By Tilak Kasturi & Doyle Irvin
The fourth industrial revolution is motivating companies to rethink how they do business — largely because tech innovations are disrupting every known industry on the planet, and those that are failing to embrace this change are likewise failing to keep up. The overall shift from ‘Big Data’ to ‘Applied Data’ opens up a number of pain points (and opportunities), so we put together the top three priorities for any organization looking to excel over the next few years.
This shift from Big Data to Applied Data revolves around one key question: how do I make sense of the data and solve specific business problems with it? This has led to Artificial-Intelligence-driven solutions slowly making their way into the enterprise environment. These early AI success stories have had uneven ROIs, however — not all AI solutions are created equal. At Predii, we focus in Applied AI for Industrial Repair and Maintenance. We see three key drivers emerging influencing AI adoption that apply to the enterprise market at large. These key takeaways are: focus on usability, leverage your domain expertise when powering learning algorithms, and employ blackbox solutions whenever possible.
1. Applied Data → 'Big Data' Focus on usability
Enterprises that are focused early on usability are able to target their "Big Data" efforts to higher ROI. In the early days of the Big Data rush, enterprises over-invested in collecting data and storing data, without a super-clear understanding of how they would use it. As a natural next step, they have been asking what the ROI is for these Big Data projects, because they soon realized that it is extremely expensive to collect petabytes and zetabytes of data. Those enterprises that did focus on use-cases and usability from the beginning are reaping much higher ROIs from their Big Data Lakes.
This is not to say that you should lose any focus on collecting data — that’s still intrinsically important. What this addresses is the fact that using the data and generating truly game-breaking insights from it is a much more difficult problem than simply collecting data, and thus requires that it becomes one of your team’s main priorities. The upside of this is that the rewards for solving usability can be massive.
In Repair and Maintenance use-cases, one of the challenges is to get insights from data that arrive in vastly different formats: unstructured data from field services, customer complaints and historical service records; structured time series data from sensors and diagnostic codes; and OEM service manuals.
The first usability challenge was extracting value from historical data, which can be filled with lots of noise. The second usability challenge was creating actionable insights from that data for a wide range of operators. Engineers and insurance companies want to know which parts are likely to fail for a given make or model, by age; the former to improve their product and the latter to set competitive rates for protecting it. Field service agents want to know precisely what the true problem is for symptoms that can be expressed by customers in a manner of different ways, so they can send a field tech with the right parts to get the equipment fixed without needing return visits. The technicians themselves are looking to be augmented with intelligent repair guidance in an era where component complexity is exponentially increasing — do they need to repair a component, or just clean it?
When retold like that, answering those questions sounds easy. But the truth is that at-scale intelligence extraction involves making sense of the data, correlating it within many different parameters, and deriving inferences that were previously unavailable — all within a problem-specific framework that is heuristics-driven. These steps are interdependent. It’s not enough to just have data. It’s not enough to just have an algorithm. It’s not enough to just have experts to guide the problem-solving process. You need all three.
2. Learning algorithms → Domain (human) expertise is your secret sauce!
Successful AI platforms need "domain expertise" to power learning algorithms and human oversight to to prescribe meaningful solutions. AI Solutions are useful because they are powered and configured by appropriate domain expertise, and, to add to that, the majority of use-cases for AI solutions have to do with empowering and augmenting humans — not replacing them. Putting data to use doesn’t mean putting people out of work.
What is going to happen is that the definition of what it means to be an “expert” will shift. This shift will be from years on the job to how effectively one can apply intelligent tools to complete a job, and how quickly they can learn a new set of tools as technology advances. Increased complexity of work environments and their products is inevitable. What that requires is increasingly intelligent tools, powered by democratized human expertise, in order to maintain the baseline where interdependent specialized teams can continue to cooperate.
What this means is that enterprises will be increasingly responsible for an “in-house training” of sorts. This applies to both training their workers to use new tools and training new tools with their existing workers. Because domain-expertise-powered algorithms provide much higher ROI to the enterprise, you can expect to see a sharp increase in roles involved with turning your heuristics into data-driven insights. That being said, this process isn't easy — or cheap — which leads us to point #3.
3. Solution → Embrace black-box AI solutions wherever possible
Black-box AI is the natural evolution of enterprise AI solutions. Embracing AI has become strategic for enterprises at the board level. Building a data-science team and harvesting gold from the "black hole" of Big Data Lakes is quite expensive for enterprises, especially when compounded by the inevitable focus on time-to-market and ROI. More and more we see large enterprises looking for black-box vendors to solve specific use cases, and for good reason: though black-box AI solutions are relatively rare, they are far more “plug and play” than their alternatives because they are purpose-built.
They are vendors that solve specific business problems, and solve them with a specialized expertise that allows them to operate with far more efficiency that their generalized-AI counterparts — their focus allows them to stay incredibly lean.
One of the main advantages of a black-box solution is that it minimizes risk. Building your own data science team takes a lot more effort than using one that’s pre-assembled, and it also comes with greater outcome variability. With a black-box solution, you get a proven track record.
This is why, whenever possible, embracing black-box solutions is a strong business move. One of the first tasks in any AI strategy is to do solid research on just what is available; there’s a lot out there. The first step in this process is understanding which of your business problems can benefit from AI.
Given the difficulty involved in Applied AI, and the reward for solving it, you can expect an exponential increase in data-related roles over the next few years as organizations attempt to ensure that they are leveraging their data as much as possible. In order to gain that leverage, competitive enterprises will make sure they have a team whose responsibility it is to bring all three of these ROI drivers together, and not just once — you need to constantly assess and re-assess whether you are taking advantage of your data. Given the pace of innovation in the AI sphere, it's a healthy practice to step back to see the big picture at least every six months.