AI is the exponential technology that’s revolutionizing our linear world. If your organization wants to keep up, it’s time to get exponential. We’ve put together a set of powerful principles for managing a team in the Age of AI Abundance.
Data scientists are researchers at heart. Too much teamwork can make them feel restricted and micromanaged. Let them set their own pace and deadlines, and wait for the genius to reveal itself!
Data science is only a tool, not a substitute for business knowledge. You know your business better — if the model output doesn’t match your gut, that’s a red flag. Work with your data scientists to make sure models reinforce your hard-earned intuitions.
Data scientists are just like anyone else — they like to cut loose at the end of a long week! Consider inviting them to play a round of golf, smoke cigars, go to a dive bar, or even work out together. A push-up competition in the office never hurts to get those data science juices flowing!
On Hiring, Building, and Retaining Teams
There’s been a lot of hype about high salaries in data science, but do you really want an employee who only cares about compensation? We recommend you actually set your offers on the low end — this will ensure you only hire people who are truly passionate about the job.
Your company has a lot of data, and this is your strategic asset. What data scientists really want is to work with large amounts of data. The bigger your databases, the fewer other job benefits you will need to offer.
The best way to find critical thinkers is to ask abstract interview questions like “How many feathers does a goose have?” If your data scientist can’t answer this, can they really be expected to make meaningful contributions to the organization?
Self-reliance is a crucial personality trait for a data scientist. If a new hire asks too many questions in their first week on the job, subtly remind them that Google is their friend.
Don’t fall into the “cloud” hype. Having multiple databases with separate owners allows everyone to control their own data. If someone else needs it, they need to talk to you — and probably should buy you lunch.
More data is always better. Be sure to use all of the features available in the entire database, even if there are more features than data points. This is what Andrew Ng’s Data Centric AI is all about.
Everyone loves dashboards, and every data scientist enjoys building out databases to feed dashboards. At the end of the day, data science is really just software engineering with statistics.
On Working with Stakeholders
“If you build it, they will come.” Train the model now, and worry about how the business will use it later.
With the proliferation of AutoML solutions these days, data scientists actually need to know very little about the data itself. Run all of your organization’s data — all of it — through these models until you find a few with good performance. Go shop these winners to the business to find users.
Business owners don’t just want to know what your data scientists did — they want to know how they did it. Make sure to explain the inner workings of the algorithms. In fact, there are now many excellent, free tutorials available on blogs and YouTube to explain how these models work. Don’t be afraid to assign a bit of “homework” so the stakeholders can follow along.
When you create graphs, more information is better. There’s no such thing as too many labels, too many colors, or too many data points. Anything less would be dishonest.
It’s 2022. Deep learning is the state of the art and can be leveraged to solve any business problem. Don’t let your data scientists disappoint stakeholders with boring logistic regressions.
Your organization needs standards, and that means standardization. Give your data scientists a small, pre-approved list of Python libraries they can use. Another great way to reduce your reliance on external code is to have your data scientists re-implement their favorite libraries from scratch.
You hired expensive data scientists for their expertise, so let them get creative. For instance, these days the “distinction” between classification and regression is really just a suggestion. If they really want to use a regression model for a classification problem, give them that freedom.
Data scientists love to learn. So they’ll love to learn data engineering, database administration, software development, and all the other skills they’ll have to pick up when you tell them to productionalize everything themselves.
If your model starts to look bad as soon as you put it into production, don’t worry! Just retrain it every morning. Your users will appreciate the constant updates to the model. And this is good practice for the engineers to redeploy models frequently.
In a well-run data science shop, the previous data scientist should have taken care of everything for you. If anyone questions the model output or your modeling approach, refer them to the prior data scientist. Keep their cell phone number in the git repository in case they ever leave the company.
What are you waiting for?
There's no better day than today, April 1st, 2022, to get started on your AI transformation.
Megan and Brooke’s passion for data science is surpassed only by their sense of humor.