Businesses looking to take advantage of the internet of things are deploying a wide array of digital devices and sensors, which gives them access to an unprecedented amount of raw data. However, for many businesses (especially small to midsize companies), managing all this data and capitalizing on it to the fullest extent can be difficult.
In my last article, I talked extensively about how businesses can overcome this challenge through a next-generation approach that applies meta-learning to machine learning — a process in which a machine is taught to automatically perform the time- and labor-intensive steps that a data scientist has to perform to build highly accurate predictive models. This would enable businesses to turn their data into insight without an army of data scientists by automating the data science lifecycle.
Acting on predictions
However, that’s only one step in the process. Enterprises can use their analytics to derive the most accurate predictions in the world, but if they can’t act on these insights in a timely fashion, then they aren’t getting the most out of them. In that case, all the time spent preparing data, creating and validating models and putting these models into production will be for naught.
It’s not just about acting on the prediction either, but creating flexible business processes to facilitate turning predictions into actionable decisions. Take one common use case for predictive analytics: predicting equipment downtime using IoT data. Simply nailing down the model to predict machine failure is a comprehensive task in itself. But what if a company deploys thousands of machines across the world, with some operating in remote locations in the field? The company in question also needs business processes dedicated to acting on predictions in every scenario to ensure the problem can be addressed in time, without any disruption.
In this instance, the enterprise not only needs a predictive result, it also needs the predictive result to trigger a specific action. This may be as simple as sending a notification to a field service agent. But increasingly, the process is growing more complex — in the near future, it may involve sending the notification to the service agent, along with repair instructions that can be displayed in augmented reality. The business application logic could also be integrated into the inventory management system to account for the parts used for repair and into customer-facing functions to reschedule service around the repair or shift service to another capable machine to keep up with production. As technology continues to improve, enterprise applications will be able to take on a bigger role in facilitating different actions — but only if businesses can develop the application with these capabilities in the first place.
From IoT insight to decisions
In the aforementioned scenario, predicting machine downtime and repairing the equipment is only a small step in a comprehensive process that impacts everyone involved, from field agents to end customers. Assuming businesses have the predictive capabilities handled, they need to tie this capability into other business processes as well. From an application development standpoint, the business would need to consider:
- Multi-channel user experiences: Depending on the location of the machine, service agents may need different tools to make repairs and adjustments. Everything from diagnostics utilities to repair manuals should be tailored to the device used to do the job, regardless of whether field agents are using a tablet, some proprietary mobile tool or even augmented reality.
- Complex business logic: The application needs to be able to receive the notification for the faulty equipment and then decide on the right course of action based on other conditions. Enterprises wouldn’t want to dedicate resources to repairing one machine when a more important one is about to fail as well. Everything from time of day to value of the machine needs to be considered.
- Integration with other relevant business applications: Making a prediction is only one step; the application needs to be tightly integrated into other relevant applications to ensure the prediction can be acted on promptly, without any other disruptions. From customer management systems to inventory management, integration must be seamless.
- Flexibility to support new application workloads: Business conditions are always changing, and the application needs to be adaptable to match that. In the scenario described above, the application pushed notifications forward to the user, who then needed to conduct the service. But perhaps over time, more steps can be handled by the application itself to reduce dependency on human operators. This only happens if the application has the flexibility to support new application workloads from the outset.
- Machine learning and predictive analytics: Once a prediction is made and a process is triggered, the application needs to continuously improve its models based on the various production scenarios encountered and the overall results. Changes in the machine environment, changes in operating characteristics and information about the accuracy of the model all need to be fed back into the model so they can be improved.
Using data from IoT and other sources to derive insights and predictions is a critical first step to improving the way your business operates. However, to truly make the most of those insights, businesses also need applications that can take predictions and turn them into implementable decisions. This requires comprehensive business applications with complex business logic, seamless integration, flexibility and a great user experience.
Building such a comprehensive business application is no easy task, but many enterprises across the globe are starting to take this approach — and they’re achieving some great results. As you look to the future and try to figure out how you can improve your analytics process and application development process, keep in mind this integrated approach. This will enable your business to take the next step in the process and go from simply making accurate predictions to actively implementing decisions.
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.