How to turn IoT data science into real business value

You’ve succeeded in connecting your devices, which are now dutifully publishing IoT data to the cloud. Now you’re in a position to remotely monitor and control your equipment in a production environment. Congratulations, that’s no small accomplishment! However, you have bigger plans for your IoT investment. You know there are valuable insights to be gained by applying data science to your burgeoning store of bits and bytes that have the potential to significantly affect your bottom line.

You’re on to something.

Having worked in industrial IoT for over a decade, It’s become increasingly clear to us at Bright Wolf that much of the value of IoT technology lies beyond the basic connect and monitor scenario, with the ability to analyze IoT data to improve products and operations as well as discover new business opportunities.

There’s gold in them thar’ IoT data hills. So how do we get it?

At Bright Wolf, we’ve seen first hand the astonishing percentage of IoT initiatives that end up either discontinued or in pilot purgatory. It’s not because the technology isn’t really cool (it is), but rather because the initiatives fail to demonstrate any real or potential return on investment. For this reason, we have a pretty unwavering commitment to a methodology we call Zero Waste Engineering™. The basic idea is to employ an iterative discovery process that doesn’t break the bank while proving out the value of your initiative. While this seems like common sense, many industrial organizations are trapped in legacy patterns of development resulting in IoT project failures like the $3 millon spreadsheet and other common missteps on the path toward digital transformation.

With Zero Waste Engineering™ as a backdrop, and an eye toward better business outcomes, let’s walk through a proven approach for applying data science to industrial IoT systems.

Define the goal

A common IoT data science application is predictive maintenance, an effort aimed at eliminating expensive, disruptive, or catastrophic failures. Other common use cases are discovering opportunities to lower operational costs, product and service improvements, and better customer support.

Build your coalition

Your coalition will certainly include management at some level; the project sponsor or champion is responsible for defining the business goals. Of equal importance are subject matter experts who have an in-depth understanding of the equipment or process you’re focusing on. They could be plant managers, operators, or maintenance personnel. Your sales and customer support organizations are likely to provide valuable insights here as well. Product owners and engineers should also have a seat at the table.

Finally, be sure to include information technology (IT) leaders, and involve them early. They will be instrumental in accessing the data needed for your research, and are likely to be involved in anything you ultimately deploy resulting from your work.

Generate your hypotheses

Identify data requirements

You’ll use other data sources to contextualize the IoT data. Enterprise data from your ERP, CRM, and other systems may be relevant. Similar for external data sources such as weather, population, and energy or fuel pricing. For example, a predictive maintenance use case might require equipment age and model — information that resides in a separate sales system.

Collect your data

Choose your analysis tools

Do your investigations

Note, it’s crucial to have a skilled data scientist with a thorough understanding of the math and methods doing the investigations. It’s very easy to use the tools to hack together something that allows you to jump to the wrong conclusions at lightning speed with numbers that seemingly back it up. It is a much more difficult task to produce an accurate, meaningful, quality model that informs the matter at hand, but that’s what you’re after.

Verify results

As with all experiments, you need to quantify assumptions and expectations. Start by defining and baselining key performance indicators (KPIs). Apply your newly found insights to the problem or process at hand. You may have a friendly customer who is willing to participate. Monitor KPIs. Is this experiment delivering the expected results? Adjust and repeat as necessary until you’ve either proved and refined, or disproved your theory. Remember, this is an iterative process.

It’s important to incorporate coalition members into the feedback loop. Share successes and failures. Even seemingly innocuous results may mean something to stakeholders and may inform subsequent activity.


This is a project in it’s own right. It’s worth a second mention: Involve IT early. Whatever you ultimately deploy will likely be integrated with your existing systems and infrastructure.

Azure Machine Learning service provides tools that make it fairly simple to create, manage and deploy ML models. AWS has a similar service called SageMaker which provides the same capabilities. This will be the subject of an upcoming article where we’ll dive deeper into ML tools.

To learn more, contact us today and we’ll be happy to share a few best practices and provide an initial evaluation for how we can help you achieve your goals.

Originally published at on July 30, 2020.

Data-centric technology provider and system integration partner for industrial IoT and connected product solutions.