The old days of the Quality movement seem quaint in retrospect: the banners, the slogans, the prizes and team-building events. Today, quality management is all about information — or, more to the point, how to keep from getting buried in it.
In the age of social media, there’s so much rich data available from consumers that companies ought to be able to convert that feedback into the creation of perfect products. If only. The problem is one of sheer volume. How can merchandisers, especially in quality-focused industries such as pharmaceuticals, make sense of all that information?
The short answer is that they can’t — not without the help of artificial intelligence and the emerging science of predictive modeling.
A drug manufacturer typically receives tens of thousands of comments and complaints from multiple sources the world round. The standard complaint-handling unit would be struggling to triage this overwhelming input, separating positive from negative, trivial from life-threatening.
In fact, the preponderance of complaints is usually low-risk in nature, according to Steve McCarthy, vice president of digital innovation with Sparta Systems, a vendor of quality-management software. Nevertheless, he says, every complaint has to be “touched.” The trick lies in filtering out the “noise” and focusing on the truly important feedback.
With improvements in natural language processing, AI has become an increasingly viable means of interpreting all of that data. The idea is to let automation sort through the raw data, then present the complaint-handling unit with an assessment of severity, possible root causes and even suggestions for action.
It remains — at least for now — the job of people to accept or reject computer-generated categorizations and suggestions. “At this stage,” says McCarthy, “we’re not removing the human from the decision-making process, but we’re trying to enhance the ability to make smarter, more effective decisions.”
In theory, the system should improve its analytical capabilities with experience. That’s the basis of machine learning, a key aspect of modern-day AI. The more it digests data and relays it to quality engineers, the higher the level of accuracy it’s likely to have, McCarthy says.
Beyond satisfying the obvious priority of patient safety, an AI-driven system is also going to be more cost-effective than a manual, human-led evaluation. (Assuming that the latter is even possible, given the flood of information with which manufacturers must deal today.)
But cost isn’t the only reason for seeking a more efficient means of processing complaints. “There’s the factor of speed and timeliness,” says McCarthy. “It’s important to get to a solid understanding of the likely severity and [level of] risk associated with a complaint, so that you can complete the root-cause investigation.”
The next step in the use of AI — and one that’s far from reaching full maturity at this point is predictive modeling. Having proven itself capable of data contextualization, order categorization and automated risk assessment, the system can then begin to anticipate the types of complaints it will receive. Through the use of trend analysis, it can actually foresee the level of severity or risk that a complaint is likely to represent.
Armed with increasing volumes of input, the AI engine can identify the number and nature of deviations from quality benchmarks. In effect, says McCarthy, it’s learning from that data before a product is even released.
“Those data sets can be brought together so that if you see a certain pattern on the shop floor that’s been linked to complaint data post-market, you can correlate those data sets and start to predict that this certain pattern could lead to a complaint,” McCarthy explains.
The feedback becomes part of a closed-loop system by which it’s incorporated back into the making of the product. The manufacturer might discover that a pressure gauge is malfunctioning. Or a flaw could be detected at the design phase, necessitating a change in raw materials and initial production setups.
While the value of an AI-driven quality process seems most evident in the life sciences, McCarthy sees the technology as applying equally to any number of other industries. Medical-device manufacturers, for example, stand to benefit hugely by being able to tweak expensive pieces of capital equipment before they go to market. “At the end of the day,” McCarthy says, “it’s about signal detection.”
The technology still has a ways to go before predictive modeling becomes a trusted and effective exercise. In recent years, AI has made great strides in understanding the subtleties and sentiments behind human responses, but it’s far from perfect. How, for example, does a maker of contact lenses parse such complaints as “scratchy,” “itchy,” “burning,” and simply “an uncomfortable feeling”?
McCarthy expects to see continuing advances in the technology, as life sciences manufacturers strive to meet both consumer tastes and regulatory strictures. The automotive industry also stands to benefit from the use of AI to improve safety and build quality testing into manufacturing.
“It’s being piloted in a number of areas,” McCarthy says. “The ability to apply predictive modeling to that process is very exciting. I can’t even imagine the capabilities we’ll have in a few short years.”