Improved demand forecasts at the last mile can save millions of dollars. Using big data and advanced analytics to drive supply chain efficiency

Transcription

1

2 Improved demand forecasts at the last mile can save millions of dollars Using big data and advanced analytics to drive supply chain efficiency November, 2017

3 3 Over the years, the focus of health supply chains has evolved from visibility to availability and now towards efficiency Visibility Availability Efficiency Data Systems Processes & Decision Making Digital Intelligence (At scale!)

4 4 Current methods of establishing stock thresholds for vaccines are rigid and inefficient, one size fits all does not work! 800 How are vaccine inventory thresholds calculated currently? # Doses Annual infant target * Dose per Infant * Wastage factor + Buffer stock + Lead time stock Jan Monthly inventory threshold Dec Inefficiencies Outdated & inaccurate census Effects of migration Fractional dosages Variability in wastage Variability in demand Proximity Seasonality and trends

5 5 Hypothesis: forecasting consumption based on historic demand at the last mile is superior to existing heuristic methods Province Nation Last Mile District Region Cascading inventory policy Consumption based forecasts Statistical models evaluated Measures of accuracy Moving averages (3m, 6m 9m) Moving median (3m, 6m, 9m) ARIMA SARIMA ARIMAX Holt Winters Seasonal trending Exponential smoothening Multivariate regression Random Forest MAPE MPE R 2 Adjusted R 2 Consumption based forecasts outperform heuristic methods 95% of the time by an average of 49 pp

6 6 The consumption based model was piloted in 1 district of a large multi echelon vaccine supply chain Universe 374 districts in 12 provinces Pilot 1 district in 1 province 11k Facilities 35 SKUs 6 Echelons 36 Facilities 11 SKUs 4 Echelons 1.8M Transactions annually 17M Birth cohort served 650M Beneficiaries served 220k Transactions annually 118k Birth cohort served 4.5M Beneficiaries served >95% Availability >96% Reporting Rate 100% Availability 100% Reporting Rate Data as of November 8 th 2017

7 7 The pilot has been running for 4 months now, the intervention was frictionless and required no additional training Order Interface Pilot Statistics Duration 4 months/ 6 months Facilities 36/36 active Users 36/36 active SKUs 11/11 active Order Status Echelons 4/4 active Pending Confirmed Shipped Fulfilled Orders 2,090 SKUs till date Doses 6.3M doses till date

8 8 In the pilot, inventory reduced by 24.7% ($0.2M cost savings over 4 months) and caused 0 stock outs in the same period 1.8 Impact achieved Inventory reduction: 24.7% Cost saving: $0.2M in 4 months Availability: 0 stock outs Inventory (# Doses M) % -1.1% -2.5% -3.0% -3.3% -5.3% -9.7% Total Population Max +$0.4 k -$3 k -$8 k -$19 k -$14 k -$138 k -$19 k BCG bopv Hepatitis B JE Measles Pentavalent TT Total Consumption Max Cost per dose as per UNICEF IPV, DPT, Rotavirus & Pneumococcal excluded

9 9 Once scaled, inventory will reduce by 17.3% thereby saving $73.6M annually, reducing availability by just 0.05 % 600 Impact projected Inventory reduction: 17.3% Cost saving: $73.6M annually Availability : 0.05 % reduction Inventory (# Doses M) % +0.8% -0.5% -2.5% -3.1% -3.7% -9.2% Total Population Max +$0.8 M +$1.3 M -$1.8 M -$6.1 M -$5.3 M -$53.7 M -$10.0 M BCG bopv JE Measles Hepatitis B Pentavalent TT Total Consumption Max Cost per dose as per UNICEF IPV, DPT, Rotavirus & Pneumococcal excluded

10 10 Minimizing the time duration materials spend in the pipeline is the single most effective way of lowering total supply chain cost Less procurement: $73.6M per annum Lower handling costs: Reduced pipeline stock Lower probability of expiry: Lower probability of temperature excursion: Less cold chain capacity:

11 11 Conclusion Simple forecasts can save >200M$ annually in the vaccine supply chain of just one country Implementing the forecasts is a frictionless change The underlying technology, methodology and insights can be applied to practically any health commodity

12 Appendix 12 The 3 month moving average method (adjusted for stock out days) was chosen as it was accurate and also easily explainable Sample 3 month moving average based forecasts # Doses stock out day 2 stock out days 10 1 stock out day Forecasted Consumption Max = #$%#&%#$ '#%()%'# *(#%(%#) = 18 0 Jan Feb Mar Apr Actual Consumption Forecasted Consumption Max

13 Thanks to our generous sponsors 10th Global Health Supply Chain Summit, Accra (Ghana); Nov 15-17, 2017