There’s been a lot of chatter about artificial intelligence (AI) over the past several months. Specifically, generative AI; and even more specifically, ChatGPT. In fact, if you’re even a little bit curious, there’s a reasonable chance you grabbed an OpenAI account and have been taking ChatGPT for a test drive. And if you haven’t, I’d encourage you to kick the tires.
AI has captured our imaginations for decades. From HAL 9000 in “2001 a Space Odyssey” to “Terminator” and “The Matrix” franchises, AI is as scary as it is fascinating. And while there are things to be concerned about (we’re not going to discuss those here), one thing is certain: a carefully thought-out technology strategy that includes the various flavors of AI can have a monumental impact on virtually every industry, and the medical device industry is no exception. In fact, we’ve seen it in use on the clinical side for a handful of years now. But operations stand just as much to gain from a well-defined plan and thoughtfully designed tools.
But first things first: what is AI, really?
This is an important point, even though it may seem obvious. To over-simplify, AI is a general term for software that can mimic human cognition to perform complex tasks and then learn from those outputs.
Great; but what does that mean?
It means that given the right set of information (data), and the right set of instructions, the software (AI) can provide analysis, answers, predictions, etc. that would otherwise require significant human intervention. And the more it’s used, the more feedback it gets about the accuracy or validity of its outputs, the ‘smarter’ it [can] get, and the more accurate its future outputs [are likely] to be. (Yes, there are qualifiers in there. And yes, that is on purpose.)
Why am I telling you all of this?
It’s important to have some background before diving into whether/how AI in any form can help your business operations.
There are two critical aspects to understanding AI in this context:
- It absolutely requires data; lots and lots of data. And that data must be clean, meaningful, and understandable. If it’s not; well, garbage in, garbage out.
- It’s basically dumb (no offense, AI). Which is to say, while AI can parse and decipher and collate and assimilate massive amounts of information in nanoseconds, it still needs someone—or many someones—to tell it how to interpret that data, what rules to follow when interpreting the data, and what actions to take, if any, after interpreting the data. It also needs to be told when it’s wrong, so it doesn’t get it wrong again or worse, build an entire dataset of answers on top of a false premise.
Make no mistake: AI is not consciousness; it’s an incredibly powerful set of tools designed to augment and accelerate the capabilities of its human authors. Over-simplified? Sure. But not any less true.
Now that we have all that out of the way, let’s talk about why AI will play an increasingly important role in medical device operations and what the operators will need to do to enable and take advantage of it.
Many businesses are rife with highly manual processes that take time, crush efficiencies, and suck the soul right out of their teams. Most of those manual processes are legacy leftovers from a time when things simply could not be automated. In general, that is no longer true. Sure, it still feels true in many cases, but that’s usually because no one has the time to figure out how to stop doing things manually.
Do we need to write things down on paper? No. Do we need to type the same thing into a CRM every time we get a repeat sale that’s similar but not exactly the same? Nope. Do we need to count every item three times before recording it on a piece of paper that gets passed to an admin who puts it into a spreadsheet that then gets uploaded to a piece of software sitting on a server under someone’s desk? Hell no.
Automation, especially automation that is predicated on past behaviors or behaviors that meet specific criteria, is a sweet spot for AI and its relatives, machine learning, and deep learning. Does automation require AI? No; of course not. But when you throw artificial intelligence into the mix, you stand to magnify the value of that automation. Why? Because AI can help decide the path of automation itself, providing much more dynamic, context-specific responses.
This can come in the form of chatbots that answer basic customer support questions, freeing up your team to focus on more complex and value-additive customer needs. Or it can come in the form of much more complex toolsets that can auto-generate things like inventory replenishments, removing multiple human steps, and only requiring human review and verification before physically packing the box. Or in the form of a system that can determine the average time to pay for a given customer, the number of times they need to be reminded to send you a PO, and then automate all those communications for you.
Analyze and Predict
This is where things start to get sexy. I mean, if you’re a data geek, that is. With enough historical data, AI can begin to analyze the past and predict what future behaviors might look like. Think weather forecasts, future traffic conditions, and Amazon shopping cart suggestions.
What that can mean for operations, especially medical device operations, is nothing short of revolutionary.
Imagine a world where digital tools can predict caseloads, and pre-emptively get those cases on the calendar in advance, alerting you of potential resource needs. A world where software can see the cases that are scheduled, predict what assets are likely to be required, where those assets currently sit, which of those assets are slow-turning, and then recommend movements based on proximity, cost, and ROA. A world where performance is constantly monitored for positive and negative patterns—patterns that may not be obvious when you’re in the weeds—and then flagged and tagged with one or more possible causes and outcomes.
Sound crazy? It’s not. In fact, in many cases that world is already here, and it is being realized in many industries. And that makes us wonder why it feels so foreign to medical device manufacturing. (There is an entire article in that last sentence, but I’m not going to boil the ocean today.)
Forecast and Plan
This is a keystone of manufacturing operations. It can be highly manual and, I would argue, a bit of an art form. Sure, you are using historical data, but you’re also very likely peppering in context, nuance, and maybe even some raw emotion when trying to determine what the business is going to look like in a quarter, a year, or five years. And then you must use that information to build a plan you can execute. It takes a ton of information and a non-trivial amount of time.
While some of this isn’t solved by machines, the simple fact that forecasting requires massive amounts of data and massive amounts of time is enough to imply a role for AI.
For example, if all your sales and case data is accessible, AI can use that to build multiple models using various criteria that you or your team set. And then, if all your asset data is available, it can use that to tell you where you’re short, what needs to be built or ordered, and—if your lead time data is accessible—it can even tell you when to order. If your business is seasonal, and the data about that seasonality is available, AI can be used to help plan staff augmentation, stock increases, and distribution.
Perhaps the biggest value of AI when it comes to forecasting and planning is the ability to leverage it to do the lion’s share of dirty work, i.e., ingesting and analyzing massive amounts of data and turning that into relatable chunks of information that your human team can review and asses as you build your plans for maximizing value and mitigating risks.
Ok, so now what?
Some of this sounds like magic beans and snake oil. But rest assured, we’re not navel-gazing into some future state years away. The tools are here now, and they are being implemented to varying degrees by companies at all stages of development. In fact, smaller manufacturers and startups have a distinct advantage here by virtue of their size, complexity, and limited access to resources. Because they are not tethered to the same large bureaucratic structures, they can experiment, iterate, and pivot much more quickly than their massive counterparts. And because cloud services have made the barrier to entry essentially zero, many of these experiments can be done, to a significant degree, with the human resources they already have with little or no upfront costs apart from people-time.
Does that mean you should jump in head-first? Well, no, it doesn’t. But you can start by considering these things:
- What does my business do that could benefit from AI?
- Do I have the right team to brainstorm appropriate use cases for AI in our operations?
- Are there risks with these use cases that we need to define and consider?
- Do I have enough data to make AI useful and the outputs meaningful, accurate, and relevant for the use cases I have defined?
- Does my team have the time/bandwidth to define and execute a proof of concept to test our use case(s)?
- If the proof of concept is successful or compelling, can we scale it with our team? If not, do we have the resources to bring in outside help?
If you get positive answers for enough items on that list to experiment, do it! How many is “enough?” That’s for you to decide. But the power and the potential are clear and will only expand over time. And those companies that sit on the sidelines will, eventually, lose.
To de-risk the exploration and speed up your time-to-value, here are a few things to try:
- Investigate existing tools that can integrate with your current processes that are implementing AI as part of their offering. This will let you dip your toes in the water without the risk of drowning
- Leverage open-source or free commercial tools in your individual daily practice to get a better understanding of what AI is and isn’t and what it can and can’t do (go get that ChatGPT account, for example)
- Search YouTube for videos on AI and then watch them
- Talk to your team about potential use cases, possible risks, and ways to mitigate those risks over time
- Review any current or planned system implementations and insist that, even if AI is not built in, it can be integrated/leveraged in the future
- Start small, run fast, and break as little as possible