to. ➢ We estimate a resupply lead time distribution from
OFDA’s past operations.
➢ We create a portfolio of disaster scenarios.
➢ One disaster scenario is a random combination of resupply lead time and disasters.
➢ Disaster scenarios are equally likely.
© 2020 MIT Center for Transportation & Logistics | Page 10
Disaster portfolio estimation process
Gamma(α=1.176, β=0.013)
Estimate a probability distribution from past shipments
Draw random scenario supply lead time e.g. 92 days
Collect disasters within lead time 4
Write disasters into a scenario 5
Draw random scenario supply lead time 6
e.g. 122 days
Collect disasters within lead time 7
Write disasters into a scenario 8
3 Collect disaster portfolio information
2 1
Total affected population
Initial estimation approach
➢ Our dataset of past responses includes the total affected population (TAP), which is the entire population in a disaster area.
➢ OFDA will carefully evaluate how many people they serve in a given disaster, considering, for example,
➢ People’s needs
➢ Capabilities/activities of other organizations ➢ Inventory available at OFDA
➢ OFDA’s mandate
➢ Similar to any insurance model, the size of the exposure matters for what the model recommends. ➢ We have to make assumptions, on how many people
OFDA seeks to serve and how they decide to allocate scarce inventory.
# of disasters
© 2020 MIT Center for Transportation & Logistics | Page 12
Beneficiaries served – anchor points
Insights
➢ Beneficiaries served varies substantially by item. ➢ Max beneficiaries served is only matched by current
inventory in case of buckets and kitchen sets.
➢ Current inventory levels can serve at least 75% of the beneficiaries served in the past.
Item Max served 3rd Quartile Median People served by current Inventory Blankets 592,514 122,500 28,938 183,750 Buckets 67,500 - 67,500 183,750 Hygiene Kits 379,200 163,900 34,280 125,000 Kitchen Sets 140,100 99,150 23,700 182,500 Plastic Sheeting 851,500 100,000 22,000 540,000 Water 622,250 182,550 24,250 183,750 3.2x 3.0x 1.6x 3.4x 𝑀𝑎𝑥𝑖𝑚𝑢𝑚 𝑠𝑒𝑟𝑣𝑒𝑑 𝑆𝑒𝑟𝑣𝑒𝑑 𝑏𝑦 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑖𝑛𝑣𝑒𝑛𝑡𝑜𝑟𝑦 Conclusion
➢ OFDA seems to have served disasters with a larger target population, than what they currently have on inventory. 0.4x 0.8x 0.7x 1.3x 0.2x 1.0x -0.5x
% TAP served per item
© 2020 MIT Center for Transportation & Logistics | Page 14
Regression
Approach
➢ To find a relationship we use linear regression ➢ Using linearization, we estimate a function that
converts TAP into a target population
Regression Statistics Multiple R 0.662544357 R Square 0.438965026 Adjusted R Square 0.431484559 Standard Error 0.95460776 Observations 77 ANOVA df SS MS F Significance F Regression 1 53.47504607 53.47504607 58.68150546 5.27146E-11 Residual 75 68.34569809 0.911275975 Total 76 121.8207442
Coefficients Standard Error t Stat P-value Lower 95% Upper 95%
Intercept 2.363161747 0.703968041 3.356916234 0.001240097 0.960785495 3.765537999 X Variable 1 0.394799314 0.051537787 7.660385464 5.27146E-11 0.292130778 0.497467851
ln(Percent_Served)=2.3632 + 0.3948 * ln(1/TAP)
Results
➢ The function can explain OFDA’s target population to some extent.
➢ We use the function to predict target population. It is a volume-dependent, scaling factor for TAP.
➢ For some items we may have to cap the predicted target TAP (see next slide).
Disaster portfolio and OFDA’s past actions
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities
Disasters OFDA’s responses in the past
• Evaluate OFDA’s disaster responses in the past
beneficiaries served percent served vs. TAP
© 2020 MIT Center for Transportation & Logistics | Page 16
OFDA’s risk portfolio scenarios – blankets
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities • Use past activities to estimate
mandate
Disasters OFDA’s responses in the past Risk portfolios: blankets
Medium-Large Large
Small-Medium Medium Small
• Creating different item-specific scenarios = OFDA’s future risk portfolio scale
• Characterize different exposure to disaster sizes Mean Std. Dev. Cap 60,000 9,400 Small 62,000 80,000 14,500 Small -Medium 84,000 180,000 48,900 Large 200,000 140,000 33,300 Medium -Large 152,000 100,000 Medium 106,000 20,100 A B C
Disaster scope Disaster scale Risk portfolio(s)
Risk portfolio development
• Mix of geographical location of disasters
• Mix of disaster types • Total affected population
I
II
III
• Share of population targeted by OFDA
• Resulting risk portfolio OFDA seeks to serve
Op
ti
on
1 • Mix of disaster locations and disaster types from OFDA’s past operations
• Total affected population from UN reports
• TAP size dependent scaling of targeted population based on OFDA’s past response activities
• Risk portfolio represents typical mix of response activities of OFDA • Varying the size of disasters in the
portfolio to show impact of various risk exposure
Op
ti
on
2 Predict• a mix of potential disaster types
and locations, and
• associated total affected population
• TAP size dependent scaling of targeted population based on OFDA’s past response activities
• Risk portfolio represents typical mix of response activities of OFDA • Varying the size of disasters in the
portfolio to show impact of various risk exposure
© 2020 MIT Center for Transportation & Logistics | Page 18
A three-step process
• Optimize the current inventory allocation for the current total inventory
• Determine performance for the implied risk portfolio
• Determine performance against various other risk portfolios • Create OFDA’s risk portfolio with
target beneficiaries.
• Collect OFDA’s supply lead time estimators.
• Create scenarios, i.e. vary OFDA’s risk portfolio
• Define objective
• Adjust inventory levels • Identify optimal allocation • Recommend actions
OFDA’s risk portfolio Inventory optimization Recommendations
1 2 3
2A 2B
Inventory allocation – the intuition
Given inventory level – varying risk portfolioConsolidate Diversify
Given risk portfolio – varying inventory level
Consolidate Diversify
Relative to the expected risk exposure the total inventory in the system determines if a disaster response agency should operate a diversified or a consolidated network
!
!
low scale medium scale
high scale low inventory medium inventory high inventory
© 2020 MIT Center for Transportation & Logistics | Page 20
Impact and implications for decision-makers
Depending on OFDA’s risk exposure expectations, OFDA has to decide what service they want to achieve.
!
!
• higher service levels • lower response times • lower logistics costs • lower service levels
• higher response times • higher logistics costs
• higher service levels • lower response times • lower logistics costs • lower service levels
• higher response times • higher logistics costs
Scenarios vs. optimization
Risk portfolio Degree of diversification Total system inventory Optimal inventory allocation effectiveness and service metrics +© 2020 MIT Center for Transportation & Logistics | Page 22
Step 2A – Optimal allocation for current ops
% of targeted TAP served % of cycles fully served Average response time
Blankets Buckets Hygiene kits Kitchen sets Plastic sheeting Water
Current inventory allocation metric *
Inventory balance metric Implied risk portfolio ** 0.9 0.5 1.2 0.9 0.65 0.5 Average costs 96% 98% 95% 96% 98% 98% 15.4h 14.7h 16.1h 15.4h 15.2h 14.7h $4,910 $2,620 $14,000 $16,400 $157,700 $26,100 81% 91% 67% 81% 86% 91% 97% 98% 93% 97% 97% 98% (-4%) (-2%) (-5%) (-4%) (-2%) (-2%) (-6%) (-4%) (-8%) (-6%) (-3%) (-4%)
* Based on OFDA’s current inventory allocation
** Identified by evaluating multiple risk portfolios and finding the optimal allocation that results in the same inventory allocation metric
Implied risk portfolio
Corresponding optimal performance metrics
If we consider the current inventory distribution as a starting point and assume an appropriate risk portfolio, OFDA can gain lead time and cost reductions through inventory reallocation!
Mean Stand. Dev. 82,000 15,300 50,000 7,200 87,000 16,500 83,000 15,200 194,000 54,600 50,000 7,200 Current inventory level 73,500 73,500 25,000 36,500 10,800 73,500
To capture the benefits, OFDA should consider reallocating inventory
Subang, MY Miami, USA
Blankets Buckets Hygiene kits Kitchen sets Plastic sheeting Water
Dubai, UAE Pisa, I 34,800 28,464 18,400 17,400 4,340 28,464 34,800 20,800 6,600 17,400 4,340 20,800 3,900 20,800 0 1,700 2,320 20,800 0 3,436 0 0 0 3,436 (+158%) (+54%) (-34%) (+168%) (+74%) (+54%) (-90%) (-4%) (-92%) (-51%) (+4%)
Optimal inventory levels*
Step 2B: Is the risk portfolio appropriate for each item and are the effectiveness and service metrics acceptable?
(+74%) (+23%) (+74%) (+32%) (-16%)
Blankets
© 2020 MIT Center for Transportation & Logistics | Page 26
Disaster portfolio and OFDA’s past actions
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities
Disasters OFDA’s responses in the past
• Evaluate OFDA’s disaster responses in the past
beneficiaries served percent served vs. TAP
OFDA’s risk portfolio scenarios – blankets
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities • Use past activities to estimate
mandate
Disasters OFDA’s responses in the past Risk portfolios: blankets
• Creating different item-specific scenarios = OFDA’s future risk portfolio scale
• Characterize different exposure to disaster sizes Mean Std. Dev. Cap 60,000 9,400 Small 62,000 80,000 14,500 Small -Medium 84,000 180,000 48,900 Large 200,000 140,000 33,300 Medium -Large 152,000 100,000 Medium 106,000 20,100 A B C
Blankets
Metrics – medium scale risk portfolio
Balance metric Inventory allocation 80 - 100% 0 - 2 Results• Reallocating inventory benefits OFDA • Response time reduction: 3% • Cost reduction: 4%
• 75% of beneficiaries served
• 93% of all disaster scenarios served
➢ Current inventory levels result in mixed service levels.
➢ Current inventory allocation is close to optimum. 97% 75% 93% 1.02 - 3% - 4% 15.7h $5,056
© 2020 MIT Center for Transportation & Logistics | Page 30
Metrics – large scale risk portfolio
Balance metric
Inventory allocation
80 - 100%
0 - 2 Results
• Reallocating inventory benefits OFDA • Response time reduction: 6% • Cost reduction: 9%
• 57% of beneficiaries served
• 80% of all disaster scenarios served
➢ Current inventory levels result in lower service and strong consolidation.
➢ Reallocation benefits for response time and cos. 94% 57% 80% 1.73 - 6% - 9% 16.5h $5,468 Medium-Large Large Small-Medium Medium Small
Metrics – small scale risk portfolio
Results
• Reallocating inventory benefits OFDA • Response time reduction: 4% • Cost reduction: 6%
• 88% of beneficiaries served
• 97% of all disaster scenarios served
➢ Current inventory is sufficient to run a diversified response network effectively. ➢ Cost and response time benefits from
reallocating inventory. Balance metric Inventory allocation 80 - 100% 0 - 2 96% 88% 97% 058 - 4% - 6% 14.9h $4,625
© 2020 MIT Center for Transportation & Logistics | Page 32
Metrics – comparing risk portfolios
Results
At current inventory levels and with a larger scale risk portfolio
• OFDA should consider to run a more consolidated network.
• Service levels drop.
• Cost and lead time increase.
Balance metric Inventory allocation 80 - 100% 0 - 2 Medium-Large Large Small-Medium Medium Small
Optimal inventory allocation
• A larger scale risk portfolio means that OFDA expects to serve more beneficiaries in each disaster and the variation of beneficiaries increases.
• The current inventory allocation seems to reflect that OFDA expects risk portfolios to be lower scale.
• OFDA benefits from inventory reallocation towards Dubai and Miami to reduce response time and cost.
• At current inventory levels, OFDA should consider to consolidate the inventory if they expect a larger scale risk portfolio.
Results
Today Small Small-Med Current
Optimum Medium Med-Large Large Inv. Allocation Metric 0.9 0.6 0.8 0.9 1.0 1.4 1.7 Dubai, UAE 20,000 24,800 33,600 34,800 42,400 60,800 73,500 Miami, USA 13,500 24,800 33,600 34,800 31,100 12,700 0 Pisa, Italy 40,000 23,900 6,300 3,900 0 0 0 Subang, Malaysia 0 0 0 0 0 0 0
© 2020 MIT Center for Transportation & Logistics | Page 34
Choosing appropriate service levels
• Directionally, what is the scale of the risk portfolio that OFDA expects for blankets?
smaller scale < > larger scale • Considering this risk portfolio, is OFDA
satisfied with the service metrics, response times, and logistics cost?
higher < > lower Questions
Medium-Large Large
Small-Medium Medium Small
Buckets
OFDA’s risk portfolio scenarios – buckets
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities • Use past activities to estimate
mandate
Disasters OFDA’s responses in the past Risk portfolios: buckets
• Creating different item-specific scenarios = OFDA’s future risk portfolio scale
• Characterize different exposure to disaster sizes Mean Std. Dev. Cap 60,000 9,400 Small 62,000 80,000 14,500 Small -Medium 84,000 180,000 48,900 Large 200,000 140,000 33,300 Medium -Large 152,000 100,000 Medium 106,000 20,100 A B C
Buckets
Metrics – medium scale risk portfolio
Balance metric Inventory allocation 80 - 100% 0 - 2 Results• Reallocating inventory benefits OFDA • Response time reduction: 3% • Cost reduction: 5%
• 75% of beneficiaries served
• 93% of all disaster scenarios served
➢ Current inventory levels result in mixed service levels.
➢ OFDA should consider to consolidate
inventory to reduce response time and save logistics costs. 97% 75% 93% 1.02 - 3% - 5% 15.7h $2,920
© 2020 MIT Center for Transportation & Logistics | Page 40
Metrics – large scale risk portfolio
Balance metric
Inventory allocation
80 - 100%
0 - 2 Results
• Reallocating inventory benefits OFDA • Response time reduction: 6% • Cost reduction: 12%
• 57% of beneficiaries served
• 80% of all disaster scenarios served
➢ Current inventory levels result in lower service metrics.
➢ A more consolidated network results in lower response times and saves logistics costs. 94% 57% 80% 1.73 - 6% - 12% 16.5h $3,160 Medium-Large Large Small-Medium Medium Small
Metrics – small scale risk portfolio
Results
• Reallocating inventory benefits OFDA • Response time reduction: 4% • Cost reduction: 6%
• 88% of beneficiaries served
• 97% of all disaster scenarios served
➢ Current inventory is sufficient to run a diversified response network effectively. ➢ Some benefits of reallocating inventory.
Balance metric Inventory allocation 80 - 100% 0 - 2 96% 88% 97% 058 - 4% - 6% 14.9h $2,665
© 2020 MIT Center for Transportation & Logistics | Page 42
Metrics – comparing risk portfolios
Results
At current inventory levels and with a larger scale risk portfolio
• OFDA should consider to run a more consolidated network.
• Service levels drop.
• Cost and lead time increase.
Balance metric Inventory allocation 80 - 100% 0 - 2 Medium-Large Large Small-Medium Medium Small
Optimal inventory allocation – buckets
Today Current
Optimum Small Small-Med Medium
Med-Large Large Inv. Allocation Metric 0.5 0.5 0.6 0.8 1.0 1.4 1.7 Dubai, UAE 31,600 28,464 24,800 33,600 42,400 60,800 73,500 Miami, USA 13,500 20,800 24,800 33,600 31,100 12,700 0 Pisa, Italy 20,000 20,800 23,900 6,300 0 0 0 Subang, Malaysia 8,400 3,436 0 0 0 0 0
• A larger scale risk portfolio means that OFDA expects to serve more beneficiaries in each disaster and the variation of beneficiaries increases.
• The current inventory allocation seems to reflect that OFDA expects a rather small scale risk portfolio.
• OFDA benefits from inventory reallocation towards Dubai and Miami to reduce response time and cost.
• At current inventory levels, OFDA should consider to consolidate the inventory if they expect a larger scale risk portfolio.
© 2020 MIT Center for Transportation & Logistics | Page 44
Choosing appropriate service levels
• Directionally, what is the scale of the risk portfolio that OFDA expects for buckets?
smaller scale < > larger scale • Considering this risk portfolio, is OFDA
satisfied with the service metrics, response times, and logistics cost?
higher < > lower Questions
Medium-Large Large
Small-Medium Medium Small
Hygiene Kits
OFDA’s risk portfolio scenarios – hygiene kits
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities • Use past activities to estimate
mandate
Disasters OFDA’s responses in the past Risk portfolios: hygiene kits
• Creating different item-specific scenarios = OFDA’s future risk portfolio scale
• Characterize different exposure to disaster sizes Mean Std. Dev. Cap 30,000 3,300 Small 30,500 60,000 9,400 Small -Medium 62,000 130,000 29,700 Large 140,000 100,000 20,100 Medium -Large 106,000 80,000 Medium 84,000 14,500 A B C
Hygiene Kits
Metrics – medium scale risk portfolio
Balance metric Inventory allocation 80 - 100% 0 - 2 Results• Reallocating inventory benefits OFDA • Response time reduction: 3% • Cost reduction: 5%
• 69% of beneficiaries served
• 93% of all disaster scenarios served
➢ Current inventory levels result in mixed service levels.
➢ Current inventory allocation is close to optimum. 97% 70% 93% 1.1 - 3% - 5% 15.9h $13,800
© 2020 MIT Center for Transportation & Logistics | Page 50
Metrics – large scale risk portfolio
Balance metric
Inventory allocation
80 - 100%
0 - 2 Results
• Reallocating inventory benefits OFDA • Response time reduction: 11% • Cost reduction: 17%
• 54% of beneficiaries served
• 79% of all disaster scenarios served
➢ Current inventory levels result in lower service and higher costs and response time. ➢ OFDA should strongly consolidate their
hygiene kits inventory.
89% 54% 79% 1.73 - 11% - 17% 16.5h $14,600 Medium-Large Large Small-Medium Medium Small
Metrics – small scale risk portfolio
Balance metric Inventory allocation 80 - 100% 0 - 2 Results• Reallocating inventory benefits OFDA • Response time reduction: 2% • Cost reduction: 3%
• 94% of beneficiaries served
• 99% of all disaster scenarios served
- 2% - 3% 14.6h 98% 94% 99% 0.37 $11,900
➢ Current inventory is sufficient to run a diversified response network effectively. ➢ OFDA should consider to diversify the
inventory across the network to reduce cost and response times.
© 2020 MIT Center for Transportation & Logistics | Page 52
Metrics – comparing risk portfolios
Balance metric Inventory allocation 80 - 100% 0 - 2 Medium-Large Results
At current hygiene kits inventory levels and with a larger scale risk portfolio
• OFDA should consider to run a more consolidated network.
• Service levels drop.
• Cost and lead time increase.
Large Small-Medium Medium
Optimal inventory allocation
Today Small Small-Med Medium Current
Optimum Med-Large Large Inv. Allocation Metric 1.2 0.4 1.0 1.1 1.2 1.4 1.7 Dubai, UAE 15,000 9,332 12,400 16,800 18,400 21,200 25,000 Miami, USA 10,000 6,700 12,400 8,200 6,600 3,800 0 Pisa, Italy 0 6,100 200 0 0 0 0 Subang, Malaysia 0 2,868 0 0 0 0 0
• A larger scale risk portfolio means that OFDA expects to serve more beneficiaries in each disaster and the variation of beneficiaries increases.
• The current inventory allocation seems to reflect that OFDA expects risk portfolios to be in the medium to large scale.
• OFDA can reduce response time and cost by reallocating some inventory to Dubai.
• At current inventory levels, OFDA should consider to consolidate the inventory if they expect a larger scale risk portfolio.
© 2020 MIT Center for Transportation & Logistics | Page 54
Choosing appropriate service levels
• Directionally, what is the scale of the risk portfolio that OFDA expects for hygiene kits?
smaller scale < > larger scale • Considering this risk portfolio, is OFDA
satisfied with the service metrics, response times, and logistics cost?
higher < > lower Questions
Medium-Large Large
Small-Medium Medium Small
Kitchen sets
OFDA’s risk portfolio scenarios – kitchen sets
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities • Use past activities to estimate
mandate
Disasters OFDA’s responses in the past Risk portfolios: kitchen sets
• Creating different item-specific scenarios = OFDA’s future risk portfolio scale
• Characterize different exposure to disaster sizes Mean Std. Dev. Cap 60,000 9,400 Small 62,000 80,000 14,500 Small -Medium 84,000 180,000 48,900 Large 200,000 140,000 33,300 Medium -Large 152,000 100,000 Medium 106,000 20,100 A B C
Kitchen sets
Metrics – medium scale risk portfolio
Balance metric Inventory allocation 80 - 100% 0 - 2 Results• Reallocating inventory benefits OFDA • Response time reduction: 3% • Cost reduction: 5%
• 75% of beneficiaries served
• 93% of all disaster scenarios served
➢ Current inventory levels result in mixed service levels.
➢ Current inventory allocation is close to optimum.
➢ Some cost and response time benefits through reallocation. 97% 75% 93% 1.02 - 3% - 5% 15.7h $16,850
© 2020 MIT Center for Transportation & Logistics | Page 60
Metrics – large scale risk portfolio
Balance metric
Inventory allocation
80 - 100%
0 - 2 Results
• Reallocating inventory benefits OFDA • Response time reduction: 6% • Cost reduction: 9%
• 57% of beneficiaries served
• 80% of all disaster scenarios served
➢ At current kitchen sets inventory levels, service is low and cost and response times are higher.
➢ OFDA should consider to consolidate inventory. 94% 56% 80% 1.73 - 6% - 9% 16.5h $18,240 Medium-Large Large Small-Medium Medium Small
Metrics – small scale risk portfolio
Results
• Reallocating inventory benefits OFDA • Response time reduction: 4% • Cost reduction: 6%
• 88% of beneficiaries served
• 97% of all disaster scenarios served
➢ Current inventory is sufficient to run a diversified response network effectively. ➢ OFDA should consider to diversify the
allocation more. Balance metric Inventory allocation 80 - 100% 0 - 2 96% 88% 97% 058 - 4% - 6% 14.9h $15,410
© 2020 MIT Center for Transportation & Logistics | Page 62
Metrics – comparing risk portfolios
Results
At current hygiene kits inventory levels and with a larger scale risk portfolio
• OFDA should consider to run a more consolidated network.
• Service levels drop.
• Cost and lead time increase.
Balance metric Inventory allocation 80 - 100% 0 - 2 Medium-Large Large Small-Medium Medium Small
Optimal inventory allocation
Today Small Small-Med Current
Optimum Medium Med-Large Large Inv. Allocation Metric 0.9 0.6 0.8 0.9 1.0 1.4 1.7 Dubai, UAE 10,000 12,400 16,800 17,400 21,200 30,400 36,500 Miami, USA 6,500 12,400 16,800 17,400 15,300 6,100 0 Pisa, Italy 20,000 11,700 2,900 1,700 0 0 0 Subang, Malaysia 0 0 0 0 0 0 0
• A larger scale risk portfolio means that OFDA expects to serve more beneficiaries in each disaster and the variation of beneficiaries increases.
• The current inventory allocation seems to reflect that OFDA expects risk portfolios to be in the medium scale.
• OFDA can reduce response time and cost by reallocating some inventory from Italy.
• At current inventory levels, OFDA should consider to consolidate the inventory if they expect a larger scale risk portfolio.
© 2020 MIT Center for Transportation & Logistics | Page 64
Choosing appropriate service levels
• Directionally, what is the scale of the risk portfolio that OFDA expects for kitchen sets?
smaller scale < > larger scale • Considering this risk portfolio, is OFDA
satisfied with the service metrics, response times, and logistics cost?
higher < > lower Questions
Medium-Large Large
Small-Medium Medium Small
Plastic sheeting
OFDA’s risk portfolio scenarios – plastic sheeting
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities • Use past activities to estimate
mandate
Disasters OFDA’s responses in the past Risk portfolios: plastic sheeting
• Creating different item-specific scenarios = OFDA’s future risk portfolio scale
• Characterize different exposure to disaster sizes Mean Std. Dev. Cap 100,000 20,100 Small 106,000 200,000 57,000 Small -Medium 224,000 500,000 207,000 Large 617,000 400,000 152,300 Medium -Large 480,000 300,000 Medium 349,000 102,400 A B C
Plastic sheeting
Metrics – medium scale risk portfolio
Balance metric Inventory allocation 80 - 100% 0 - 2 Results• Reallocating inventory benefits OFDA • Response time reduction: 3% • Cost reduction: 3%
• 73% of beneficiaries served
• 94% of all disaster scenarios served
➢ Current inventory levels result in mixed service levels.
➢ OFDA should consider to consolidate
inventory to reduce response time and cost.
98% 73% 94% 1.07 - 2% - 3% 15.7h $166k
© 2020 MIT Center for Transportation & Logistics | Page 70
Metrics – large scale risk portfolio
Balance metric
Inventory allocation
80 - 100%
0 - 2 Results
• Reallocating inventory benefits OFDA • Response time reduction: 9% • Cost reduction: 14%
• 56% of beneficiaries served
• 82% of all disaster scenarios served
➢ Current inventory levels result in lower service.
➢ OFDA should consider to consolidated
inventory to lower reduce cost and response time. 91% 56% 82% 1.65 - 9% - 14% 16.0h $172k Medium-Large Large Small-Medium Medium Small
Metrics – small scale risk portfolio
Results
• Reallocating inventory benefits OFDA • Response time reduction: 1% • Cost reduction: 2%
• 96% of beneficiaries served
• 99% of all disaster scenarios served
➢ Current inventory is sufficient to run a diversified response network effectively.
Balance metric Inventory allocation 80 - 100% 0 - 2 99% 96% 99% 032 - 1% -2% 14.2h $140k
© 2020 MIT Center for Transportation & Logistics | Page 72
Metrics – comparing risk portfolios
Results
At current inventory levels and with a larger scale risk portfolio
• OFDA should consider to run a more consolidated network.
• Service levels drop.
• Cost and lead time increase.
Balance metric Inventory allocation 80 - 100% 0 - 2 Medium-Large Large Small-Medium Medium Small
Optimal inventory allocation
Today Small Current
Optimum Small-Med Medium
Med-Large Large Inv. Allocation Metric 0.65 0.3 0.65 0.7 1.1 1.5 1.7 Dubai, UAE 3,280 4,240 4,340 4,480 6,980 9,600 10,617 Miami, USA 2,500 2,520 4,340 4,480 4,020 1,400 383 Pisa, Italy 4,760 2,120 2,320 2,040 0 0 0 Subang, Malaysia 460 2,120 0 0 0 0 0
• A larger scale risk portfolio means that OFDA expects to serve more beneficiaries in each disaster and the variation of beneficiaries increases.
• The current inventory allocation seems to reflect that OFDA expects a small-medium scale risk portfolio.
• OFDA benefits from inventory reallocation towards Dubai and Miami to reduce response time and cost.
• At current inventory levels, OFDA should consider to consolidate the inventory if they expect a larger scale risk portfolio.
© 2020 MIT Center for Transportation & Logistics | Page 74
Choosing appropriate service levels
• Directionally, what is the risk portfolio that OFDA expects for plastic sheeting?
smaller < > larger
• Considering this risk portfolio, is OFDA
satisfied with the service metrics, response times, and logistics cost?
higher < > lower Questions
Medium-Large Large
Small-Medium Medium Small
Water
OFDA’s risk portfolio scenarios – water
• Disaster responses in the past • TAP and location
• Adjust TAP according to OFDA’s disaster response activities • Use past activities to estimate
mandate
Disasters OFDA’s responses in the past Risk portfolios: water
• Creating different item-specific scenarios = OFDA’s future risk portfolio scale
• Characterize different exposure to disaster sizes Mean Std. Dev. Cap 60,000 9,400 Small 62,000 80,000 14,500 Small -Medium 84,000 180,000 48,900 Large 200,000 140,000 33,300 Medium -Large 152,000 100,000 Medium 106,000 20,100 A B C
Water
Metrics – medium scale risk portfolio
Balance metric Inventory allocation 80 - 100% 0 - 2 Results• Reallocating inventory benefits OFDA • Response time reduction: 3% • Cost reduction: 5%
• 75% of beneficiaries served
• 93% of all disaster scenarios served
➢ Current inventory levels result in mixed service levels.
➢ OFDA should consider to consolidate inventory 97% 75% 93% 1.02 - 3% - 5% 15.7h $29,100
© 2020 MIT Center for Transportation & Logistics | Page 80
Metrics – large scale risk portfolio
Balance metric
Inventory allocation
80 - 100%
0 - 2 Results
• Reallocating inventory benefits OFDA • Response time reduction: 6% • Cost reduction: 10%
• 57% of beneficiaries served
• 80% of all disaster scenarios served
➢ Current inventory levels result in lower service
➢ OFDA should consider to strongly
consolidate inventory in order to reduce cost and response time.
92% 57% 80% 1.73 - 8% - 13% 16.5h $31,500 Medium-Large Large Small-Medium Medium Small
Metrics – small scale risk portfolio
Results
• Reallocating inventory benefits OFDA • Response time reduction: 4% • Cost reduction: 6%
• 88% of beneficiaries served
• 97% of all disaster scenarios served
➢ Current inventory is sufficient to run a diversified response network effectively.
Balance metric Inventory allocation 80 - 100% 0 - 2 96% 88% 97% 058 - 4% - 6% 14.9h $26,600
© 2020 MIT Center for Transportation & Logistics | Page 82
Metrics – comparing risk portfolios
Results
At current inventory levels and with a larger scale risk portfolio
• OFDA should consider to run a more consolidated network.
• Service levels drop.
• Cost and lead time increase.
Balance metric Inventory allocation 80 - 100% 0 - 2 Medium-Large Large Small-Medium Medium Small
Optimal inventory allocation
Today Current
Optimum Small Small-Med Medium
Med-Large Large Inv. Allocation Metric 0.5 0.5 0.6 0.8 1.0 1.4 1.7 Dubai, UAE 31,600 28,464 24,800 33,600 42,400 60,800 73,500 Miami, USA 13,500 20,800 24,800 33,600 31,100 12,700 0 Pisa, Italy 20,000 20,800 23,900 6,300 0 0 0 Subang, Malaysia 8,400 3,436 0 0 0 0 0
• A larger scale portfolio means that OFDA expects to serve more beneficiaries in each disaster and the variation of beneficiaries is higher.
• The current inventory allocation seems to reflect that OFDA expects a small scale risk portfolio.
• OFDA benefits from inventory reallocation towards Dubai and Miami to reduce response time and cost.
• At current inventory levels, OFDA should consider to consolidate the inventory if they expect a larger scale risk portfolio.
© 2020 MIT Center for Transportation & Logistics | Page 84
Choosing appropriate service levels
• Directionally, what is the risk portfolio that OFDA expects for water?
smaller scale < > larger scale • Considering this risk portfolio, is OFDA
satisfied with the service metrics, response times, and logistics cost?
higher < > lower Questions
Medium-Large Large
Small-Medium Medium Small
A three-step process
• Optimize the current inventory allocation for the current total inventory
• Determine performance for the implied risk portfolio
• Determine performance against various other risk portfolios • Create OFDA’s risk portfolio with
target beneficiaries.
• Collect OFDA’s supply lead time estimators.
• Create scenarios, i.e. vary OFDA’s risk portfolio
• Define objective
• Adjust inventory levels • Identify optimal allocation • Recommend actions
OFDA’s risk portfolio Inventory optimization Recommendations
1 2 3
2A 2B
© 2020 MIT Center for Transportation & Logistics | Page 86
Options
There are different options to shed further light that need to be driven by OFDA’s decision-makers
• acceptable risk portfolio • acceptable service metrics
• keep total inventory
• reallocate inventory to reap response time/cost benefits • no change in diversification
1
• higher/lower risk portfolio • acceptable service metrics
• keep total inventory
• reallocate inventory to reap response time/cost benefits • change diversification
2
• acceptable risk portfolio
• higher/lower service/effectiveness metrics
• higher/lower total inventory
• reallocate inventory to reap response time/cost benefits • change diversification
3
• … • …
Forward-Looking “Disaster Portfolio”
“Disaster Portfolio” is an input to current model
• Based on OFDA historical response data (slide 10)
Stochastic Linear Program Model (SLP) Item/Usage Characteristics Inventory Portfolio Carrier Portfolio Disaster Portfolio
1 USGCRP, 2018: Impacts, Risks, and Adaptation in the United States: Fourth National Climate Assessment, Volume II [Reidmiller, D.R., C.W. Avery, D.R. Easterling,
K.E. Kunkel, K.L.M. Lewis, T.K. Maycock, and B.C. Stewart (eds.)]. U.S. Global Change Research Program, Washington, DC, USA, 1515 pp. doi: 10.7930/NCA4.2018.
Opportunity to improve model by incorporating a Forward-Looking “Disaster Portfolio”
• Impact of disasters on people around the world is increasing in recent years and will continue to
increase in the future
1• Calculate expected total affected populations (TAP) based on probability of future disasters
• Will allow OFDA to be less reactive and more responsive to future events
© 2020 MIT Center for Transportation & Logistics | Page 88
Potential Methodology
Forward-Looking “Disaster Portfolio”TAP
(Total Affected Population)
for specific hazards specific regions & specific return periods
Population Impacted by Hazard Vulnerability Indicators (country specific) Global Population Dataset Forward-Looking Probabilistic Hazard Dataset 1 2 3
Potential Methodology
Forward-Looking Probabilistic Hazard Dataset 1 Global Population Dataset 2 Vulnerability Indicators (country specific) 3Source:
1. ThinkHazard! (https://thinkhazard.org/en/)2. Gridded Population of the World (GPW) (https://sedac.ciesin.columbia.edu/)
3. The World Bank (https://data.worldbank.org/indicator)
• To identify relevant vulnerability indicators, first create model with historical hazard dataset and isolate vulnerability indicators with highest predictive power (ability to predict resulting TAP).
© 2020 MIT Center for Transportation & Logistics | Page 90
ThinkHazard! Datasets
All stored on the Geonode:
ThinkHazard! Example
© 2020 MIT Center for Transportation & Logistics | Page 92
What is S&OP?
“Sales and Operations Planning (S&OP) is a set
of business processes and technologies that
enable an enterprise to respond effectively to
demand and supply variability”
Why S&OP? (Adapting to the context)
• Emergency response makes planning in advance
difficult
•
Trade-off considerations: speed vs. cost
• S&OP can enable the organization to predict the
potential impact of unexpected events
•
Show stakeholders the impact of unexpected demand
surges (“scope”) or supply shortages (“scale”) via
sensitivity analysis
© 2020 MIT Center for Transportation & Logistics | Page 94
Proposed S&OP Process: Bi-annual Planning
• CO and Regional teams assess potential hazard levels and threats and vulnerability of the population to identify anticipated demand. • Take Operating Year Budget into
consideration
• Factors considered: • Locations
• Type (Natural vs. Conflict) • Total Affected Population • Population planning figure • Sectors (WASH, Shelter)
Identifying Demand
1 2 3
Demand & Supply Planning Converge, align, and agree on plans • In this final stage, both Program and
Supply Chain come together to discuss major factors that may impact demand and supply (i.e. disaster type, budget, existing resources, etc.)
• Supply Chain analyzes current supply levels (owned or at external
organizations). Provide global commodity level forecast. • Factors considered:
• Locations • Type • Caseload
• Existing resources from
Gov/Com/Hum sectors in the region
• Items sent • Forecast items
• Missing opportunities
Data-driven Decision Tools
The MIT model seeks to improve the overall S&OP process by:
3. Assisting with quantification of disaster
“scaling” mechanism
• Analyze the impact on inventory allocation
1. Providing OFDA with an unbiased tool that can optimize the
inventory network
• Incorporate “scope” and “scale” set by S&OP process
2. Assisting with demand & supply planning
• Disaster forecast
© 2020 MIT Center for Transportation & Logistics | Page 96
Quarterly Review Ad hoc Review Metric Review
Additional Review Considerations
• Model updates may include new disaster data or inputs
• Business decisions may generate network changes
• Business decisions may generate budgetary changes
• Additional disaster data
• Additional warehouse locations • Additional mode considerations
• Changes to “scale”,
• Changes to target fill rates • Changes to inventory profile
• Changes to model constraints • Changes to optimization
Prerequisite for turning research into action: a decision-making process
• Optimal inventory allocation
• Forward-looking disaster portfolio
Decision-support tools
1. Identify demand
2. Demand & supply planning
3. Converge, align, and agree on plans
Supply Chain Planning
• Fastest response • Highest service level • Highest quality
• Lowest cost • …
© 2020 MIT Center for Transportation & Logistics | Page 98
Software development process
1.
Analysis and planning
2.
Requirements
3.
Design and prototyping
4.
Software development
5.
Testing
6.
Deployment
7.
Maintenance and updates
• Software development is informed by a stable and transparent decision-making process
• The process determines the requirements (inputs, outputs, functionality)
Research questions
Inventory Locations Research Questions
➢ How well does the current network perform against a portfolio of disasters?
➢ Should OFDA hold more or less inventory?
➢ Should OFDA redistribute inventory to improve performance, and if so, where should the inventory be located?
➢ How efficient is OFDA’s current prepositioning network based on existing costs and capacity? ➢ What alternative prepositioning strategies should
OFDA consider considering cost and response time? ➢ What metrics could OFDA use to inform
decision-making?
OFDA holds a strategic stockpile of key disaster response commodities to support people world
wide in crises situations.
© 2020 MIT Center for Transportation & Logistics | Page 100
Key steps
Data inputs?
➢ Disaster – scale / scope ➢ Items and their
characteristics
➢ Transportation modes, cost, and time
➢ Travel distances ➢ Supplier lead times ➢ …
Methods?
➢ Served population estimation
➢ Disaster scenario creation ➢ Supply lead time estimation
Outputs? ➢ Inventory allocation ➢ Inventory level ➢ Transportation mix ➢ Cost-time-trade-off Metrics? ➢ Inventory balance ➢ Inventory allocation ➢ Effectiveness metrics ➢ Efficiency metrics Model? ➢ Optimization model? ➢ Simulation/scenario model? ➢ Stochastic vs. deterministic?
Insights
Our results – methodology & model
Stochastic linear programming model
Methods to prepare data & model inputs
© 2020 MIT Center for Transportation & Logistics | Page 102
Our results – insights
© 2020 MIT Center for Transportation & Logistics | Page 104