A common fallacy exists for people building data science teams that: smart hires translate to successful data science teams. Flawed concepts might include:
“If I can just get my hands on a Stanford Ph.D. data scientist everything will work out”
“If I have two hundred applicants my top cum laude pick should be just fine.”
“This individual worked at Google/Uber/Facebook, therefore, they must be amazing”
“Rebranding and promoting our best internal analysts as data scientists = retention win & cheaper salary spend”
“These founders are the smartest people I have ever seen pitch => $14.5M seed”
Meet the money-stupid data science team
The number one reason smart data science teams fail to win and provide value at the rate that they should is money. Sure you pay them well, but they just don’t get the business drivers. They can’t speak the language your board members, managers, and customers need to hear. Despite their data genius, they are idiots in the business world.
Symptom (1) Get me a translator!
These symptoms are manifest to managers when your data science team lead communicates in confusing data jargon (AUCs, confidence intervals, random forests, deep learning, clustering). The best thing to show an exec is a number with a dollar sign. If you are reviewing confusing plots and struggling to understand the business value your team is sick with a data-to-business translation issue.
Symptom (2) Must be management’s problem
“My managers just don’t value/enable our data science organization. We seem to be doing more BI and custom analysis work rather than tackling the major wins.” – sad data scientists everywhere
Sure you have the whole range of executives from genius to incompetent, however there is a good chance the person making this complaint has failed to communicate in terms of relatable dollars. Instead of doing anything about it, they see it as being something outside their control. Remember you are typically dealing with a bunch of introverted nerds, not the type to cause the necessary tension for change.
Symptom (3) Not Failing Quickly
There is nothing noble about continuing to invest in a bad idea. Learning to fail quickly and being willing to kill your own projects can be very difficult. Even better, bringing in customer feedback before chasing a project can really help. In the spirit of failing quickly coming up with loose assumptions and shortcuts can help here. Too many data science teams double down, triple down, and ultimately fail on a bad idea. Understanding the money and the business drivers can help clarify when time ranks above statistical rigor.
Symptom (4) Stupid solutions beat smart ambitions stupid
What is the best way to do this? What is the most challenging way to do this? What would my data science peers respect the most? What would make me look the smartest? This is a temptation that all data scientists fall into. We are all really smart, we want to work on challenging problems. We want to learn new things. Understanding the money helps a good data science team focus on the lowest hanging fruit. What is the most valuable win for the least amount of time/money? Even a very basic bayesian or logistic regression can be a fantastic starting place for most products. Keep It Simple Stupid (KISS), we’ve all heard it, always start simple and focus on the business win. After you have a simple win then sure level up on the approach and take it to ridiculous but never start with your most advanced idea.
Symptom (5) Know when to outsource, chase acceleration over job security
Too many data scientists have the attitude of “Oh I think we can do that.”. Yes, yes maybe you can in 6-12 months, but this product does it today at scale. Remember how time is money? Smart data scientists fail to grasp that. Having a data science team that resists outsourcing to third party solutions is an example of a red flag on inefficiencies. Sometimes managers fall into this pit as well where they think their IP or secret sauce has to be custom built. It doesn’t. For most companies, their IP is in their data, not in the specific modeling method.
Symptom (6) You’re not in graduate school anymore Dorothy
I love this one from James Taber in the comments below. So many data scientists never leave graduate school. Their timelines are dramatically different from the GSD attitude needed to survive in industry. For the business leaders, they want stuff ASAP and they could care less if the quality is sufficient for a white paper. Need a sentiment, spam, churn, model? The data scientists who have moved beyond graduate school can package up a finished solution and story before lunch. The others will do what I call “academic spinning” where their timeline will include no urgency and multiple iterations on things that don’t matter.
If your data science team hasn’t provided tangible business value after their first 6-12 months of existence you have some serious issues that need fixing.
What are the number one reasons you think smart data science teams fail to offer business value?
