The massive shift to working from home has put virtual collaboration companies like Zoom in the headlines. The same dynamic is playing out in healthcare as hospitals—in an effort to prioritize in-person care for coronavirus patients—are trying telehealth product suites, such as TelaDoc’s, to manage cases and scale up.
AI’s role in getting us through this pandemic remains less straightforward. That’s because AI can’t be as readily plugged into organizations that have been forced to go virtual nearly overnight. In addition, AI serving in point solution roles has limited impact.
Benchmark Capital’s Chetan Puttagunta was insightful on the first point. As the pandemic accelerated in the U.S., he reminded us that during previous downturns, technology vendors whose solutions could be implemented quickly and easily rose to the top. If it takes too long to implement a technology, the customer is in a holding pattern.
Paul Strassman, the Pentagon’s head of IT in the 1990’s, noted an important distinction between managerial productivity and operational productivity. In fact, Steve Jobs shared Strassmann’s thinking on this to a class at MIT in a memorable lecture, where he showed on a chalkboard how managerial productivity helps you do a few things well, and then the value tapers off. That’s opposed to operational productivity, which enhances everything the enterprise does.
So AI needs to take the form of an operational productivity solution that has broad impact on the industry it serves. It can’t be a tool offered out of context with an industry’s workflows. It has to be purpose-built, capable of addressing an industry’s unique challenges.
The pandemic is likely to winnow the pack of AI companies down to a smaller group of enterprise suite vendors. Some will run out of cash and others will realize they need to go back to the drawing board.
Effective tech leaders are looking for early lessons from the new world created by Covid-19. Fundamentally this crisis means utilizing integrated digital systems to recognize and respond to emerging risks and consumer demands, rather than simply automating existing activities and workflows. That is the test AI solutions vendors face.
Until now, most predictive AI technologies have been based on assessing two variables retrospectively. AI can explore multiple variables and how they change through time in relation to each other. Making this easy to plug in and deploy as a suite that delivers operational productivity makes new solutions possible.
For example, in the insurance industry, it could mean agile claims intake processes and alerts to emerging threats in a multi-variable world. In financial services, it can mean real time understanding of liquidity and capital reserves and when best to utilize or increase these reserves. In healthcare, it can mean empowering front-line clinicians with tools that pull in data for collective use, and also help them make more informed treatment decisions.
Let’s return to banking for an example that’s hitting home this month.
Because the stimulus loans can be forgiven and the circumstances around how they are forgiven are still confused, it presents a target rich opportunity for organized crime, or just plain opportunism, which means fewer funds for the small businesses that really need them.
Examples include fabricating the number of employees, taking the loan and then firing the employees and multiple other frauds that occur not just at the application but throughout the lifecycle of the loan. Add to it that there has been an eruption of hacking and digital crime generally as the pandemic has intensified, and it’s not unheard of for a program this large, that is launched this fast, to have as much as 80 percent of initial applications be fraudulent.
Finally, because of the volume and speed of the applications and money involved, you can’t rely on past historical data to predict where fraud will occur. Doing things faster and at higher volumes changes their characteristics. An event like this one can attract new fraudulent actors who can’t sit on the sidelines when they see so much at stake or are so desperate, and they may show new behavior that you would not have expected.
The solution is not to re-haul the system—there’s no time for that. But establishing a set of risk key performance indicators through the loan servicing process is crucial. There is a decent chance these loans will start to be securitized at some point in the future. So good clean risk assessment will be an absolute must. Otherwise we are back to 2008 and the toxic mortgage packages no one understood.
The complexity of the moving parts over a long time make this a nightmare for all banks—AI enabled or otherwise. Most banks using AI have not scaled their projects to the type of volumes and computational complexity that will be needed. Some have and they will feel more comfortable both now and as the secondary markets for these loans evolve.
Banks will have to deploy an enterprise class AI engine quickly to segment their customers, adopt behavioral intelligence in decision making and establish a longer term loan risk understanding based on immediate data and expected behavior in the future.
Keep in mind that the mortgage housing crisis was like a slow motion stimulus program. Money was given out to support a social good—home ownership—which stimulated the economy. But ignoring the quality of the loans caught up with all of us.
Now we have to do the same thing on a massive, high speed scale. TARP was overall a well managed stimulus; yet fraud still accounted for about $11 billion. This time around we should expect that number to be, at best, $50-$70 billion.
There’s a lot at stake. Speaking recently to a group in his Connecticut hometown, Strassman pointed out the tensions in the global economy that could be pushed too far by today’s events. His small book on the subject, Growing Old in a Turbulent World, has already sold out on Amazon—possibly an example of panic buying during uncertain and pivotal times.