In season two of HBO’s Silicon Valley series, VC Russ Hanneman recoils in horror when startup CEO Richard Hendricks mentions plans to generate revenue. “If you show revenue, people will ask how much, and it will never be enough!” Hanneman warned.
New technologies sometimes enjoy a golden phase when nobody focuses on financial results. Early providers sell the promise and early adopters cut a path for the technology. Then one day, things change and companies find themselves in a world that demands measurable results.
So it has been with the industrial internet of things.
IIoT has developed along two parallel tracks over the past several years. On the marketing track, IIoT has been in the golden phase where its promised benefits are infinite, anything is possible and everything can be measured.
This effect has been sweeping. Companies with decades-old technologies have suddenly awoken to find they’re in the internet of things business. Every company that can conceivably claim its products are IIoT-related, from physical connections on up, has focused its message on the overall expected benefit when the full-scale IIoT will be up and running.
But on the parallel implementation track, the wheels are just creaking into motion. Most of the focus among IIoT technology suppliers has been on connectivity: connecting machines and collecting the data.
The result? Connectivity, while clearly essential, is not enough. We hear the laments over and over again from manufacturers:
- “We spent all this money, we built a data lake, but it doesn’t do anything for us.”
- “We’ve finally got the data together and found that half of it was bad.”
- “Our data scientists spend most of their time conditioning the data.”
- “Each analysis project is a one-off; we haven’t been able to repeat or scale any of them.”
One of manufacturing’s virtues is its practicality. For industry, the acid test of new technology is simple: does it increase your productivity? Industrial companies have quickly discovered that getting all their manufacturing data together by itself doesn’t solve any problems. The challenge is making sense of that data.
That’s hard. And in manufacturing it’s probably more difficult than in any other business because of the “variety problem.” The typical shop floor includes hundreds, if not thousands, of data sources, each continuously generated in multiple streams. Time series data, ordinal, image and point cloud, text, serialization — all on different time frames, from different sources and of different types.
To use a horribly mixed metaphor, a data lake filled with machine-generated data usually becomes a liquid Tower of Babel. The companies that have been able to make sense of their data often do so by throwing teams of data scientists at specific problems. That might possibly work for that one problem, but there is often no way to apply the result to other challenges. Applications and models, each built for a specific need, accrete, and then the problem becomes how to get all of them to talk to each other.
Relying on teams of data scientists is also expensive and hard to scale. According to a survey by CrowdFlower, data scientists spend 79% of their time on low-value tasks like collecting, cleaning and organizing data, and only 13% of their time mining the data and refining algorithms. In order to start using their data productively, companies need to shift that 79% of a data scientist’s day away from data conditioning and into data analysis.
The goal is to use data scalably. Those companies that figure out how to systematically collect, condition, model and analyze their data will secure a strong advantage. The rest will invest heavily in connectivity, only to find they are drowning in data.
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.