Today’s cutting-edge research on cloud solutions for manufacturers highlights how artificial intelligence and machine learning have the potential to prevent downtime, improve safety, and reduce material waste. That’s exciting for industry leaders, who are always looking for ways to refine these core efforts.
At AWS’s re:Invent conference in November, AWS CEO Andy Jassy dedicated a considerable portion of his keynote to machine learning, which hasn’t happened before. His message was clear: businesses need to start developing an understanding of machine learning now, because this technology will be critical in the future.
Even though machine learning is on the horizon, current ML and AI applications aren’t the simplest or even the best-suited solution to the problems manufacturers face today. Given the complexity of these technologies, most manufacturers would be wise to consider these three questions before turning to ML or AI solutions.
1: What Problem Are You Trying to Solve?
AI and ML are undeniably powerful and exciting technologies. For manufacturers, who produce an enormous amount of data simply by running their operations, The temptation to bring in these technologies to see what they’re capable of is understandable.
In reality, this strategy doesn’t work.
AI and ML are tools. In some cases, they’re the right tool to solve a problem. Unfortunately, they don’t have the power to transform a business in and of themselves. For example: let’s focus on the shipping or logistics arm of a manufacturing company. Currently, it uses human inspectors to verify that boxed orders match what was submitted digitally. Turnover, though, has been high. The company is spending too much time and money recruiting and training employees for this group.
An IT leader at the firm suggests replacing most of the human workers with an algorithm that could visually inspect shipments and determine whether they match orders. In this case, it’s possible that AI or ML might be a workable solution: the right algorithm could learn to recognize items in different orientations and various lights. It could save money and speed up the QA part of the shipping process.
In this case, the IT leader identified the problem first and then offered an AI- and ML-based solution, rather than first deciding to implement AI and ML and then looking for a problem to solve with these tools. Even if the order of operations is right (as they are here), it’s important to ask yourself whether AI and ML are the best solution for the problem at hand.
2: Can Your Existing Tools Solve This Problem?
Before considering whether to explore AI or ML solutions, it’s important to first determine whether the tools you already have at your disposal are sufficient. While new technology may be the most exciting way to solve a problem, it usually isn’t the most practical. Only certain problems truly lend themselves to AI and ML solutions. Besides which, the newest technology is rarely the most time- or cost-effective.
For example, large and legacy companies often find that their technology evolves unevenly. While public-facing departments (like marketing and sales) may rely on cutting-edge software, internal systems tend to be much older. The problem that looks like it needs an AI or ML solution may in fact require the modernization of older internal systems so that the company can operate more efficiently.
Let’s return to our previous example. We’ve determined that the smart QA algorithm may indeed eliminate the problem of worker turnover in the shipping department by fully automating the process.
When we peek under the hood at the software HR uses to manage employee records, we notice that the department never fully operationalized its employee retention software, which has a handy workforce insights feature. If the HR team starts using all relevant features of the software, they might be able to find underlying causes of the shipping QA team’s turnover and address them via new internal processes.
In this case, getting the HR software fully functional would be significantly faster and less expensive than developing a smart QA algorithm tuned specifically to your business. To determine what makes the most sense in your situation, you’ll also have to consider one more question.
3: Do You Have the Resources to Develop and Implement an ML or AI Solution?
Assuming you’ve identified a problem ML or AI can solve, and you’ve determined that these tools are the best way to solve that problem, it’s time to determine whether you have the infrastructure to implement AI or ML. In other words, do you have the AI / ML software engineers who can build the solution you need or modify an existing solution to suit your purposes?
These specialists are hard to find right now because of high demand. Due to this demand, they are also extremely expensive.
If you do have these team members on staff, how long will it take them to develop, test, and implement this solution? How much will that cost? What will they not be able to work on during that time? Will you have to hire additional staff to take over that work? Given all these costs, how long, once the solution is up and running, until it has a positive ROI?
Often, even if an AI or ML solution has the potential to deliver cost savings, the reality of building the solution cuts into that savings – in some cases, substantially. The operational planning and financial modeling necessary to pursue these projects is a critical success factor. Don’t skip it.
Manufacturing leaders today face mounting pressure to demonstrate their use of cutting-edge technology as a sort of shorthand for their ability to innovate. In reality, though, adopting AI or ML solutions without a clearly defined path to ROI can cause more problems than it solves.
Today’s technology means that manufacturers have at their disposal more tools than ever to solve the challenges they face. But the newest tools aren’t always the most effective. Before bringing them to your company, be sure they’re the ones best suited for the job.
Eric Dynowski is the Chief Technology Officer at ServerCentral Turing Group.
Credit: Google News