Analytics for IoT - Turning vision into reality


Have you heard of Kevin Ashton? The British technology pioneer co-founded the Auto-ID Center at the Massachusetts Institute of Technology (MIT), the organization responsible for developing an international standard for RFID. But more importantly for us, more than 20 years ago, he had a vision of computers gathering information about everyday objects, and predicting when things would break, so that they could be repaired in advance. Does this sound familiar? Today we call this the Internet of Things (IoT), and the pre-break repairing is predictive maintenance.

It has already become standard to gather information from ‘things’, whether production equipment, computer tomographs or agricultural machinery. Controlling these complex devices and machines is only about intelligent behaviour, and this requires three steps:

  • Collect data;
  • Analyse the relevant signals from the data, interpret them and reveal interesting relationships; and
  • Make suggestions for action.

But here’s the thing: The IoT only makes sense with analytics! And that is a big part of the problem. According to a study by the MIT Sloan School of Management, 58% of companies still consider that they lack sufficient analytical skills to get value from the IoT. Why is this?

The Three Vs of Big Data

We can put it down to the three Vs of Big Data: Volume, velocity, and variety. Organisations struggle to manage the sheer mass of data acquired from the IoT and sensor technology. It is no longer possible to take a “quick look”. Descriptive analytics or Excel are not enough to gain any insights from the huge mountains of data. The runtime for a particular query is too long, and users feel like they are looking for a needle in a haystack.

Speed is also a hurdle. Data emerges so quickly and is so short-lived that decisions must be made immediately, where they are created — at the edge. But how can you decide which events are important, and which ones can be ignored? What data should be removed for later analysis? All these questions must be answered immediately and close to the IoT device. The relevance of the data must be assessed in the data stream, and decisions made in real-time on the basis of the IoT data analysed. The classic paradigm “Save => Analyse => Decide” no longer works.

Finally, it is easy to take into account a huge variety of variables because questions and analysis go beyond simple rules and descriptive statistics. Decisions in the IoT environment must be taken in a multivariate context. For example, suppose machine failure occurs when certain sensor values ​​are in combination. If so, it is not enough to look at a single value. Instead, you have to look at several variables at once.

No ‘rule of thumb’

One final point: Decisions in the IoT context are immensely important. A machine failure has high costs. In the medical field, health and patient safety are at stake. Therefore, the models MUST be robust and allow reliable statements. A quick ‘rule of thumb’ is not an option.

Two studies demonstrate why companies often fail to analyse IoT data effectively. An IDC survey showed that only 15% of analytical companies see themselves in a position to successfully operationalise and integrate analytics into their business processes. And a survey conducted by SAS found that data analysis in real-time was described as the number one challenge in IoT initiatives by 21% of respondents.

There is a lot of variation in analytics, not least because the IoT involves more and more devices, sensors, interfaces (actuators) and data streams. Shortened representations such as “Analytics = Gaining Insights” are attractive, but do not actually provide solutions. Analytics without operationalisation has no value in the IoT area. In view of these obstacles, powerful analytical platforms seem to me to be essential. These platforms can automate, scale and manage complex modelling, parametrisation, model comparison, deployment and model monitoring. I advise companies who want to use IoT and analytics to link their classic data to IoT data and to filter relevant data intelligently. They should be able to analyse data in the data stream and carry out modelling in the terminal. Ultimately, they will realise value ​​by engaging deeply in business processes and storing business rules so that analytical findings can be converted into meaningful actions.

The data is there, the algorithms are there, and analysis in real-time is also possible with high-performance analytical environments. Everything is ready to make IoT a viable playing field, and transform Ashton’s vision into reality.


About Author

Andreas Becks

Manager Business Analytics DACH

Dr. Andreas Becks ist als Manager Business Analytics DACH mit seinem Team aus Business Experten und Data Scientists für die Beratung rund um die analytische Plattform von SAS verantwortlich. Themenschwerpunkte sind datenbasierte Innovation auf der einen und Industrialisierung von Analytics auf der anderen Seite. Seit mehr als 15 Jahren konzipiert Herr Becks innovative Lösungen für datenbasierte Entscheidungen in komplexen Geschäftsanwendungen. In seiner Zeit vor SAS hatte er verschiedene leitende Positionen in Forschung und Entwicklung, als Business- und Lösungsarchitekt sowie im strategischen Produktmanagement eines Softwarehauses inne. Andreas Becks ist studierter Diplom-Informatiker, hat an der RWTH Aachen im Bereich Text Mining und Wissensmanagement promoviert und ist Master of Business Administration der Universität St. Gallen. Besuchen Sie ihn bei LinkedIn oder auf Xing .

Related Posts

Leave A Reply

Back to Top