Artificial intelligence has a long and storied history. It evokes images of robots taking over society, doing our dishes, balancing our checkbooks, winning at chess. The New York Times recently found it performing standup comedy.

Artificial intelligence also has a long history as an overhyped dud.

The popular imagination about robots has led to bouts of overexcitement, hyperbole, futuristic evocations and finally impatience and anger whenever AI technologies fail to develop quickly enough for the masses. That disillusionment has led to numerous periods of “AI Winter”—specifically in the 1970s and 1990s—eras when funding dried up from impatient investors. And a new winter is likely coming, warn the Cassandras.

That hype part of AI is still very real, says serial entrepreneur and technology consultant Sultan Meghji.

“In 25 years, I have seen in the private sector six companies do artificial intelligence properly, six in total, and not half of them are financial services,” says Meghji, the founder of Virtova, a tech advisor to private equity firms and financial services companies. In other words, he thinks top-to-bottom AI, where every link in the chain is automated, is not really being done. 

But put the delusions of grandeur aside, and the reality is that AI has come a long way in the last few years. Its algorithms have become more robust, Meghji says. The data it mines (AI’s lifeblood) has become much richer and more accessible. People are now building ETFs out of artificial intelligence, too. Advisors have learned to find clients with it, learned when to engage clients with it. Pershing even says AI can tell you when a client is about to leave you.

A lot of fintechs are waving the AI flag, Meghji says, and in the back office, Morgan Stanley, TD Ameritrade and JP Morgan are building robust AI programs. Last October, Greenwich Associates said in a report that 15% of jobs in finance are at risk of being replaced by AI or machine learning in areas like research, trading, analysis and sales.

A Tool For Advisors

If Amazon knows what sweater you like and offers you another, just think of what the right program could do for a person looking for 529 plans and estate plans.

That was the thinking at Morgan Stanley, which dived into its first artificial intelligence programs five years ago, says Jeff McMillan, the firm’s chief analytics and data officer. After years of development, Morgan Stanley rolled out a pilot program in  June 2017—a platform called “Next Best Action” that allows advisors to offer clients products, investments and otherwise, based on what it knows about them. Some 16,000 firm advisors now have access to this tool.

The seed of the idea was time efficiency. On any given day, thousands of things are happening in the market, McMillan says. The percentage of clients reachable on the telephone is declining, and no human being can adequately monitor all the market conditions going on or understand every action a client might want to take in response.

That’s what AI is for. It learns to select and subtract topics and prompt advisors about whom to call and when among high-net-worth clientele. “What it does is it profiles every single client, every single portfolio, every single day and intraday and evaluates literally thousands of ideas,” McMillan says of the firm’s new rollout.

These algorithms can look at new data—such as the downgrade of a company or a client’s equity exposure in Japan. The algorithms might also prompt a client to use a client aggregation tool, or send a safe driver manual to a client whose child is turning 16. The technology scores clients on the relevancy of this data to their situation and then looks at what actions they’ve taken before.

“So for example,” McMillan says, “over the last six months I have sent you 31 single security equity ideas, both downgrades and upgrades, and you have opened 94% of them. You have clicked through on the content 83% of the time. And you’ve transacted on 47% of the ideas that I’ve sent you.”

Those ideas get sent to the advisor every day and morph over time based on clients’ behavior, he says.

Next year, the firm wants to launch another program in which it uses natural language processing to understand human questions, the way the Amazon Echo might, from a client conversation. “We have 40,000 something employees. We have knowledge about every single financial aspect of every market in the world. There is somebody somewhere who knows about trusts and estates in the state of Utah. How to dispose of a Monet you find in your attic to where do we think the long bond is going. Traditionally, that knowledge is built into a bunch of static PDFs and very hard to access.”

So the firm is collating that knowledge using artificial intelligence in the form of natural language processing and building bots for financial advisors, serving as intelligent assistants when clients ask advisors those questions. Advisors couldn’t and wouldn’t likely be experts in everything, after all. “If a client called and said my mom was just diagnosed with early-stage dementia, what should I be thinking about? Or my 2-year-old son has been diagnosed with autism.” The AI arms the advisor with the knowledge. “Because the truth is, there is a person at Morgan Stanley who can answer those questions.”

Picking Stocks

The company EquBot was founded to use AI to pick stocks for an ETF.

EquBot’s exchange-traded fund, the AI Powered Equity Fund (AIEQ) uses machine learning as part of its portfolio selection using IBM Watson and Google DeepMind. One of the ways it learns is to read databases full of newspaper articles to see how companies perform in different environments.

“Each day,” says its prospectus, “the EquBot model ranks each company based on the probability of the company benefiting from current economic conditions, trends, and world events and identifies approximately 30 to 70 companies with the greatest potential over the next 12 months for appreciation and weights those companies to seek a level of volatility comparable to that of the broader U.S. equity market.” Humans then rebalance the portfolio.

By picking the best mix of stocks for the current environment, based on performance and risk, it’s not wrong to compare it to the shifting baseball teams in Moneyball, say its managers. The algorithm learns to put words like “steel tariffs” in context to decide whose ox gets gored by them.

“AI is specifically useful where there is a significant amount of data,” says the firm’s CEO, Chida Khatua, a former director of engineering at Intel. “And there is a way we can find or train the systems to understand the correlations.” That means evaluating both Excel spreadsheets and newspaper stories.

The $167 million fund’s AI model is based on how investors do due diligence on a company. They look at management, financials and how insight from news will influence companies’ profitability or equity prices. The idea is that the market trains the system to work better.

The fund charges a 0.75% expense ratio. The firm launched an international fund in June (AIIQ).

Robots Hunting Clients

Advisors are also using AI for marketing. Shirl Penney, CEO of RIA services firm Dynasty Financial Partners, recently told attendees at a Pershing Advisor Solutions’ RIA Symposium in New York that one of his firms on the West Coast did an entire marketing campaign based around artificial intelligence to bring in some $21 million in new client money.

“We wrote some simple white papers, very short and easy on ISOs, stock options, for Snapchat employees,” Penney said. The firm then used LinkedIn to target executives at the company. “The strategy was very simple: To position this group of advisors as experts in stock option cash-flow analysis, diversification, etc., running a concentrated position, in this case for Snap. And then we had a very simplistic AI tool that interacts with the client.” The advisors set the software and then let it do its work, Penney said. If a client clicked through (“raises a hand”) the software sent more educational materials on options. Ultimately, a meeting could be inked in.

“In the last campaign that we ran, [we got] 16,000 impressions. Nine hundred times the employees clicked through. It automatically, through the system, set up 15 meetings. Of the 15 meetings, the advisors closed seven of them. It was $21 million in new assets. Let me show you how much that campaign cost: $600.”

Why Clients Leave

Ram Nagappan, chief information officer at Pershing, says machine learning can show patterns with clients and help predict when they might defect for another firm.

“We are using machine learning to give some kind of head’s up that this client could potentially leave,” Nagappan says. “We take data on who is the client, what gender, what status, tendency, tenure, all types of information that you have on the book of the business, then we use the previous history of why a client left and recreate a pattern. … This is a learning technique. You want to learn whether it is right or wrong and then you can correct the algorithm.”

Nagappan says the machines look at the assets a client has, the assets she’s transferring over a period of time, what similar clients did during book transfers and what transactions they pursued—as well as the clients’ ages, genders and employment statuses, dependency and how much online interaction they have had. The algorithm then shows patterns in a book of business similar to the patterns of clients who left in the past, Nagappan says. If someone is doing money transfers, or logging in and checking more frequently than ever, those are warning signs.

He mentions the big improvements in AI tools by Microsoft, Google and Amazon. Pershing offers these through its own portal and offers tools through its advisors’ own portals.

Not A Replacement

But still, even in world of digital advice, clients still want to engage with human beings, says McMillan. Morgan Stanley, for one, has no intention of replacing them.

“We fundamentally don’t think that’s the game we want to be in, nor should we be in,” McMillan says. AI and machine learning technology does things that are unique, he says, including storage and memory, things humans aren’t equipped for. “They are able to process millions and billions of permutations. They don’t forget things, they look for anomalies.” But humans are able to infer things from other people, read gestures and faces. “And they are able to contextualize multiple data facts and assimilate them in ways that [robots] simply can’t do yet.”

Hype aside, there is still a powerful gee-whiz quality to AI that allows the visionaries to come out, says Meghji, and there is lots of room for disruptors. He sees a big one coming in the insurance industry, which he says is a field about to be invaded by robots. Robo-platforms here will likely do the same things they do for investment platforms—finding not ETFs but insurance policies for people who need to be covered and have only to pop in their risk factors on their home computers.