This article by Simon Moss, CEO of Symphony AyasdiAI, an artificial intelligence software company for the financial services sector, explains how Enterprise AI, which has been a favorite topic for technology leaders, is at a moment of reckoning – we are now seeing how much of that talk was hype, and how much of it translates into the results we need when we need them most.
The hubris of enterprise AI businesses is rooted in the enterprise software industry’s resilience in the face of the Great RecessionOpens a new window . Corporate America kept paying for IT despite the carnage, and the assumption is the current meltdown will be the same. People like Gavin BakerOpens a new window at Atreides Management say this time is different.
If he’s right, tech companies and AI in particular need to stop trying to find the best way to explain how their amazing and abstract technology works. We can only focus on fully grasping the source of our customers’ worst immediate panic, and ask what they have in the bag that can make solving it simpler, faster, and better… right now.
Don’t explain how it works, explain what it’s going to do, and how quickly it’s going to do it.
Banks
Take commercial banking. Governments are pouring truly massive amounts of stimulus funding through lenders and lulling financiers into believing speed is more important than anti-fraud rules like Know Your CustomerOpens a new window . It’s a bonanza for financial crime.
Bad actors can leverage the impressive number of identities stolen in the last decade, and the huge numbers of shell companies established to run $4 trillion in laundered money a year through the banking system. It creates a perfect environment for wholesale theft.
Bankers worry regulators can’t be taken at their word. But if that’s true, banks are damned if they move loans out at such high velocity that they risk breaking the rules. And they’re damned if they go too slow or refuse to participate. Simply put, the banks better sort out once and for all an absolute and intimate understanding of who their customers are, what they do, and how they can be effectively worked with.
The problem is standard machine learning can scale up a fraud detection effort, but it will be based on past data and biases. This means you are only catching the criminals who haven’t done their homework, and not the ones already inside your system, or who run sophisticated and likely large operations.
Alternatively, you can shake down more customers by casting a wider net in order to find as many fraudsters as possible. However, bank customers are already frustrated with programs like the PPPOpens a new window effort. Subjecting them to a new layer of investigations, and angering the “false positives†in your customer base is a recipe for problems. That’s where unsupervised, hypothesis-free artificial intelligence comes into play. Auditable findings, perhaps using topology helps.
But a company trying to explain that to a bank needs to explain less of the math, and more of the concrete timing so that black box can be validated. Bank CIOs don’t have time for digital transformation theory. And they don’t have time to understand the difference between topological data analysis, supervised machine learning, and unsupervised machine learning.
Like Star Wars’ Admiral Akbar sitting between the Death Star and the Imperial Fleet, they’re trappedOpens a new window . Enterprise AI has to provide a lifeline that can be deployed right now. That lifeline has to create value right now. It has to deliver a return on investment within a few short months. It cannot demand huge systems integration costs.
Learn More: The Case for API-Based Automation in Banking & Finance in Post-COVID 2020Opens a new window
Too Big Not to Be Cut
Cursed by our own success, the software is now about 50 percent of corporate IT. And right down to dental practicesOpens a new window , businesses are asking and getting deep cuts in their subscriptions in the wake of the pandemic’s economic crash. Software is too big not to cut. That’s especially true for AI which continues to have a high hype to low clear value ratio for most.
Back to banks – yes they desperately need to understand their own customers better. Way better. But at the same time, they have to look at their costs. Right now millennials, the largest generation in history, inheriting $30 trillion in the next 40 years, are wooed by the many digital and diversified financial services providers and traditional loyalty is being diluted.
So as banks digitally transform and think about AI and machine learningOpens a new window (ML) they also need to cut their costs by about 40 percent in the next decade to fund the pivot to new nimble models. On top of the impact of COVID-19Opens a new window , it’s an existential challenge that current events will separate the wheat from the chaff.
Learn More: What Would Be the Impact of Artificial Intelligence on AccountingOpens a new window ?
Banking Tech Redeems Itself
When it comes to fraud, what is needed now is a massively scalable accurate identification and interdiction approach to discover “unknown unknowns.†That includes risks and attacks across jurisdictions, operations and borders.
AI can and should deliver analysis that dynamically, and with minimal manual supervision, evolves as attacks mature and change. Establishing risk metrics to understand both default and “forgiveness†risk enable confidence in pricing and securitization. Low cost, rapid implementation, and low impact deployment is necessary.
That means moving away from business intelligenceOpens a new window , where one is reporting on “known knowns.†In that case, you only have reporting and dashboarding of structured and ordered data. It’s limited to a historical view of what’s happened before.
It almost amounts to profiling. A bank thinks it knows a particular type of bank customer. And bank officers presume a gasoline service station may be involved in organized crime because of where it is, evidence of certain transactionsOpens a new window , and without understand the vetting that franchise has gone through as part of the oil distribution business. In the end, an innocent business may be shaken down, a good customer lost, and a bank’s reputation tarnished within a network of businesses.
Learn More: How AI & ML Can Power Advanced Analytics for Corporate FinanceOpens a new window
Supervised machine learning is better, and allows bank executives to collaborate on “known unknowns.†The programming is done through example, guesswork or history. It is basically a rules engine, however, It’s a process supervised by the knowledge of a past outcome. However, if the majority of the data entered into the system is still from law enforcement files that are a decade old, and doesn’t account for how COVID-19 and political disruptions are changing the world quickly, problems occur. Good customers making small mistakes are turned upside down, and bigger threats lurking in the system get bigger.
Through the unprecedented events we’re experiencing today, we are seeing the launch of unsupervised and auditable machine learning. Topological data analysis is a particularly strong variation. Now we’re discovering “unknown unknowns.†We’re doing subject and object analysis, where the outcome of one thing affects other things you care about. This complex stuff is unknown when you start, but discoverable.
A bank has a higher likelihood of not chasing down the wrong businesses and can concentrate on the truly nefarious problems that have found a way to hide inside the bank’s system like an undetected cancer.
Learn More: Is PKI Crucial for Securing Digital TransactionsOpens a new window
The Good Times
These are the times when the different types of AI start to stand out from each other, and software as a whole does not survive. Only the software that works best.
Robert Smith of Vista Equity Partners famously said “software contracts are better than first-lien debtOpens a new window .†But at that time software was a smaller part of a businesses’ budget, and was deemed essential enough during the Great Recession to escape cuts. Hardware and everything else took it on the chin.
That perception of invincibility may have contributed to the inflation of software valuation in investors’ eyes. It may have led to some less essential companies to float along when the debt was cheap, taking their time to find their purpose in life. It certainly stoked the confidence to launch a raft of AI companies with the idea that there was time to find the value in the software.
But now customers are finding nightmare scenarios where they have to ramp up expenses to deal with big and sudden risks. At the same time, they are trying to preserve cash. The two don’t work together.
Banks in our example are moving vast amounts of small business loans and massively speeding up reviews of whether an applicant is who they say they are, and does the applicant do what they say they do. That takes a lot of well-compensated and qualified specialists when the C-suite and board are trying to keep cash from going out the door. The whole liability of these loans must be making the chief risk officer and chief financial officerOpens a new window lose a lot of sleep.
Like the problem-solvers down on Endor who saved Admiral AkbarOpens a new window ‘s fleet, enterprise technologists and AI need to stop thinking they’re essential and stop going on with their original plan. They need to figure out how to solve the most urgent and burning problems right now with what’s on hand. Make their customers smarter, faster, more transparent, more productive.
We will likely see a dramatic drop in explainer videos from AI companies attempting to educate customers up to the level of understanding of how a particular black box works. But it’s more than solving corporations’ most dire problems. That helps get through the immediate crisis.
Let us know if you liked this article on LinkedInOpens a new window , TwitterOpens a new window , or FacebookOpens a new window . We would love to hear from you!