How Lytica became a unique analytics company: Part 9
Sometimes I feel like a 7-year old riding a bicycle. I am moving forward quickly, taking a daring step as I try to maintain control and balance with my hands in the air.
We are at a point with the ATC and our first round of AI development projects where things are getting very exciting; it’s exhilarating from a number of perspectives. Trusting that a machine can learn and draw conclusions much as a person does is unsettling yet invigorating. It’s like being in or out of adrenaline – charged control. The normal hands-on engineering discipline with its attention to detail is supplanted by the training of a machine with examples of correct and incorrect data and watching it learn. In reality I have never really known how others think but I can ask them about what they are doing and how they are doing it; this piece is missing from the machine’s education! The Russian proverb, Trust but Verify seems appropriate but not too comforting.
The scale of what we are doing in a relatively short time is astonishing. It’s outpacing what can be achieved by mere mortals and my developers are being turbo charged by a 7000 times productivity boost; that’s a 0.01% human to 99.98% machine ratio. Our three major projects in part number cleansing, component match rate enhancement and price prediction are coming to fruition. We are accommodating tens of thousands of manufacturers and hundreds of millions of components with times measured in minutes – or better. Our targets seek perfection in areas we seek to improve.
The intricacy of these learning systems, combined with the scale of our initiatives, make verification complicated and difficult. We have had to think outside the box to find ways of simplifying this complexity. As part of our approach, we have been involving our customers (as evidenced by those who are helping with MPN cleansing) but the basic task is just an extraordinary amount of work. Verification is the stage where we prove that the machine knows what it is doing because we can’t just ask it.
I thank those customers who have supplied us with data to help in our cleansing verification. We have been using their information for training and adjusting our systems. While we have not yet provided any results to them, their input has also helped us to understand the state of the industry better and appreciate how much more stringent our standards have become. Some of the industry problems we see and believe we have overcome in our development environment are:
- Incomplete recognition of manufacturers by name
- Out of date manufacturer names post acquisition
- Not recognizing manufacturer part numbers when packaging information is missing from the character string or when characters have been added
- Incomplete recognition of manufacturer part numbers
- Strong bias towards North American suppliers with limited Asian content
I believe that some of these deficiencies are caused by the 80/20 rule being used in the pre-AI era. This was a time when the cost of working to completion outweighed the benefits. I have never liked the 80/20 rule and in the information age, as computers made it increasingly easy to manage the long tail, it doesn’t make sense. What I really dislike is that 80% plus 20% equaling 100% suggests completeness – it is not. It’s not math; it’s axiomatic drivel. The result is that none of the databases we have seen are complete enough to meet our needs, hence our AI initiative.
I have hinted before, if not outright shouted, that AI will change the game in supply base design and supplier management. Lytica is getting close to driving this change as we roll out better results and new capabilities to our customers. Look Ma, no hands!