by Chris Curran

Do you have a plan for harnessing IoT data?

Jul 30, 2018
AnalyticsArtificial IntelligenceBusiness Intelligence

As things become more connected through the Internet of Things, artificial intelligence will become more powerful. But, raw IoT data alone isn’t enough to uncover the true value of AI.

binary neural network - artificial intelligence - machine learning
Credit: Thinkstock

If you trust the headlines, 2018 is either the year of blockchain, the year of AI, or the year of augmented reality. The Internet of Things (IoT) doesn’t get the same buzz it did a few years ago, but investment in IoT is still strong. For many enterprises, however, IoT still has the potential to be the most transformative emerging technology.

AI may be the most transformative of all in the end, but it is limited by access to lots of clean, understandable and relevant data. Before, companies were only able to collect periodic, analog information on their physical infrastructure, products and supply chain, but IoT sensors allow companies to generate hundreds of readings per day.

A single data point is not valuable, but a real-time stream of data can reveal trends. It’s what I call the Internet of Business Things.

As things become more connected through IoT, artificial intelligence will become more powerful. Don’t be fooled though, raw IoT data alone isn’t enough to activate AI. You will also need to determine which combinations of sensors and machine information together represent complex, real-world situations.

Humans in the loop

The vast amounts of data coming from IoT sensors could be used to train machine learning data, but it is largely unlabeled. That is, you may know where each sensor is and what it is measuring, but you don’t know specifically how that corresponds to an event in your business processes and outcomes, unless those two very different silos of data are integrated. And in nearly every case, you need humans to help label that data.

For example, if a restaurant supply company wants to use machine learning to predict when food would spoil, sensors could measure the temperature in the refrigerator, the amount of times the door opens and the weight on the shelves, but humans would have to observe and record when the food spoils to create the training data to better predict and prevent premature spoilage.

Similarly, sensors in a hotel room could record the ambient light, noise, temperature and humidity, but the database that records guest complaints is likely kept in an entirely different data silo. And again, it’s updated by humans.

Very few companies have begun to integrate these data silos. Even fewer have trained their human functional specialists how to label data to be usable by machine learning algorithms.

Investment in IoT and machine learning often go hand-in-hand, but before deploying these technologies, companies need to plan how they’re going to work together. Who is going to label IoT data? How will the enterprise use it and put it to work?

Just the first step

To take full advantage of IoT and AI, collecting and labeling data and feeding it into an algorithm is just the first step. Most companies will need different teams to define the business questions they want to answer, identify what IoT data is (and could be) available and label data based on the value it provides to each of their business functions.

As PwC reported in our AI Predictions paper, with IoT and AI spreading throughout the enterprise, companies will need more functional specialists – who have a basic understanding of data science – than advanced engineers or programmers.

Once labeled and analyzed, companies can prepare to act on what they’re learning from the terabytes of IoT data pouring in and the patterns that machine learning reveals.

First, the demands on digital infrastructure across the enterprise will increase. Machine learning in the cloud may be the only architecture solution to process multiple live streams of data across a large organization. And business leaders will need more advanced business intelligence (BI) tools to visualize the insights produced by machine learning and IoT data.

Second, the organization needs to become more nimble to make decisions based on these insights. Traditional decision-making takes weeks or months. By then, sensor data might be obsolete, or the business opportunity may have expired. As I advise clients, “Collecting more distributed sensor data enables making more distributed decisions.”

The CIO clearly has a role in planning an IoT deployment and infrastructure transformation, but for acting on IoT insights, the COO and CEO need to be involved. In a more consumer-facing organization, so does the CMO.

Together, the leadership needs to put a mobilization plan in place to adapt the business to what it’s learning from AI and the company’s countless touchpoints with the real world.

Don’t go with your gut

IoT is just about information and transparency into everything the business touches. AI is about learning. It’s the business’s responsibility to act on those learnings and adapt in real-time. If deployed successfully, machine learning and IoT sensors enable organizations to fine tune their operations, taking advantage of feedback loops: learning, improving, learning even faster and improving more.

By harnessing these feedback loops, an organization can improve the effectiveness of a workplace, a truck route, a warehouse, an assembly line or a retail space. Before we had this data, we were just using “our gut” to design work processes. In a world of digital transparency, we can continually optimize and redesign the way every business works.