Most companies know what ethical behavior is, but many still aren’t good at practicing what’s preached in corporate mission statements and HR policies. Headlines filled with insider trading, bribery, sexual harassment, and customer deception remind us how difficult ethical behavior is to enforce. Could advanced technologies, including artificial intelligence (AI), finally help companies solve an age-old problem?
AI has proven itself in several areas including automating repetitive tasks, anticipating customer needs, and enhancing customer experiences with voice interfaces. Keith Strier, AI advisory leader at EY Global and Americas, explains that corporate ethics could prove be one of AI’s most surprising frontiers.
“One of the best-use cases for AI is to look at information and to identify patterns that stand out,” he says, adding that this applies just as readily to looking for unethical behavior as it does to recognizing cat pictures. “As long as there’s data related to it and you can analyze it at scale, AI can detect anomalies that humans simply can’t.”
(Machine) Learning To Follow The Money
Finding many ethical violations is a simple case of following the money, yet conventional audits are remarkably ineffective. Just 15% of instances of corporate fraud were detected by audits, compared to 40% that came from whistleblowers, according to a 2018 report by the Association of Certified Fraud Examiners.
The problem is volume, explains Robin Grosset, chief technology officer at MindBridge, a company that sells AI-powered auditing software. There are too many transactions in most companies for auditors to go through manually.
Auditors cope by randomly selecting a subset of transactions and validating those. “Then, when they’re producing the audit opinion, they’re essentially arguing that because they found nothing in the sample that the books are in good shape,” Grosset says.
MindBridge uses machine learning models to hunt for inconsistent and suspicious transactions that could point to unethical behavior. Each model is trained to look for specific red flags. These can include scenarios such as unusually large transactions, capital flows between unusual participants, or manual system entries close to the end of the year. It highlights the red flags in a report and assigns a risk score, explaining where and why human auditors might want to look further.
Other systems can go deeper, using AI to do human-like detective work. AI auditing platform AppZen for example, uses three interrelated AI technologies to help analyze transactions, according to CEO Anant Kale—deep learning, computer vision, and natural language processing.
The company’s computer vision system can look at everything from a picture of a restaurant receipt to a printed expense report. “Our AI engine essentially reads every document, every image, every travel itinerary, whatever the employers provided, just like a human would,” he says.
AppZen does the kind of snooping that a dedicated auditor with unlimited time might do, says Kale. Let’s say a sales manager in a far-flung office somewhere submits an entertainment claim, attaching a receipt for a place called Sizzle. A time-strapped auditor might wave it through assuming it’s a steak restaurant. Someone who took the time to check out the address online might likely discover something else.
“Our AI engine does the same,” says Kale. “It finds out about that merchant, looks at the social reviews they might have, and finds out whether there are expenditures that the company would not likely be associated with.”
The semantic analysis it performs can also detect different forms of fraud. If it reads a business trip car rental receipt and sees a child’s car seat listed, it would flag that as an inappropriate expense.
The kinds of transactions flagged in auditing software could be one-off infractions, but some experts are using AI to help understand and predict more systemic patterns of unethical behavior. The hope is that AI ethics engines might be able to stop ethics violations before they happen rather than catching them after the fact.
AI can be good at this because it analyzes behaviors that aren’t always consistent or rational. Before Jeff Lui became AI director at Deloitte, he coded his own AI system to help an enterprise consulting client find ethical weak spots across the company. In doing so, he had to adapt the system to account for human relationships.
His AI algorithm found a hotspot of inappropriate expense claims in one specific office across the whole country. Investigating that office, managers found a very close-knit team that would always socialize outside work. It was more like a family, he recalls.
Peer pressure, it turns out, nudged people toward unethical expenses. “These teams were more likely to go down that ethical slippery slope because they knew each other. Once we had that insight, we built that into the algorithm too,” he says. “It was neat to build that softer aspect into a very cold-blooded financial algorithm.”
Leveraging AI To Change Behavior
Denver-based Convercent is taking this one step further—using software tools to look beyond expense claims to other data sources in a bid to predict when unethical behavior is likely to occur. Its platform scours HR systems for data around salaries, office locations, travel itineraries, and training histories. It even looks at how people browse through corporate codes of conduct to detect potential problems.
AI algorithms crunch these numbers to produce statistical models that can detect risky situations, says chief strategy officer Philip Winterburn. “A salesperson flying to China, for example, is immediately in a high-risk category. If they’re going there and entertaining government officials, they’re in an extremely high-risk category,” he says. Convercent’s software can send that person a text message reminding them of the Foreign Corrupt Practices Act and connecting them to the section of their company’s employee handbook warning against bribes.
“Those kinds of nudges at that point in time when you’re embarking on something risky are the most effective way to modify behavior,” says Winterburn.
To be sure, it’s still early going to use AI as an ethics tool, says EY’s Strier, and many companies remain unaware of its potential. But as they become aware of its possibilities, he hopes that the technology will evolve still further to move beyond simple transactions.
“There’s one version of this that looks at financial data or human behavior through some kind of transaction,” Strier says. “The other version is looking at actual human behavior in the real world in real time, and is able to identify potentially illegal or anomalous behavior.”
Strier predicts that AI could soon analyze audio or even video feeds for unethical employee behavior. “In financial services, conversations that financial advisors have with consumers are highly regulated. If they deviate from the script, it could border on unethical,” says Strier. AI tools could eventually monitor those interactions as a consumer safeguard.
Even the most advanced machine learning models will still have trouble detecting or preventing the most egregious actions. The most effective means of bringing attention to unethical behavior is for human colleagues to call it out.
But for the thousands of accidental or opportunistic ethical infractions that put companies at risk every day, emerging AI tools could nudge tens of thousands of workers toward doing the right thing a lot more often than they may do now.