Data Defines Your Destiny in IoT
By Lou Lutostanski, Vice President of IoT, Avnet
Data can be daunting. It’s intangible, invisible and typically isolated in physical things like the trucks in your fleet, monitors in your hospitals, or machines in your factory. Once extracted, aggregated and analyzed, however, this data can prove enormously valuable. It’s the raw material from which actionable business insights are forged.
The best decision you can make with your data is to harness it in a way that creates value for your business. Falling hardware prices and soaring computing performance mean you can tap data in places you never would have imagined a decade ago. Companies that fail to seize this opportunity are taking a serious risk.
Here, then, are a dozen best practices for addressing IoT’s data challenge.
1. Define the use case
Given what you know about underused data in your company or solution, what exactly would you like to accomplish? Let’s say you want to control natural light in your hotels to save on utility costs and make guests happier. That’s the use case. You’ll need sensors in your rooms that measure ambient sunlight and temperature (the data) so that the blinds will open and close accordingly.
Another use case could be creating a data-generating platform on which a myriad of applications can be built. Automakers do this as they embed their vehicles (the platform) with sensors and computing power that amass data for a multitude of current and future applications. A smartphone is a similar platform, offering a camera, accelerometer, GPS and other capabilities to capture data for new apps in the future, each one a new use case.
2. Quantify the value proposition
Value creation is the most important part of your plan. In our building automation example, the value is your savings on utility bills and the enhanced profit resulting from higher occupancy. Although your use case may be enhanced customer comfort, the value is expressible in dollars and cents. One caution: IoT can unearth all kinds of elegant, intriguing and exotic data, but that’s not the same as generating hard value for your customer or your own business. Mind the gap.
3. Harvest your history
Good IoT implementations “learn” by refining algorithms based on experience as businesses encounter anomalies and find solutions. But experience doesn’t start on day one of your IoT implementation. It started years ago. Whatever business you’re in, it’s likely you have decades of historical data that can inform your predictive algorithms today. What were your seasonal outputs? How were you staffed? What time of the year saw the most equipment failures? When did sales spike and ebb? Even discarded machines can be sources of important historical data to inform your algorithms.
4. Break down the silos
You might plan to use IoT to manage inventory, but the inventory itself is just one source of IoT data. You might also analyze data from HR to compare staffing to factory output, CRM to look at demand projections, finance to glean materials costs, third-party logistics to predict delivery timing, and macroeconomic indicators to gauge general business activity patterns. Plants can be silos, too. Have you pooled the data for all of your factories around the world?
5. Choose a sample rate
Some IoT applications—say, a manufacturer checking on consumers’ connected refrigerators—may require a data sample only once a day. Other applications, like monitoring a global manufacturers’ production lines, may require analyzing thousands of data points per second. The goal in both cases is harnessing just enough data to predict events, like breakdowns, before they occur.
6. Ensure your data is clean
Data can be dirty in a thousand ways, but imagine that your smartwatch misses one out of every five steps you take when your hands are in your pockets. That’s dirty data. Any insights derived from such data will be useless. Think it through early. Once you’ve confirmed your data is clean, then talk about augmenting it.
7. Ensure data is complete
The data you collect from your devices will often need to be supplemented by additional information. If a generator in your factory is vibrating, for example, the manufacturer can determine the severity only if that vibration is compared against thousands of generators that have failed over time. In my experience, half of IoT leaders will realize along the way that their data is incomplete.
8. Choose a storage site
If you want to perform complex analytics on massive data sets, you’ll need to store it. The cloud is one cost-effective (but not free) option. A data warehouse might be an even more cost-effective foundation for adding use cases and scaling your IoT over the long term.
On the other hand, a quality-control device on a factory production line may never store a byte of data; it will scan parts, discard the data, and alert key staff only if there’s a product flaw. Maybe you need a hybrid setup, as with a smartwatch that collects all your data on a chip then transfers it into your smartphone when connected.
9. Choose where to put the intelligence
Closely related to selecting a storage location is selecting the right place to put your artificial intelligence (AI), which analyzes data, consolidates it, and shares resulting predictions with the IoT. Many organizations assume the right place for AI is in the cloud since that’s where they’re moving their data, IT and computing power. That’s not necessarily the case. Many of the AI and machine learning applications that are reshaping our world require real-time responsiveness (consider a driverless car approaching a jaywalker). When real-time response and low latency are critical, you need AI processing data at the edge. Before long, however, 5G will move data 100 times faster than today, for the first time enabling real-time IoT insights derived from cloud data. The possibilities are huge.
10. Keep learning
Your IoT should never stop improving. Imagine your operation suffers a disruption today, and you solve it. That disruption-solution scenario is new data that can solve that problem instantly the next time it occurs, anywhere in the world. In ideal cases, this machine learning occurs automatically.
11. Be realistic about security
Security is among companies’ biggest fears and a common reason they postpone IoT projects today. With adequate protection from hackers, you can trust that your data is real (versus altered in storage or transit), an important consideration in many IT applications. We have made major strides in protecting the IoT—to the point where data integrity concerns should no longer hold you back. (Read more here.)
12. Finally, make your pilot prove the concept
Too often, pilot IoT implementations are designed to “succeed” instead of to objectively evaluate the opportunities and challenges. We commonly encounter proofs of concept that cannot scale and are, in a sense, IoT silos themselves. If you’ve thoroughly addressed the considerations we’ve laid out—and built your foundation on proven hardware and software—you’ve likely set yourself up for scalability.
Give some thought to these considerations as you plan your IoT initiative. What is my use case? How exactly will value be created? How much data do I need and from what sources? Where will I store it, where will I apply the intelligence, and how am I going to secure it? Will it scale?
Tackling the challenge with best practices (and a good advisor) makes data less daunting, and the IoT payoff bigger.
Lou Lutostanski, is Vice President of IoT, Avnet
Today's case for AI at the edge
AI might feel far away, but engineers are innovating everyday. Should you consider edge AI? ...
Specialists vs. end-to-end partners: the pros and cons for developers
Getting a product to market requires more than just the right idea; it also needs the right partner....
Making sense out of sensors
If creators still don’t think that sensors are getting the attention they deserve now, then there re...