Knowing AI vs Applying AI

Why everyone talks about AI, but only few really apply it

By now, almost all companies see AI and advanced analytics (AA) as transformational technologies. McKinsey estimates a annual potential between $9.5 and $15.4 trillion. Let that sink in. That’s somewhere between 250 and 450 times the net income of Google. AI represents over a third of this just by itself. 

Thus, the case for investing in AI and analytics is clear. And with such clarity, investments are being made. O’Reilly’s 2019 AI survey showed 79% of companies planning to spend at least 5% of their IT budget on AI. The number of positions for ‘Data Scientists’ on Indeed has risen by 55% between 2017 and 2019.

Yet, I see many companies still unable to capture these opportunities. Despite the small army of data scientists that works hard every day. Despite the investments in data quality and data platforms to make sure they have the right data. And, despite the executive sponsorship to drive AI and analytics forward. At the end of it all, many companies can show prototypes and perhaps limited cases in production. If they’re lucky the production cases might have recouped their original investment.

What is it, that makes the capture of value from data so hard for companies? 

The value of ‘applying’ data science over ‘knowing’ data science

One thing I see is that companies overvalue data science knowledge over the ability to apply these skills to business problems. 

Think for a second about computer science for an analogy to the data science world. Many companies hire CS graduates to engineer software. Their experience with computers makes them great at solving problems with computers. These roles are aptly named Software Engineers. With that title, expectations are clear from the outset: you are here to engineer software. From a company perspective, the best software engineers are productive, business savvy, and deliver measurable value. 

Now, consider the story for data science. Companies want to capture their share of the $250+ trillion opportunity. Someone told them they need Data Scientists to do so. They go out into the market and hire a set of people with a data science degree into a role named… ‘Data Scientists’. 

Quite likely, they now have a team of researchers with a specialty in data science. What they need, however, is a team that delivers a valuable and working AI/AA solutions. These two are not the same. The stereotypical data scientist values knowledge and intellectual depth. I’ve even had data scientists ask me to remove them from projects with CEO-level visibility because it wasn’t challenging enough from a science perspective.  The stereotypical AI Developer or Analytics Specialist – or however we want to call them – values how well he can apply his AI and data skills. He or she would delight in the opportunity to work on such a high-impact project.

At Intel […] it almost doesn’t matter what you know. It’s what you can do with whatever you know or can acquire and actually accomplish [that] tends to be valued here

– Andy Grove (Intel)

Andy Grove made the same distinction when comparing Intel to Fairchild. What you do and what you achieve is how you create value. Knowledge, and the ability to acquire more knowledge, are only useful to the extent it helps you deliver.

Corporate data science today needs more ‘do’ers’ and less ‘scientists’. We need more data scientists using WD-40 and duct tape to squeeze value from data. We need data scientists that focus on shipping data based solutions, instead of polishing them. In todays world, a 80% solution already deployed is worth more than the 99% solution stuck in the lab.

A focus on ‘doing’ sets expectations to business focused data science teams. Your success is no longer defined by academic credentials, publications, or the complexity of your solution. Instead, you live by the feedback of your users, the simplicity and elegance of your solution, and the value you deliver. It asks teams to take responsibility beyond modelling. Delivering a state-of-the-art model without a business case is still failure. 

And, the accompanying ‘bias to action’ limits your risk of unexpected failure after months or years of investment. It’s agile for data! Getting something up-and-running today means getting feedback tomorrow. It means discovering deployment, data and sponsorship woes before investing months on modelling.  

Conclusion

Business data science needs more ‘do’ers’ and less ‘scientists’. Companies must refocus on delivering measurable value, and delivering it quickly. As data scientists, value is your most important deliverable. Start shipping it. Start measuring it. Start tracking it.

Finally, if this sounds like you, come talk to me. We are hiring and we would love to have more do’ers in the team