We hear a lot about ‘the value of data’ (including here) and how companies need to monetize their data better. And how every company needs to go out and hire a bunch of Machine Learning experts and data scientists, and then monetize that data in incredible ways.
So surely, there must be dozens of examples, right at the tip of your tongue of companies that have done that. Go ahead, I\we’ll wait…Yeah, I couldn’t think of many either. If you answered ‘hedge funds’, that is incorrect and as remedial homework you need to go watch all of season one of Showtime’s Billions. The obvious examples are Google and Facebook. But what about everyone else? What about companies that deal in analogue not digital products? The best example we can think of is the way that many of industrials companies are tying data analytics to service contracts. You could also argue that there are companies like Blackhawk systems which are doing something with consumer credit data, but even there we not convinced that is really how they make money. We can think of a lot of companies that are saving money through better data, but very few that are actually generating revenue from data.
Part of the problem is that most data is not really that useful. We recently spoke to the CTO of a large industrial company that manufactures big industrial systems. Like everyone else, they were trying to develop an IoT strategy. We sat with him while a software vendor was pitching their vision of the future, full of monetization possibilities. He was polite, at first, but after a while he broke in and said, “Before you go any further, you have to realize that almost all the data we capture is wort nothing.” That conversation only went down hill from there. However, he made a valid point. His company have been adding all kinds of sensors to their equipment for years. They could capture petabytes a day, probably more. But 99.9% of that data essentially translated into “Status: Unchanged”.
We are not arguing that all data is worthless. However, we think it is clear that capturing value from data in the physical world is still a very poorly understood process. During the last Bubble in the 1990′,s we read a profile of a software company that had pitched its order system to a mid-sized produce distributor. After months of evaluation, the distributor determined that their existing fax-based system was still much more efficient than the fancy web-based system. It probably took another decade for software to bridge that gap. We think it may take that long for machine learning to make much difference to most companies.
There is hope. We recently met with a recently founded company in the machinery industry. They had put a serious amount of sensors into all of their products from Day One. These sensors probably added 10% to the cost of their product. However, they have been gathering that data from customers for some time now, and actively use that data to refine the operation of their product. They were able to do that because they are only a few years old, and designed their product from the beginning with sufficient computing capacity to enable ongoing improvements to its performance. Retrofitting that amount of compute to decades-old designs of large machines is probably not feasible. So it will take many years to redesign it all.
There are many other examples of this in all analog industries, and we are in very early stages of applying machine learning to ‘big data’ or any data.
In future posts, we plan to explore interesting models for extracting value from data, but we think it is safe to say that this is going to take a while.
Pingback: Start Up: AlphaGo conquers chess, enter Amazon, Silicon Valley’s model problem, bitcoin’s future, and more | The Overspill: when there's more that I want to say·