Price prediction and beyond

How Lytica became a unique analytics company • Part 2

The fact that no one knew what I should be paying for a component puzzled me.

Each year I would receive cost reduction targets for materials. These were dictates communicated by accounting and were either set to achieve margin goals or by using some pseudo-scientific method such as putting a wet finger in the air. A third approach was to target a repeat of last year’s cost reduction number or guess a number close to what the supplier sales people were whispering. In reality, maybe none of these methods were actually used but there also seemed to be no real connection between the target and what the market was doing or how my company was positioned in that market.

During the time that I was CPO, Nortel was the darling of the stock market. It was growing at unbelievable speed due to leadership in optical networks. Every manufacturer and supplier wanted to be in our supply chain. Everyone was offering me great prices. I must have had the best prices in the world – everyone told me that I did. I had 1600 employees around the world and they all assured me that I had the best prices. Suppliers and manufacturers would visit and promise me that no one had better prices. Yet, when I would take a circuit card to one of the top EMS, their costed BOM would come back with some components priced better than mine.

How could this happen? I would ask my staff, the manufacturers and the suppliers, all of whom had assured me that I had the best pricing, and I would be answered with blank stares and uncomfortable body language. I would go to expensive consulting companies and ask about my pricing. I was told my pricing was very good but I could do better on some components. No one could tell me exactly which ones were priced right and which ones had opportunity. If I held up a component – say a 10 UF 50V electrolytic capacitor – and asked what my price should be on this, no one could tell me. I could get quotes but there was no real information to say if I should accept them as reasonable or not. Across the high tech industry this reality was not unique. I was not alone. Still today, I would not be alone.

Various methods are often tried to give an indication of what one should pay for a component. These include white sheet analysis, discounting internet or distribution pricing, phoning a friend in another company and sometimes being friends with the salesperson. None of these are adequate because they do not factor in your unique circumstances, market trends or the value of intellectual property. White sheet analysis attempts, through educated and uneducated guesses, to estimate the manufacturing cost of a component and apply some level of margin to get a price. In the end it is a lot of work and still a guess. Discounting distribution pricing is also a guess as to what discount is appropriate for you. Even if you get it right once, does the same discount apply to all part types? Asking a friend may work but are all of the conditions that influence their pricing the same as yours? The bottom line is that before Lytica’s no one really knew.

The problem of knowing how much cost reduction was available in the market place and which components had cost reduction potential had to be thought through to identify a new solution. For this process, it is useful to consider two prediction parallels – weather and blood pressure.

Before we explore these, let’s agree that you cannot extract more cost reduction than the market has to give through negotiation. As a result, in order to meet aggressive margin targets beyond what negotiation can deliver additional measures such as design change or material substitution on the cost side and having pressure applied to Sales & Marketing to sell at higher prices need to be taken. It’s a bigger problem that cannot be fixed by negotiation alone.

If you ask a procurement professional what determines a component’s cost you will get a list of variables that will contain items like relationship, volume, payment terms, return privileges, manufacturer’s cost and more. My list currently contains more than 50 items impacting price. Can we put these 50 items into an equation which can be solved for the price on each component? Not likely. I am told that weather forecasting needs 9 or 10 variables and their ability to predict beyond a few days is still basically horrible unless you live in a climate, like San Diego, with perfect weather every day.

Let’s consider blood pressure. We all know many of the factors that contribute to high blood pressure and we can list some of them; heart rate, weight, water retention, blood vessel elasticity or constriction, stress, blood viscosity. Again, the result is a list with too many variables. I know of no doctor that predicts blood pressure from an unsolvable equation. The ones that I know measure it and compare the measurement to reference tables made up of blood pressure data from healthy and unhealthy people.

This characterization, or benchmarking, can be done with component pricing if you have access to the appropriate price reference tables. Characterization is the first part of the prediction solution and was the first feature in the early prototypes of The second part of the solution comes with the realization that you do not need to predict prices on components that benchmark well. You are not likely to get much cost reduction on a component with a market leading price or a price that is suitable for you. However, a component that is mispriced will fall outside of your characterized competitiveness level and can have an appropriate price estimated. By using statistical methods aligned with the price distribution and the benchmark characterization of the component of interest, an appropriate price can be estimated. Mispriced components are outliers; something is not normal about their pricing and that needs to be corrected. If a student who always gets an “A” on math tests comes home with a “D”, something is wrong. If you fix what was wrong and they rewrite the test, they should get an “A”. In procurement, you can always rewrite the test.

Much like predicting your health through blood pressure characterization, predicting your price competitiveness through works because, like your blood pressure, your pricing is determined by a system. That system is influenced by all of the factors that went into the equations that could not be solved. All of those factors are important and all have an impact. Correlations can be found within these parameters (such as price with payment terms or blood pressure with weight) but this is not prediction. Prediction needs characterization. If you change the system, you can change the result. Losing weight, exercising and reducing salt intake can lower blood pressure just as shortening payment days, building better relationships or consolidating volumes can reduce component prices. If you don’t like where your characterization positions your competitiveness level, change the system. If you want to know how much you should pay for a component that is an outlier in your current system, use Lytica’s prediction technology.

With a theoretical solution at hand, we now faced many hurdles to prove its utility.

The first big hurdle was to get enough companies to give us their data so that we could form statistical distributions with enough data points on each component to be relevant. This problem had two elements, one legal and one related to scale.

The legal issues centered around confidentiality of data and limitations on how we could use customer information. We made some early decisions that helped, such as not revealing customer names and only using data in aggregate, which led to contracts that were acceptable to the legal teams of our customers. Additionally, customers could always choose to withhold any information that they didn’t want to share. The second element, the scale issue, involved getting enough data to achieve critical mass at our fledgling stage. This was helped by the fact that Lytica was a consulting company with solid customer relationships. Many of these customers were intrigued by our concept and became early adopters. The fact that our first reports, though preliminary and thin, saved them significant money didn’t hurt (and also proved our hypothesis). Word of mouth amongst the industry brought more customers and as our database grew so did the quality of our reports. Today EMS is our largest vertical market, followed by industrial, communications and medical as well as customers in all other major market segments. Many of our customers are in the top 10 ranking for their vertical with sixty five percent of them headquartered in North America and the rest in Asia and Europe.

New challenges emerged as we gained more and more customers. One significant one was that our match rate leveled off and did not grow as anticipated with the addition of data. Our investigation into this revealed a surprising situation, in many cases we had too much data on components and, as expected, for some we had too little. A deeper dive to root causes traced the problem of too little data to typos and input errors in clients’ Manufacturing Part Numbers (MPNs). We had too many single component groups with mistyped MPNs which would have been assigned to larger groups had they contained correct inputs.

Added characters and typing errors had segregated them into single occurrences. The problem of too much data was caused by anomalous price group distributions that failed our filtering tests. There were many sizeable groups where valid pricing statistics couldn’t be extracted.

We have engineered solutions to overcome these challenges but the enormity of the implementation task vastly exceeds our capability to accomplish it using traditional engineering approaches. I need a massive increase in engineering productivity and Artificial Intelligence is the key to enhancing this. With AI, Lytica can reorganize our data, fix MPN input errors, mine data for relevant new information and more. By focusing a significant portion of investment on AI in our Advanced Technology Center we can address the known issues, introduce enhanced features in existing products and release new products that offer fresh solutions to the problems of cost, security of supply and compliance.

All of this is great but it raises two questions pertaining to our journey; how does a small company access A.I. in a meaningful way to solve the productivity problem, and why does Lytica need an Advanced Technology Center to do it?

Follow this blog series on SCM Roundtable to find out the answers to these questions…and more!

Ken Bradley is the Chairman/CTO & founder of Lytica Inc., a provider of supply chain analytics tools and Silecta Inc., a SCM Operations consultancy.

Ken Bradley
Ken Bradley

Ken Bradley is the Chairman/CTO & founder of Lytica Inc., the world’s only provider of electronic component spend analytics and risk intelligence using real customer data.

Articles: 116