Characterization and Prediction using SPCKen Bradley
How Lytica became a unique analytics company: Part 14
Characterization and prediction are two of the things we are working on in our Advanced Technology Center (ATC). When you look at what all of our effort in AI technology and data science deliver, it is really the ability to characterize your supply chain performance against accurate reference sources and then make predictions about how specific actions could improve your situation. All of our work is focused on getting reference information on cost, security of supply and compliance. We are continuing to build reliable and trusted information that can be used to make good companies more competitive.
Characterization consists of benchmarking, to determine how you stack up against a reference based on the performance of peers, as well as an assessment that gives insight into which factors impact the performance. For example, in a chemical vapour deposition manufacturing process, characterization would give insight into the effect of temperature, pressure, gas concentration and time on the film thickness being deposited.
Characterization isn’t new to manufacturing organizations as it is the basis of process control and improvement. Most companies today use advanced statistics and continuous improvement methods in their manufacturing lines. SPC (statistical process control), Six Sigma and Lean seem to be the most espoused approaches of leading and operationally excellent firms. These methods have been applied with great success over decades, yielding quality improvements, reductions in cost, gains in efficiency and more. SPC and basic statistics are especially interesting as they support process characterization. They identify process capability and help to distinguish the root causes of variation. These are must haves for any operations process and are applied, it seems, in most manufacturing processes except those that impact e-component purchase prices. Why is this?
Our work at the ATC has had a dramatic impact in our ability to capture e-component data and create reference distributions based on real prices paid by our cost benchmarking clients. This in turn allows us to benchmark spending performance and, as our priced component database has grown significantly in recent months, to begin characterizing why components may have been priced as they are. For this we use statistical control techniques. After all, purchasing, buying, negotiation or whatever you want to call it, is what determines your pricing. Pricing is part of a process that can – and should – be under statistical control and subject to continuous improvement. The application of statistical analysis and SPC is now possible because our Lytica Competitive Index (LCI) can be calculated for almost all components to provide market price normalization. This normalization has been the missing piece that prevented SPC from being applied to pricing. Normalization allows all components to be analysed as part of a purchasing system rather than being discrete, isolated and unrelated price points.
Let’s assume that all of the prices of all of the e-components (MPNs) that you buy are normalized through Lytica’s Competitiveness Index (LCI) and are plotted as a histogram of price competitiveness with frequency. What might you see? Your histogram shape might be thin (small sigma), wide (large sigma), skewed left or right or maybe bimodal.
What would this tell you?
- If your histogram is bimodal, it means that two factors are influencing the result. If you have turned over partial management of your pricing to an EMS, one of you is doing a better job negotiating than the other. Maybe it’s the EMS or maybe it’s you. Shouldn’t you find out who is better and take steps to get both teams performing to that standard?
- If your distribution is wide, one could speculate that your purchasing process is not in control and that your negotiators don’t have an accurate picture of what market prices really are. If the distribution is narrow, you might find the opposite to be true and would want to see how this distribution is moving over time by applying SPC techniques.
- Skew in the distribution may indicate that there are differences in tactics being used by different commodity managers and that, possibly, training is needed to bring all staff up to an acceptable level of skill.
There is a lot of speculation in the illustrative examples above but the idea is clear. The techniques that you have used to make your manufacturing processes better can now be applied to e-component prices because of market price normalization.