1. Do SKU and Network Complexity Drive Inventory Levels?
Good morning, thanks for coming, name is, over next few minutes we’ll share our thesis research on the question of whether
Joseph mccord & David novoa Garnica
Scm 2015

2. Take-aways
Greater complexity does not translate into higher inventory levels.
Inventory quantities mirror simulated inventory management heuristics rather than traditional optimal inventory models.
Two potential measurements of complexity: number of SKUs per brand and number of stocking locations per SKU.
In case your coffee hasn’t kicked in yet and you only catch the next 30 seconds of our presentation, we wanted to bookend our presentation with our key take aways. These are the very high level conclusions which we’ll explain more in depth through this presentation, and they are the following:

3. Outline Introduction Methodology Results Conclusion
Using an outline that will become

4. The Challenge of Complexity in the CPG Industry
Products are added to categories faster than they are removed
Many production and stocking locations
Wide product variety
How does this affect inventory requirements?
Project Sponsor: Unilever
Two billion interactions per day
+400 brands in 190 countries
Introduction
Methodology
Results
Conclusion

5. Two Specific Forms of Complexity
Stock-Keeping Unit (SKU) Complexity
How many SKUs are in each brand?
The more SKUs in a brand, the greater the forecast error for each SKU
Network Complexity
How many stocking locations are used for each SKU
The more customer-facing locations for an SKU, the greater the forecast error for that SKU
Mechanism
“Square root law” – going from 1 to n locations, IOH increases by sqrt(n) (under several assumptions)
Introduction
Methodology
Results
Conclusion

6. Research Question
Do increases in SKU or network complexity result in anticipated increases in observed inventory levels?
Introduction
Methodology
Results
Conclusion

7. Overall Study Design
Ordinary least-squares regression against a power curve
Replicated across geographies, product categories, and time periods
Also: developed a simulation exercise to model inventory levels under varying hypothetical inventory management approaches
This assumes in each case that the complexity factor is the sole source of forecast error. Obviously it isn’t, but this approach allows us to see what the level of correlation is
In several cases, where possible, we averaged inventory levels across several near points in time to get something closer to average stock levels
Introduction
Methodology
Results
Conclusion

8. Data Applied Source Content Use “MIO” Inventory Optimization Software
Ave daily demand quantities, forecast error per SKU and SKU location
Calculate complexity, demand data
Inventory Records
Inventory quantities per SKU location
Calculate inventory days on hand
Sales Records
Sales quantities per SKU
Eliminate obsolete SKUs
Introduction
Methodology
Results
Conclusion

9. Analysis Process For each iteration of the analysis:
Removed discontinued and obsolete SKUs
Linked inventory and demand datasets
Generated pivot tables
Created scatterplots and calculated regression statistics using spreadsheet software
Simulation process: created database with similar characteristics to actual datasets, calculated inventory quantities according to various common inventory models:
Base-stock policy
“ABC” (sales-based inventory control)
Introduction
Methodology
Results
Conclusion

10. Data Summaries – Daily Demand
Cluster 1:
Strong skew toward a daily demand of zero units.
10% of SKUs have a daily demand of 500 units or more,
More than 2000 SKUs have a daily demand of 25 units or less
Other cluster present the same characteristics
10 % of SKUs represent the majority of sales
Introduction
Methodology
Results
Conclusion

11. Nodes in the network Cluster 1: Cluster 2:
Different markets
Network complexities differ across clusters. Geographies and markets dictate network design
Introduction
Methodology
Results
Conclusion

12. Regression Results – SKU complexity
Cluster 1:
Strong skew toward a daily demand of zero units.
10% of SKUs have a daily demand of 500 units or more,
More than 2000 SKUs have a daily demand of 25 units or less
Other cluster present the same characteristics
No correlation and similar across clusters
Introduction
Methodology
Results
Conclusion

13. Regression Results – Network complexity
Cluster 1:
No correlation
Network complexity does not appear to act as an influencing factor.
Very wide range of inventory levels at each discrete network size
Introduction
Methodology
Results
Conclusion

14. Simulation results
Expected outcomes if inventory levels were in fact managed according to various known inventory control policies:
Inventory managed safety Stock Equation (CSL 95%)
Inventory Managed over Lead Time Using Safety Stock Equation
“ABC” Method
Graph: avg expected inventory + sqrt (sku complexity)
Introduction
Methodology
Results
Conclusion

15. Simulation results (2) Inventory managed using:
- Base level driven by SKU complexity and lead time
- Safety Stock Equation (CSL 95%)
Graph: base(avg expected + lead time)+ sqrt (sku complexity)
Introduction
Methodology
Results
Conclusion

16. Simulation results (3) “ABC” Method
A - Fast movers: low inventory levels (managed)
B – Intermediate movers: higher inventory (managed)
C – Slow movers: high inventory levels
Similar results to the actual results
Categirize sku depending on their average demand ignoring lead time and variability
Introduction
Methodology
Results
Conclusion

17. Interpretation
Do increases in SKU or network complexity result in anticipated increases in observed inventory levels?
Answer NO
Left: real
Middle: expected taking into account variability
Right: similar to what we had ABC method – local planners did not always relied in methods driven by variability
Introduction
Methodology
Results
Conclusion

18. Limitations
Several factors limit the ability to extend the results of this research:
Data analyzed only represents the experiences of one firm within the consumer packaged goods industry.
This study assumes independence of demand between products.
Other unexplored sources of complexity which could drive inventory levels could be production or sourcing lead time variance, frequency of product mix change, and overall variance of demand within brands.
Other firms in the same industry or firms in other industries may operate more centralized inventory control methods which would affect the relationship between complexity and inventory levels in more direct ways.
All products have same average demand
Introduction
Methodology
Results
Conclusion

19. Take-aways
Greater complexity does not translate into higher inventory levels.
Inventory quantities mirror simulated inventory management heuristics rather than traditional optimal inventory models.
Two potential measurements of complexity: number of SKUs per brand and number of stocking locations per SKU.
Introduction
Methodology
Results
Conclusion

20. Thank you!
Questions?