10 Things We Learned Building an Advanced Analytics Center of Excellence

Advanced Analytics, Big Data, Artificial Intelligence, Machine Learning.

It all sounds good, but how do you actually operationalize a fully functioning Advanced Analytics Center from the ground up?

Good question, glad you asked.

Here are our key learnings from doing this very thing at ATD over the past 12 months:

1. Start at the very top

Your leadership has to be on board — otherwise you won’t have the resources to staff your team, important stakeholders won’t back your work and your ideas won’t get implemented. Educate your C-suite, introduce them to analytics leaders and show them the resources required and the results that are achievable.

2. Solve real problems

It is fun to build convolutional neural networks and apply the latest bleeding edge technology. However, don’t look for a nail that fits your hammer. Instead, solve problems that actually exist. If that means less custom and more standard, then that’s fine. There will be a time for the fancy stuff once you have picked the low-hanging fruit.

3. Don’t sound like a textbook

Chances are your success will depend on salespeople and warehouse workers using your models and tools. If they don’t understand what you have built, they won’t adopt what you give them. So, drop those acronyms and be easy to understand. If you cannot do that, you probably haven’t fully understood the theory behind your model yourself.

4. Know your pitch

So, do data scientists want to work in Huntersville, NC for a tire distributor? Some folks may prefer free coconut milk at some Silicon Valley tech giant.

But do data scientists want to be the founding member of an analytics start-up with access to decades of untouched data, direct access to thousands of customers and the ability to own work from beginning to end? To get the best talent, you need to know what makes you special and then find people that respond to that.

5. Make money

Every model produces great accuracy measurements and leading indicators of success. Some of those outputs are truly academic. Ultimately, your model worked if your company made money. So, design for measurability and hold your technologists accountable for bottom-line impact.

6. Beware of technical debt

I get it: Everyone wants you to go super-fast. And that’s fine, because it creates buzz and excitement around your work. But there will be a time when you are managing 25 use cases at a time and should maybe employ some best-practices, even if that means projects taking slightly longer.

7. Push the envelope

It is tempting to solve easy-to-define problems at the beginning of your journey. At some point though, you need to also work on those “holy cows,” those “we have always done it this way” topics. This will require subject matter expertise, so spend time with the business units to truly understand the topic and think about analytical ways to solve them piece by piece.

8. … and push back (a lot of pushing, I know)

“Oh, can you just pull me this data set…?” With data access comes power and you will encounter tons of colleagues that want you to help them out. It’s fine once or twice, but if little one-offs take up more than 5% of your time you have a problem. Make people understand that the analytics team should use their core competency — which is advanced analytics, not business intelligence.

9. Get people fired up so they come to you

Evangelize whenever you can and talk to people about the “art of the possible.” You want folks to compete for your resources and come to you with their problems. It is a sign that you have truly arrived in your organization.

10. Fail once in a while

A lot of companies are not used to admitting failure. We realized that it’s a sign of humility and actually shows two amazing things: you realize when things do not work (which means you can be trusted when they DO work) and you are setting ambitious goals.

Please let us know in the comments which lessons resonated the most with you — and what we may have missed!