8 Mistakes Leaders Are Making with Data Science and How to Fix Them
And how to more effectively use AI.
“More money than brains” was one of Gramma Jo’s favorite sayings. She hated seeing people waste money. So she’d be upset to see what’s happening with AI today. Surveys show 92% of companies are spending more on AI, but 11% are deploying it effectively.
What’s going on? How can so many companies invest in AI and get so little from it?
The answer is rooted in naive leadership, hype, and fear-mongering.
Here are eight mistakes leaders make with data science and how to do it right, as shown by one of the most innovative data science projects of the year, the National University Healthcare System in Singapore (NUHS).
Mistake 1: You think algorithms make decisions. Forget every movie about AI you’ve ever seen. Algorithms don’t reason, and robots aren’t out to take your job away. They’re here to help and augment human decision-making.
Better to compare AI to the humble blindspot indicator in your car. In driver-assisted car technology, algorithms quietly assess surroundings in real-time. When a car enters your blind spot, a tiny light in your rearview mirror turns on. This is a very, very smart algorithm. It’s cleverly designed and subtle. It’s helpful. It’s AI in the right place.
Smart companies view AI like blind-spot indicators: an assistant, a recommender, a sentinel. And they realize that humans still make final decisions.
NUHS does this. Here’s how:
You go to your hospital
Frontline medical staff takes electronic notes (#1, above) which are transmitted to NUHS algorithms in real-time.
Your data is “scored” by algorithms that match your data to an anonymous, 3 terabyte database to evaluate health risks from cancer to appendicitis.
Recommendations (#2) are returned to your doctor in real-time.
You and your doctor make decisions.
Action is taken (#3). Actions may be as simple as a question the doctor asks or scheduling an X-ray.
All this happens in the blink of an eye. Like the light in your rearview mirror.
So, think of AI like the blind-spot indicator in your car, not a sentient robot.
Mistake 2: You let data scientists lead data science projects. At NUHS, doctors are in charge of algorithms, not mathematicians. A lot of mainstream companies get this wrong—they put data scientists in charge of data science instead of decision-makers and domain experts.
Even the popular mantra—“BE DATA-DRIVEN!” is wrong. The article “Be Decision-Driven Not Data-Driven” was a viral hit on LinkedIn because it argues that decision-makers should grab the wheel. Even data scientists agreed.
But most companies get this backward: they hire a CAO, put their data science team on a pedestal, and isolate it from decision-makers. That’s wrong. Take a cue from NUHS: put decision-makers in charge of AI.
Mistake 3: You don’t see statistical shades of grey. Most decisions are not black and white, and AI doesn’t make them any less grey.
For example, the red highlights in the physician note below are keywords used by AI to assess the probability this patient has appendicitis like RIF PAIN, RIF TENDERNESS, PAIN SINCE AFTERNOON. AI’s assessment, in yellow: 95.5237% of patients with these notes have appendicitis.
The NUHS system sends doctor’s notes to algorithms in real-time to spot health problems. This patient has a 95% probability of appendicitis.
You might think that this patient should be rushed to surgery. But doctors know better: the right action is to take an x-ray!
Effective AI businesses use statistical probabilities in this way. They view decision-making as a uniquely human job, with AI-fueled statistics that augment decisions with shades of grey.
Mistake 4: You worry about security too late. One of the most important pieces of data in the NUHS note above is what you DON’T see: personal information. The only context we have is the patient ID. Their system, which is called Endeavor AI, was built with secure, private, anonymized information flow from day one. This is essential because even the smartest system is useless if it isn’t secure.
Effective AI leaders consider security and privacy from day one, and in everything they do.
Mistake 5: You don’t have a forensics-first mentality. AI is only effective when it’s trained on data. Lots of it. So the next principle of effective AI is to build a forensic-first mentality. As healthcare visits happen in Singapore, each interaction is added to an anonymized database. This data helps data scientists train algorithms and see the forest for the trees.
For example, this slice of NUHS data shows, in red, a cluster of patients that have a high probability of breast cancer.
Source: NUHS
By clustering data points, researchers can discover similarities and improve early detection algorithms, in some cases, over 50 weeks into the future.
NUHS has a culture of forensic data exploration; you should too.
Mistake 6: You don’t prioritize visual storytelling. If a picture is worth 1,000 words, a great data-infused visual story is worth a million.
One of the reasons I love using the NUHS story to explain these mistakes is that their published visualizations are so compelling.
Yet, sadly, less than 5% of the visual explanations I see—and I see a lot—are compelling. I think this is a lack of understanding of the importance, the art, and science of data storytelling.
How can we fix this? In recent years, there’s been an outbreak in popularity of executive “Chiefs of Staff”—the right-hand-man-or-women for executive leaders. Business leaders: consider hiring a “Chief of Storytelling” instead. A well-constructed, data-infused story is more effective than a million of your inspirational words.
Mistake 7: You misunderstand why real-time AI matters. The NUHS system is real-time. Doctor notes are evaluated in milliseconds. Insights are delivered while insights still matter. Some technologist will object: “That’s not real-time! Decisions aren’t fully automated!” Don’t listen to them.
Effective AI is a silent, constant, real-time presence. Like the blind spot indicator in your car, it’s always there, constantly evaluating. That light might come on once a day. But when it does, it matters.
Effective AI is real-time AI.
Mistake 8: You invest in the wrong people and the wrong tools. Most money today is spent on tools that help data scientists do more data science. That’s like buying a Ferrari for a drunk-driving offender: you’re just going to get into more trouble, faster.
There are two ways to correct this mistake:
First, shift investment from data scientists to analysts that know how to APPLY AI. These are people who question decisions, understand business problems, and use AI where it makes sense.
Then, buy those people tools to help them apply AI. For example, model operationalization tools (ModelOps) allow business users to adopt, understand, deploy and manage algorithms on their own and in collaboration with data scientists. NUHS uses ModelOps to build a bridge between data scientists and healthcare operations.
This isn’t to say you shouldn’t hire data scientists and buy data science tools—it’s a matter of BALANCE. You need both.
More money than brains.
My grandmother’s favorite saying was “more money than brains.” She hated wasting money. She’s hate to see what’s happening with AI today. The good news is, these mistakes are easy to fix, and the solutions require leadership and cultural change. In 2022, do this:
Put the business in charge of AI
Use AI to augment decision-making, like the driver-assist technology in your car
Learn to view decisions with statistical shades of grey
Think about AI security from day 1
Adopt a forensics-first mentality
Hire a chief of data storytelling staff
Employ real-time AI
Shift investment from data science to applied data science
Correct these eight mistakes and you’ll make more effective use of AI in 2022, and reap the benefits of smarter, more automated, more empowering decision-making with AI.
