Business speaker and author Tom Davenport says the 'moonshot' approach only works for big firms when deploying AI - and everyone else should take small steps first
Named as a “top ten voice in tech” by LinkedIn and one of the world’s top 50 business school professors by Fortune magazine, Tom Davenport is regarded as one of the top thought leaders on the planet when it comes to technologies including artificial intelligence. He sits down for a conversation with Dan Robinson
Many people go into business with visions of changing the world.
No one wants to temper such ambitions but unless you’re Amazon or Google, perhaps the best route to achieving transformation is taking baby steps first.
Tom Davenport calls it the “low hanging fruit” approach. While some entrepreneurs may want to reach for the stars immediately – the “moonshot” – evidence suggests the best results come from tackling a series of easier challenges first, particularly in the deployment of AI.
“You need collections of low-hanging fruit,“ the prominent tech professor and advisor tells Compelo in an interview at the IPsoft Digital Workforce Summit in New York City.
“They’re projects that aren’t going to get everyone excited but they seem to prevail over moonshots.
“They’re much easier to accomplish and are more suited to AI because it does particular tasks rather than whole processes.
“If you want to achieve something big like autonomous vehicles, that’s a lot of different capabilities that you have to string together.
“We’ve been expecting driverless cars for more than 30 years and they’re still not here, so it’s better to take smaller pieces and then connect them – then together they’ll add up to something.
“Big and very tech-sophisticated companies are the only ones that can achieve those moonshot aims and, even then, the likes of Amazon have been very challenged to make it successful.”
Tom Davenport gives examples of the low-hanging fruit approach in business
Tom has built big reputation in the technology industry as someone to listen to, with 20 books and 250 journal articles under his belt.
He is the president’s distinguished professor of information technology and management at Massachusetts business college Babson College and a fellow of the MIT Center for Digital Business among other roles, while he advises businesses including Deloitte Analytics and has taught at institutions such as Harvard Business School.
In his most recent book, The AI Advantage: How to Put the Artificial Intelligence Revolution to Work, Tom highlights the work of Singapore-based multinational DBS Bank as an example of the low-hanging fruit approach to innovation.
“It tried to do something really ambitious for recommending stock and bond investments using [natural language processing system] IBM Watson but it didn’t work out because the technology couldn’t read charts and graphs properly,” he says.
“So it said ‘what else can we do?’ DBS had a digital bank in India so it launched a chatbot for that to keep costs low.
“The bank noticed its ATMs were running out of cash in Singapore so it started to use machine learning to predict when to replenish them, and improved it 55-fold.
“It then realised how important its sales people were for bank relationships so it used machine learning to identify people who would most likely leave the bank so staff could try keep them.
“All those are very practical and realistic things to do given today’s technology, and now DBS is doing a lot more of this stuff.
“It’s very important to have those things add up to something big so you keep people motivated and the money flowing into AI.”
There are numerous other examples of taking small steps towards a bigger goal working out.
Pennsylvania wealth management company The Vanguard Group combines automated and human investment advice to recommend things like portfolio rebalancing, tax loss harvesting and retirement income scenarios.
Tom admits it’s not the most sophisticated of tech but has reduced costs by 0.3% and led to lower wealth thresholds than most human investment advice sources.
Only $110bn (£84bn) of the firm’s $5.3tn (£4.1tn) portfolio of assets under management are automated but this is rapidly growing and he claims it now has the biggest pot of robo-advised clients.
Vanguard has also piloted Amelia – US software firm IPsoft’s “digital worker” tool that uses sophisticated AI to enable natural conversations between humans and machines – in its call centres.
“Humans are still around as advisors,” says Tom. “But they work more in financial psychiatry now – telling you not to jump out of a market if it goes down by a few points, for example.”
Credit card company Capital One, meanwhile, has become an “AI-first” business by deploying more than 1,000 minor artificial intelligence applications to achieve its overall aim of “reducing friction” in customer experience.
These include creating Eno, a virtual assistant for customer transactions and early fraud warnings, and moving all data to the cloud to make broad machine learning easier.
“Capital One is right up there with the most impressive non-tech companies for using AI,” says Tom.
Even Amazon – a company that has achieved potentially life-changing moonshot goals such as the Amazon Go self-service stores that are set to be rolled out across the US and its new drone delivery service – has placed a large emphasis on a series of “invisible” machine learning projects that ultimately improve its operations and customer relationships.
These include achieving marginal gains in demand forecasting, product search ranking, product and deal recommendations, merchandising decisions, fraud detection and machine translation.
In a 2017 letter to shareholders, founder Jeff Bezos said: “Though less visible, much of the impact of machine learning will be of this type – quietly but meaningfully improving core operations.”
Tom adds: “This says to me that, bit by bit, they’re making Amazon more effective – even at a huge company, the low-hanging fruit approach is paying off.”
Tom Davenport on why the moonshot route often fails
Not all companies go straight for the low-hanging fruit approach, however.
In 2016, pharma giant Pfizer launched a partnership with IBM to use the IBM Watson for Drug Discovery platform to accelerate research in immuno-oncology, which aims to recognise and eliminate cancer cells using the body’s immune system.
It sought to use the machine learning and natural language processing technology to better identify new drug targets and more efficiently select patients for clinical trials.
But Tom claims it didn’t work and it was reported last month that IBM was ending sales of the drug discovery product after failing to yield large enough financial returns.
While Pfizer’s “moonshot vision” hasn’t delivered the desired results, it has still used machine learning in about 150 smaller projects, which include identifying which patients are most likely to take all four of the required hepatitis vaccine doses and those likely to stop after the second dose.
Another case that backs up Tom’s point was at the MD Anderson Cancer Center, in Houston, Texas.
IBM Watson was once again the technology being used for a project launched to much media fanfare in 2013, this time with the aim of effectively curing cancer.
In practice, this meant speeding up clinical decision-making and matching patients to clinical trials more efficiently.
The AI programme cost more than $62m (£47.6m) – including $50m donated by Low Taek Jho, a Malaysian financier now on the run from US authorities after being accused of money laundering – but reportedly did not treat any patients and a 2016 report by auditors at the centre called it a failure.
Despite this, the centre used augmented intelligence software by Austin company CognitiveScale to create a “care concierge” system that helped patients’ families with directions when visiting, identified patients that needed help with paying bills and gave staff IT support.
Together, these smaller schemes – while not as newsworthy or as significant at a wider level – improved patient satisfaction and financial health at the centre, with other similar projects now underway.
Tom says: “AI isn’t quite ready to cure cancer and that was a really ambitious objective. But there are some smaller goals it can achieve, and it’s a clear example of a low-hanging fruit paying off when the moonshot approach didn’t.”
Tom Davenport on artificial intelligence in the enterprise
Earlier at the Digital Workforce Summit, Tom gave a keynote speech in which he touched on the state of AI in large US firms.
The MIT Center for Digital Business fellow’s own research found that between 20% and 30% of senior executives were “AI-aware” and actively employing multiple technologies in their business.
Only a few of these are “AI-first”, with Tom highlighting the investment bank JP Morgan Chase as a good example in spending $11.5bn on technology research and development each year and employing more than 50,000 people in IT.
“Large firms are setting up infrastructure at a pretty impressive rate – I don’t think we’re heading towards an AI winter anytime soon,” he says, referring to a period in the 1980s and 1990s when a perception of the technology being over-hyped led to a starvation of funding that was only revived from about 2012.
Machine learning is the most popular tech, with 63% of firms engaging with AI adopting these models – a rise of 5% compared to 2017.
Half are using deep learning – the biggest riser as it represents a 16% increase in the past two years – while 62% use natural language processing and 49% employ robotic process automation, whichTom describes as the “dumbest type of AI but getting smarter all the time”.
The aims of businesses using AI can vary but the most popular was to enhance current products and services, which was the goal among 44% of executives.
Other common targets include optimising internal and external operations, making better decisions, freeing workers to be more creative and creating new products.
Tom says: “What’s interesting is the diversity of things companies are trying to do with AI and they’re pretty much the same as what they were doing a few years ago.”
Tom doesn’t buy into the notion of companies wanting to displace human jobs through automation – this was second bottom as an aim in his research – although he admits to being disturbed by findings in the Deloitte State of AI in the Enterprise survey last year.
It found that 63% of companies want to cut costs by automating as many jobs as possible, with only 36% believing job cuts resulting from AI-driven tech being an “ethical risk”.
Only 10% preferred to keep and retrain current employees, while 78% wanted to either keep or replace staff in equal measure – or replace its workforce with new talent.
“I’m an optimist,” says Tom. “Augmentation, which is machines and humans working together, is going to be more likely than automation.
“That’s what’s happened so far in the vast majority of cases because AI only does narrow tasks, not entire jobs.
“Pure automation locks you into a few things and is a bad strategy because your people won’t want to work with you to make the AI happen because they’ll be pushed out.
“But no one knows for sure what will happen. If we see AI systems that are better than us for everything, then all bets are off.”
Tom Davenport on industries best using AI
While the big tech firms like Facebook and Amazon are naturally among the biggest proponents of AI, other industries are getting in on the act too.
Tom identifies financial services firms as the most aggressive outside of this domain, adding: “The companies that are very aggressive tend to do a lot of things.
“They have a tonne of people related to risk and machine learning, and they’re trying to do some targeted marketing with AI.
“Then you have the insurance companies getting pretty aggressive with things like underwriting, which is becoming more algorithmic.
“You crash your car and send the insurer a photo, and it will tell you whether or not it’s totalled without a human having to see the vehicle.”
Tom says telecoms are deploying AI effectively to identify which customers are most likely to leave through natural churn in a bid to retain their service, while retail and pharmaceuticals are other industries beginning to adopt the tech.
“Beyond that, there’s not a whole load going on,” he adds. “Some manufacturers are starting to put sensors in their products to identify when something will break down but there’s certainly more that could be done.
“The area of personalised recommendations has been very slow to develop. Amazon’s algorithm is quite old and it’s not using state-of-the-art technology for it.
“Some of its acquisitions are doing a better job, like online shoe retailer Zappos, which has sophisticated recommendations.
“Eventually we should be able to use AI for highly personalised recommendations but it’s stubbornly slow to arrive.
“A lot of the time, companies aren’t using attributes of their products to understand what customers want.
“Someday we’ll get there but we’ve been hearing that for a long time.”