The Internet Of Things: A New Hope
The IoT is the future. And as the saying has it – the future is already here, it’s just not evenly distributed.
Tesla, for example, already streams sensor data from its vehicles and has analysed that data to make at least one significant change to its cars. All by merely pushing out a software update, without having to recall a single vehicle.
The IoT – and the sensor data that it produces – will enable us to optimise existing business processes in ways that we are only just beginning to understand. And allow us to create entirely new data-driven products and services. If you doubt how profound the changes are likely to be, just ask a London Cabbie what he thinks about Uber (stand well back).
The Hype is Strong with This One
But even the most optimistic among us commentators should accept that the IoT has been somewhat over-hyped recently. The 21 billion smart devices that are forecast to be deployed by 2020 aren’t suddenly going to be able to magically talk to each other, just because they each have an IP address.
There’s no C-3PO waiting in the wings – not yet, anyway – to translate all of the very many different sensor protocols that already exist at the start of this revolution, never mind the very many more that will join them in the near future.
And as we have already seen in parts 1, 2 and 3 of this series, even if there were, the data that those sensors produce are an unwilling, unreliable – and sometimes even downright deceitful – witness to the events that we want to understand. The smart money says that it is another few years yet before your fridge starts conducting reverse auctions with three different grocery chains to replace the milk.
The Data Strikes Back
All of which has implications for how we capture, store, process, analyse and manage the sensor data that underpin the IoT revolution.
Those data need to be managed, because sensors sometimes lie – or go offline – and we may need to interpolate and to extrapolate. And to understand where and how we have done so – so that we know if what we are looking at are raw data, or cleansed and smoothed data.
We should seek to capture sensor data at its lowest level, wherever we can. And to understand how it has been filtered and summarised where we cannot.
We will typically need to do lots of analytics before we can start the analysis that we care about, so we need to ensure that we can support “multi-genre analytics” – at scale – so that we can build a useful analytic data sets from noisy time-series data.
We need to think carefully about where we score those models – and how we deploy and manage them – in order that we can successfully and reliably operationalise IoT analytics, so that we can take insight out of the lab and use it to change the business.
And last but not least, we need to have a presumption for integration. Of sensor data with other sensor data – and with other transaction and event data from around the organisation. If you plan to capture raw sensor data in a Data Lake, prioritise integrating it with the transaction and event data in your Data Warehouse.
Awaken the Force
All of which means that we can instantly disprove that we are on the cusp of a revolution fuelled by the IoT on its own.
Recently, at Teradata Universe, a guest speaker from Monsanto gave us some insight into this when he said, “The thing people don’t like talking about is that IoT needs infrastructure. That’s not cheap or easy.”
When building out that infrastructure, it is easy to focus on the sensors and the networks that connect them. But don’t overlook the supporting data and analytic infrastructure