Using Machine Learning to Generate Quick Savings and Improving Price Consistency

It could quickly become onerous for a large group and their Purchasing team to ensure price consistency in a specific category.

• Why pay $ 5.6 for a 10cm long, 2mm diameter extruded tube and $ 20 for a 20cm, 2.5mm diameter tube?
• Or a certain price for a red product and a double for its blue version?
  These shortcomings may be normal but need to be controlled and understood so that suppliers or R&D can be challenged.

Even with corporate organization and systems, business units are often independent or, worse, siled. They could implicitly refuse to coordinate globally because they have to move quickly and remain independent. But, beyond organizational and cultural behavior, the main obstacles are technical and IT:

• Products have different specifics and characteristics making it difficult to compare prices and match
• There is no common repository combining technical and economic information
• Price history is rarely reviewed
• ERP system not warning of price differences
• The price of small suppliers is not properly controlled
• Price review routines may vary from group to group
• Multi-contract with similar suppliers
• Suppliers apply a different cost for doing business per BU / region

It’s for sure that the traditional approach should cost and the solutions can help solve these problems.

However, they are often complex to implement on a large scale, as they require a lot of information on manufacturing processes, strong technical expertise and long and tedious cooperation with suppliers (open book policy is mandatory).

Consult our article “Optimization of the Supply Chain: avenues for a full cost approach“

Many companies have developed an innovative and efficient approach combining advanced analytics and predictive algorithms (derived from artificial intelligence) to generate opportunities for rapid savings, only by processing your existing data.

The Methodology is Based on 5 Pillars:

• Create a cost model based on shared cost factors taking into account BU / region specificities
• Use benchmarking to identify gaps in BUs / regions and optimized sourcing strategies
• Identify overvalued products and quantify savings opportunities
• Use data mining to determine optimization levers and prepare arguments for negotiations with suppliers
• Specify the management systems and needs required to support the process

The Following Advantages Have Been Observed:

• Identify savings opportunities (negotiation, VAVE, resourcing)
• Generate quick wins through “analysis-based” negotiations with suppliers
  Control supplier quotes on new projects
• Improve procurement strategies through optimized allocation of suppliers by cluster
• Drive continuous improvement through a robust cost model, increased skills and improved cross-functional collaboration

Machine learning-based valuation solution that estimates the price of a new product or service by processing current / historical data using a sophisticated algorithm, such as random forests. Random forests are a nonparametric statistical method that allows learning over multiple decision trees based on slightly different data subsets generated by Bootstrap techniques (see Breiman, L., Random Forests, Machine Learning, 45, 5 -32 (2001)).

This type of method makes it possible to estimate the price of a product / service on the basis of pre-identified parameters called “cost-drivers”. The estimation is very fast and precise (30% more precision compared to traditional statistical methods).

Compared With Traditional Methods, the Main Advantages of Costing Based on Machine Learning are:

• The user should not be an expert in the production process of the product.
• The estimate is based exclusively on the characteristics of the product (“cost drivers”) which are information that it can access internally (vs. asking its supplier).
• It can mix an infinite number of cost drivers, the latter being able to be continuous or discrete, technical (weight, color function, raw material,…) or commercial (country, volume, supplier,…).
• Ability to process databases for which the number of variables greatly exceeds the number of observations
• Ability to automatically identify and weight the most important parameters, and therefore the cost factors that have the most impact
• Ability to interpret results
• Ability to manage missing values ​​/ incomplete database

For all these reasons, machine learning-based software helps create a very robust, out-of-the-box costing model.

Beyond the above, the machine learning based costing solution processes all current purchase prices and identifies inconsistencies / deviations from estimates, making it easier to identify savings opportunities , including negotiations with suppliers.

Finally, some solutions integrate benchmark functionalities which make it possible to compare each BU / Region for a specific category (even through products of different design and characteristics). They are used more and more in the industry and, therefore, they add external reference knowledge for each product every day (intellectual property and confidentiality respect are respected). This makes it possible to create benchmarking communities and to share more (life sciences, automotive, etc.).

Analytical Model

Benefits

• Explanatory model and focused on operations
• Estimation of the “best landed price” and definition of the target price
• Allows you to optimize production prices and control progressive supplier plans

Limits

• Difficulty accessing and maintaining process references over time
• Intrusive approach to suppliers
• Expert model difficult to deploy
• Timeout for setting up and running encryption
  Precision?

Software Examples

• Siemens PLM
• Apriori
• Facton

Statistical Parametric Model

Benefits

• Easy and quick to use
• Consistency of estimated prices and accuracy (conditional)
• Non-intrusive approach to suppliers
• Product and service applications
• Very relevant in the upstream phases of the life cycle and for consistency analyzes

Limits

• Requires minimal data and good history
• Model that is not very “explanatory” to moderate the progress plans of suppliers
• Less relevant model for setting target prices and the “best landed price”
• Difficulty in modeling qualitative parameters

Software Examples

• Seer
• EstimFEC

Non-Parametric Statistical Model

“Random forests”

Benefits

• Easy and quick to use
• Consistency of the estimated price and accuracy increased by 30% compared to parametric models (conditional)
• Non-intrusive approach to suppliers
• Product and service applications
• Very relevant in the upstream phases of the life cycle
• Also relevant in downstream phases for the analysis of price consistency and the identification of opportunities thanks to the explanatory properties of forests
• Incorporates a lot of cost factors, including qualitative factors
• Detects technological breakthroughs
• Prioritize cost drivers
• Handles missing values ​​and can work with a limited sample

Limits

• Less relevant model for establishing target prices and “best landed price”

Software Examples

• easyKost