Cost-Regression Analysis

Sometimes, procurement and its extended cross-functional team will not have enough knowledge or power over suppliers to build a detailed bottom-up cost model as described under cost-based price modeling (G1) or factor-cost analysis (H2). Many suppliers will not be willing to provide a cost breakdown, or they will manipulate the data. In addition, the procurement team may be under time constraints to negotiate a final price and may not have enough time to develop a detailed model. In this case, cost -regression analysis is the right method because it uses data that already exists within the company to assess suppliers' cost positions.

Procurement will typically have access to two sets of information about each part or service:

Commercial data such as price, volume rebates, and delivery and payment terms

Specifications for parts (technical parameters such as weight, volume, material type, tolerances, and other technical parameters that define the part) or services (element of service level agreements, such as frequency, scope, technical knowledge, and the capability level of the people delivering the service)

This data can often be found in past requests for proposals. A method is needed to help understand the dependency of the price from the respective (mostly technical) specifications. Cost-regression uses statistical tool regression analysis to help estimate the relationship between many variables. In particular, it describes how the dependent variable—the price (p) —will change when the independent variables—the specifications (x) —change. For example, the price for using Uber services depends on three variables: distance of travel, type of car, and the current usage level of Uber cars in service.

Mathematically, the dependence of the dependent variable p from the independent variables xi is described as follows:

p = α_0 + α_1 f_1 (x_1) + α_2 f_2 (x_2) + ... + α_m f_m (x_m) + e

where α_i is a weighting coefficient, f_i (x_i) is any function of xi, m is the number of different parameters, and e is the error (residuum) of the model. In procurement, you have usually a linear dependency. Therefore the price -specification relationship simplifies to the following:

p = α_0 + α_1 x_1 + α_2 x_2 + ... + α_m x_m + e.

Given a set of data points with prices and specifications (pk; x1k; x2k; ...; xmk), the cost-regression analysis statistically calculates the weighting coefficients α_i in a way that the sum of the squared residuum δ is minimized—that is, identifying the set of α_i for all:

p_k = α_0 + α_1 x_1k + α_2 x_2k + ... + α_m x_mk + e_k that minimizes δ = ∑_1 ^ k▒e_i ^ 2.

Knowing these weighting coefficients, you can calculate a price for any combination of specifications xm, the target price.

As shown above, cost-regression analysis is a linear, multivariate regression model. This model helps you to:

Understand what parameters have an influence on the price (dependent variable) and how strong (quantitatively) this influence is

Forecast the price (target pricing)

Because this model is based on statistics, several preconditions need to be fulfilled to get reliable results:

Certain level of complexity of specifications (cost drivers): A minimum level of technical complexity of the products to be analyzed is required to prevent certain parameters from becoming dominant. If this happens, a simpler linear performance pricing can be used instead. We recommend this tool for products that are assembled from different sub-components, require a high content of manufacturing value added, or are based on different manufacturing technologies.

Enough variation among the specifications: Even if we have enough parameters identified, we will need sufficient variation of these parameters between the products to be analyzed. If one parameter shows only minor difference between products, the parameter will most likely be useless because the statistical model will not be able to determine its influence on the price.

Sufficient number of products to be analyzed: To ensure a proper statistical environment, a sufficient number of parts need to be included in the model. Based on past application among many different category groups, we do not recommend applying cost-regression analysis on a sample with fewer than 40 parts. If there are too few parts, the regression will have too much influence by outliers, resulting in poor quality of regression (statistically described by low R2 ratio).

For application in daily procurement practice, verify the availability of the data within the organization upfront. Procurement usually has all commercial data, including prices, supplier names, and volumes, already on hand from the ERP systems. Nonetheless, technical specifications often need to be collected from the engineering department in a structured format. The team will have to evaluate how much effort will be required to collect these parameters. Most of the time, it can be done by a junior engineering resource in a limited amount of time.

In general, a cost-regression analysis occurs in four phases, beginning with collecting and preparing data. In this phase, it is essential to define the right set of parameters that are expected to drive costs. We recommend doing this with your technical department or perhaps with the supplier, which significantly increases the acceptance of the results. The collected data needs to be checked for completeness and plausibility, and qualitative information has to be quantified. For example, for material specification, we may estimate material cost levels; for quality , we may use proxy to describe the level of quality; and for color, we may use price differentials between different color groups.

During the second phase, the regression model will be built and tested for its statistical relevance. In several iterations, the model will be adjusted to optimize the quality of the regression (increase R2) by potentially removing parameters that show mutual dependencies (such as weight and density or volume), and linearizing some parameters to better fit the linear regression model. After the model has been optimized, the results will be validated within procurement and with engineering or other departments that provided technical parameters. Here, outliers will be discussed to understand a potential rationale and see if any parameters have been omitted that should instead be included to explain the difference.

Finally, data will be extracted from the model to aggregate the results and prepare a negotiation strategy. Renegotiating with suppliers based on target price identification is the core application of the cost-regression analysis within procurement. With this model, procurement can quickly analyze the cost position of a particular supplier or of selected parts and identify opportunities to reduce costs. These reductions are calculated by taking the difference between the purchase price of a selected part and the respective value of the target-price curve. To increase your identified savings potential, you can also calculate the target price function based on the first one or two quartile parts. The results of this method give you a strong argument to renegotiate for rapid cost reductions with a supplier. Suppliers also welcome this approach because it gives them transparent feedback about their cost position with like-for-like suppliers as all suppliers included in the analysis are del ivering parts or services to the same company.

In addition to renegotiations, cost-regression analysis can be used for other applications. For example, it can be used to compare prices for the same parts from different business units. It can be used to identify which variants should be removed during a complexity reduction effort by comparing similar parts with similar technical parameters and identifying less competitive ones. Finally, it can be used for target cost during early stages of the product development process to be able to quickly calculate the cost of a new specification.