Advances in artificial intelligence are in the news a lot these days, but what those capabilities mean for businesses is often far from clear. Chief Executive recently spoke with George Casey, who leads RSM’s advanced analytics practice, about how mid-market companies can participate in this AI-fueled future. Excerpts of that conversation, edited for clarity and length, follow.
Chief Executive: When you mention artificial intelligence, people think of robots in sci-fi movies. Tell us what you mean when you talk about incorporating AI in business solutions?
George Casey (RSM): In a lot of cases, it is more about what it is not. There is a lot of misinformation and misunderstanding about what AI is, so the way I describe it is we are talking both advanced algorithms and computer systems that perform tasks that typically require human intelligence—things we previously would have expected a human to do. That includes visual recognition or natural language processing, understanding speech or reading text and interpreting context. The big difference is that it is not just prediction. I’m predicting what you are asking of me, but I’ll also act upon that and make that decision. That is what differentiates AI from what we would call machine learning or predictive analytics.
Can you share some of the more successful applications of AI that you’ve seen?
When we think about the context, we are really thinking about automation of process. Is there a longer-stream process I can automate for enhanced production or improved efficiency or reduced human error? You might, for example, create a customer engagement opportunity that is always on and can really interact with a human. As opposed to a “click here” or “read this FAQ,” you would describe how you would like it to talk to customers and the AI will interpret how to answer them.
We also see this in manufacturing. We talk about IoT being the Internet of Things and the connected workforce or connected factory—but through the sensors or measurement readings we have on equipment now, we can understand outcomes like failure, decreased efficiency, overheating or fire, and respond by scheduling maintenance. That is where we are taking data that we know, like what a machine looked like before it failed, and then recognize indicators in the future and assign a corrective action—finding the big opportunity to maximize productivity on the shop floor.
There’s a perception that these technologies are out of reach for middle market companies. You need to be big to benefit—to justify the expense of buying and implementing technology. How has that changed?
In the past, it really was the domain of data scientists and really advanced computer power and massive data sets. With the advent of cloud computing, we can very easily scale up to increase the computing power we have as opposed to investing in the on-premise hardware to achieve the maximum capacity we need. We now have massive data sets in the cloud and cloud-based AI services. We look at machine learning as a way to empower that citizen data scientist and put some of these capabilities in their hands with enough guidance. So, it’s, “This is what I really need from you, the designer, because I, the computer, can handle the rest.”
You provide the data, and I will recommend what type of problem we are solving, whether we are trying to classify something as fraudulent or not fraudulent, or predict a time series, what demand will be next month, or interpret natural language or what customer sentiment is on Twitter. Those cloud-based tools are making this much more accessible.
We are also seeing the idea of embedded AI, where AI technology can be embedded into your business application software. The simplest example is when you start typing in Microsoft Outlook and it prompts you about what you might want to say next. It is saying, “when I have seen these words before, this is what typically follows so I will suggest that and, if you like it, you can confirm.” There are a lot of open-source AI platforms available that let companies tap into these massive libraries and models. They are the last mile of the process as opposed to the entire journey. These are the things that make this a lot more accessible to the middle market than building it all from scratch.
How should companies approach this? Is it about looking at business problems and seeing if there’s a new way of doing it?
The more things change the more they stay the same, in that with any new technology we should start the conversation with what is the problem to be solved. It might be “I have a quality control problem, I’m shipping too many bad products.” However, instead of a human workforce to check every product, can I introduce an AI-based computer vision system where every part passes under a camera that instantly detects defects? So, the problem is quality control affecting costs and customer satisfaction, and we would look at next-gen solutions versus just using human checkers.
It all starts with business value. We think about value in a few dimensions. There is value created through revenue or income—can I grow revenue or income? If this will help, that’s a value story. Second, can I reduce cost? Is there an efficiency where something that used to cost $100k can be done for $10k because I’m more efficient?
Third, can I reduce risk, risk being unrealized cost?If the things I am worried about come true, there will be a future cost. For example, identifying fraud before it happens, or errors as they happen versus waiting for the audit. Those are some of the ways to think about finding value. Then the question is how might AI or machine learning help me address this valuable opportunity?
What are some specific applications middle market companies should explore as they look for opportunities to incorporate AI into their operations?
Customer churn, fraud detection and quality control are a few. Another that we haven’t talked about is bringing an AI flavor to forecasting. We have a partner named Prevedere, and their key solution is looking at whether external indicators can be useful in predicting outcomes. So, if I am trying to predict my sales as a computer product company next year, I might want to look at what outcomes correlated to economic conditions arose in the past. I might say, how does inflation affect us? When we have seen inflationary pressures in the past, this is how our demand line responded so I can make some assumptions as to how it might respond in the future. Using those massive data sets to look at what else in the world is signaling what might happen in my company is another interesting opportunity.
Tools like ChatGPT and Midjourney are getting a lot of media attention. What is the potential business application of those technologies for middle market companies?
First, this is an extremely exciting technology that is opening people’s eyes to what is possible. With that creates uncertainty, and that is where we get fear. The first is to ask, is this replacing humans? It really isn’t. The opportunity is how can I empower them, how can I assist them in their task, so their productivity goes up and better supports their needs, whether by lead generation, being able to assess content, help organize presentation or help research different ideas. It still starts with someone asking the question and creating the prompt. So, skills will be shifting toward, “How can you write a really good prompt for ChatGPT?” whereas 20 years ago it was, “How can I write a good Google query?”
The risk with all AI is that it is still based on previous data. It is not coming up with a new answer; it is saying what worked in the past. So, we have potential for bias in the data set the model was trained on. Therefore, it can quickly skew to making inappropriate decisions because it has the wrong data set. The final concern is the confidence or certainty with which I have the right answer. When we simplify, we are also taking out some of the awareness of how good this answer is. Let’s say I am going to use AI to look at a radiograph to determine whether a tumor is cancerous. When I do that in the purest machine learning sense, I get statistics—it might say there is 99% likelihood versus a 51% likelihood of the tumor being cancerous. Whereas with both of those results, if you boil that down and just ask ChatGPT whether the tumor is cancerous, it is going to say yes, which is a very different answer.
What are the biggest challenges of leveraging AI capabilities? What will companies need to have in place or get in place to be prepared for a successful implementation?
I still think data science skills are relevant and meaningful in identifying the patterns and trends that will be useful in these models to then append the data set. For example, “here is how we might add more data to make this a better answer.” The other challenge will be quality. At the end of the day, models attempt to simulate real life, but it is still a simulation. So, the idea that it is 100% accurate is not realistic. The other risk is explainability; why did the model say this? If we go back to the earliest techniques around predictive analytics, it was a linear projection: When you estimate house value, you look at square footage, the number of bedrooms and bathrooms, and then it gives a weighting of how these things affect home value. Deep learning is more difficult to explain. You won’t necessarily be able to explain how it got the answer it got.
What else should CEOs be aware of with regard to AI going forward?
Start by thinking about opportunities to experiment, encourage people to be curious, evaluate what could I potentially be doing with this. Create an innovation committee in the organization to start evaluating opportunities. The key is to approach these opportunities with equal parts of optimism and risk avoidance. We see people immediately jump to risk concerns, and what they are not considering is the risk of not innovating and not getting better at these problems. The biggest risk most companies face is lack of innovation. There have been 100 years of proof of this. You can control risk all the way to the point where you are out of business because ultimately companies that don’t innovate go out of business. So, recognize the risk, but also explore, be curious and find these new opportunities.