Taking Control of Artificial Intelligence

Do’s and Don’ts

While only executives can determine an AI strategy, experts who have studied the issue have some recommendations on where to start. The first thing that has to be clear is the CU’s digital strategy and data strategies, says Kirk Kordeleski, CCE, former chief strategy officer for Best Innovation Group, now executive benefit consultant at OM Financial Group, a CUES Supplier member based in Boston. 

“Everything starts with the data,” explains Kordeleski, who also previously served as the CEO of a large credit union. “AI and machine learning are tools for getting what you want from the data. You need the data, the skills to manage it, project road maps and a strategy.”

Starting with data is the right approach, agrees Zach Hill, CTO of CUES strategic partner Think|Stack, Baltimore, which advises and provides service to CUs on technology issues. 

“Have a plan to consolidate and package your data to make it more actionable,” he advises. “Most credit unions are in the early stages of doing that.”

Executives also need to make an honest assessment of whether their CU’s culture is a good match for AI, Kordeleski points out. 

“You need the right people and the right policies,” he says. “Not every credit union is ready. The value has to be there because it’s an investment that won’t necessarily pay off unless you’re ready for a pretty drastic change. It has to be a financial decision but strongly supported by the culture. Can you get the rewards in lending, marketing and operations from these tools to justify the investment?”

The big task for the C-suite, Kordeleski says, as they think through and formulate an AI strategy, is resource allocation. If you’re frugal, you could fall further behind because artificial intelligence is evolving quickly, he warns. If you’re enthusiastic, you could overspend. 

“There are many ways to go,” he continues. “Some credit unions may not need a comprehensive AI strategy. But what you plan needs to align with what you will spend, or what you can spend needs to determine what you plan.”

Most CUs have badly underestimated the time and money it will take to embrace AI, Hill observes. “You can’t flip a switch and see results. You have to nurse it like a plant. There are a few easy wins—like using Snowflake (a cloud-based analytics tool) on sales and membership data—but for a strategic move to AI, most are woefully falling short.”

Don’t be too slow to consider AI strategy, cautions Vladimir Kovacevic, co-founder and managing partner at Inovatec, Burnaby, British Columbia. AI isn’t coming; it’s here, he says, permeating business processes. Because of this, CU leaders need an AI-first mindset and should make specific business decisions around what will connect to their AI premise. The most immediate opportunities are in risk management and work automation, he notes, with enhanced member experience close behind.

Three Levels

C-level executives need to understand three kinds of automation technology, according to Michael Carter, EVP at CUESolutions provider Strategic Resource Management, Memphis, a business that interfaces between CUs and the providers of AI technology.

The first is robotic automation, which works well for simple “if this, then that” situations. Each metaphorical “robot” can be coded for different functions to work together and make a process happen automatically, repetitiously, until a human programmer changes the code, he says. 

The array of possible applications of robotic automation are still being developed, but the C-suite folks at most CUs probably are familiar with projects to tackle manual, repetitive back-office work routines and look for ways to replace resource-consuming tasks with robots that improve efficiency and quality, Carter says.

Carter thinks C-suites should start small with robotic automation. “Don’t take a plunge,” he says. “Start with a toe in the water, a process that can be automated and then improved.” 

He suggests auditing invoices as a place where robotic automation can be effective. “When credit unions audit invoices, it’s after the fact, and some invoices are complex, taking as much as two hours for a human to process. They often skip the deep dive and compare the invoice with the prior month or information on a purchase order. If the numbers align, usually they pay it, without comparing the invoice to the contract, which is a dynamic document that probably has triggers that impact the fees paid, like types of services, numbers of users or levels of transactions.”

In contrast, AI robot has instant access to the various bits of information that might determine the proper price the CU should pay and can almost instantly produce the right information. It does more than the human would do. For a human, a deep dive is complicated and time-consuming. For a properly programmed robot, it’s quick and easy.

The second type of automation technology driven by artificial intelligence is machine learning. Here the tech is programmed to receive and respond to input, like find a second path to the goal if the first path is blocked, Carter says. This tech will also remember that path if the situation recurs. It doesn’t replicate the job of a teller cashing a check or a call center agent answering a routine question like robotic automation could do, but rather the job of the person who sets priorities and designs and adjusts the workflow.

Deep learning is the third and most complex and powerful application of AI. The machine can observe and identify correlations between data sets and within processes, Carter explains. Deep learning applications are not given a specific set of criteria like machine learning applications, he says. They are intelligent enough to make deductions and decisions without humans as guides.

Tool Providers

The more complex and advanced AI gets, the more CUs are relying on outside experts to help them make it work and let staff focus on the business and member relations side of the operation. Only the very largest are trying to put AI experts on staff, Carter reports. 

AI is “deeply mathematical,” Kordeleski notes, so it’s nothing to build in-house. “You need involvement by data scientists and business analysts with specialized skills”—which raises the critical question of where those skills will come from. CU execs can’t rely on mainstream vendors to supply them, he warns. 

“Ten years ago, you could count on your vendors to provide a road map and the tools to stay pretty current with the technology,” he reports. “Now it’s a humongous challenge. The traditional vendors are not up to it.” Even so, you should consult with key vendors—such as your core provider—because they are talking to data scientists and business analysts, he adds. 

Leaders should also have vendor review initiatives to align their CUs with providers that can most effectively access their data and process it in an AI-friendly way, Hill recommends, and ones that can work well with other vendors to achieve interoperability.

And new vendors are popping up that need to be on the     C-suite’s radar. Kordeleski cites CUES Supplier member CU Rise Analytics, Vienna, Virginia, for “amazing, credit union-specific analytics” and Quinte Financial Technologies, New York, for fraud analysis. Sherpa Technologies, Columbus, Ohio, offers chances for credit unions to participate in “fintech accelerators.” Also helpful in the right situations, according to Kordeleski: InfoBuilders, New York, which supports business intelligence efforts; CUESolutions provider AdvantEdge Analytics, Madison, Wisconsin, a “full-service analytics provider that uses a cloud-based solution and offers analytics apps and services”; and CUES Supplier member Trellance, Tampa, Florida, a “full-service data warehouse platform.” 

Using What You Have

As many C-suite leaders look for an AI strategy, managers at progressive credit unions already have tested AI tactics and can bring some experience to the table, Kordeleski says. The decisioning process for approving indirect auto loans automatically is a common example. 

Automated document indexing and recognition is another activity where CUs are chalking up wins, he adds. “Here AI-powered tools can really shine because they can recognize and adapt to new types of documents without having to constantly update and load forms. AI can learn on the fly how to recognize new types of documents and classify them.”

And keep working on the member experience, Carter urges, which doesn’t mean just digital, but high-performance digital. 

“Most credit unions are converting to digital— digital account openings, loan originations, mortgage underwriting—but just having it doesn’t mean it’s working well. I bought a house in Memphis recently and did it online but had to input the same information three times. That’s back-office inefficiency. AI could have eliminated that. It can expedite processing. That can mean pulling a lot of relevant data from different systems and using it almost instantly to complete a process or make a decision.”

Outlines of a Vision

The likes of Citibank and Bank of America have aggressive plans to convert service decisions to AI, but that doesn’t mean CUs should plan to do everything the largest players in the market are doing, suggests Peter Scott, a futurist and founder of The Next Wave Institute (nextwaveinstitute.org), Vancouver Island, British Columbia. 

Instead, CUs should implement AI selectively, on a scale appropriate to the institution.

“For loan applications, AI can tell you yes or no and how much and quote a rate, but it can’t tell you why and how the applicant might change that outcome,” he explains. “Here’s where the friendliness and accessibility of a credit union can shine.”

Scott gets into concepts that may be new to some C-suites. AI can create “digital avatars” that become “co-workers” alongside people in adjoining cubes. In these scenarios, a person and an avatar can work together as a team, he says. He cites innovative work in this area by IPsoft’s Amelia and mentions IBM’s Watson, Magic Leap’s Mica and Samsung’s Neon.

“Amelia, for example, combines natural language processing, selectable domain-specific knowledge, and on-the-job learning, to make her the ‘smart kid in the corner cube’ ... who does some of the rote tasks and answers specific questions,” Scott elaborates. “More specifically, Amelia can accept a mortgage application, conduct a conversation with the client to refine the responses, and provide team members with recommendations, following instructions for advancing the application or answering questions about its status.”

Implications for the C-Suite

Of course, the elephant in the room is whether AI can replace the people in the C-suite. Scott thinks it’s possible because, essentially, a C-suite is a system with functions that can be recognized and automated.

What you have now, Scott explains, is people sitting in meetings pooling their collective intelligence. Corporate staff and business unit leaders talk about what they are seeing, and then the CEO makes a decision based on their input. That’s a lot like what AI does, Scott notes, only AI does it faster. 

McNair, on the other hand, is not concerned that AI will ever replace executive management. “It will never control decisions,” he predicts. “It will take over routines and be programmed to make low-level process decisions, giving management more time to be strategic and creative. It’s just a tool for helping people do their jobs better.”

But executive team members might consider changing their roles, Scott suggests. “For centuries, people have worked to gain business wisdom from studying financial statements. AI can now do that for them. They can pivot to a more human orientation, more about helping people realize their dreams or find stability and less about poring over numbers.”

With that change, the composition of the C-suite also needs to change. “Traditional roles are losing value,” Scott notes. “Leadership needs to be re-engineered to complement the role of AI systems.” And it will absolutely take a new kind of board, he insists. He recommends reading The Fifth Discipline by Peter M. Senge.   

What the C-Suite Should Know About AI

The C-suite learning agenda should include three questions about artificial intelligence, according to Vladimir Kovacevic, co-founder and managing partner at Inovatec, Burnaby, British Columbia.

1. Does your lending infrastructure work with AI? This is critical, Kovacevic says, because AI can’t be bolted onto a legacy system. To properly leverage various services and solutions, a lender needs core systems built to support or embed AI from the ground up. If the answer is no, look to replace and upgrade the legacy infrastructure. 
2. Are current processes and policies precisely documented? And is there a workflow that’s consistently used and followed? If not, fix this. You need that vision to identify what parts of the process can be automated, improved or augmented by AI solutions. Defining your process also makes it possible to build proper business cases and demonstrate the return on specific AI investments.
3. Is there an in-house AI center of expertise? Or will a partner be engaged to assess and recommend the appropriate solutions? Jump right into this alongside steps 1 and 2, as it will take time for in-house resources to get trained and up to speed, even to work with an expert partner. cues icon

Richard Gamble writes from Grand Junction, Colorado.