Breaking Open the Black Box

Just finished a fantastic conversation with a top data scientist at IBM as preparation for a possible podcast. He made some extraordinary points that I want to share now:

1. Data science will be a cornerstone of life, but data scientists are having their 15 minutes right now. Human powered data science is already commoditizing, and you can see early indications of this in (gently) falling compensation levels. Automation, AI, block chain, quantum computing and other developments will continue to have the same effect on data science that they've had on the data center and other areas of company operations.

2. Companies have tried to centralize data management and data analytics. The results are clear that these approaches have not only not delivered value, they have destroyed value in many instances. Today, we are seeing companies decommissioning data lakes that they just spent millions to create and manage. Centralized data science teams are being split up and diffused into functional teams. And most importantly, democratized, fit-for-purpose analytics solutions are being adopted to accelerate insight and push better decision making power to the edge. 

3. Data analytics teams are seen by typical company workers (whose work is being analyzed) as human-powered black boxes. The result is that those typical employees have ultra-low trust in the data science team and very low utilization of the generated insights. There's a big lesson here for software companies too.

4. The culture of data science is based in science and strives for the ideals of precision and accuracy. In this, data science is not wrong, but it is at odds with the pragmatic, fit-for-purpose philosophy of business decision-making. The business definition of accuracy is "would more accuracy change the business decision?" In practical terms, the priority that data science places on precision-crafted answers means that the analytics are often irrelevant by the time they're delivered.

5. Current approaches to data science are not scaling effectively to meet demand because they are powered by humans operating with tools that require a lot of work to use. As we've seen in the IT service desk and other similar corporate services, the definition of a data scientist in the corporate environment going forward will be more and more sherpa and less and less service provider.

These observations square with what we continue to see at Proof, and I found them very compelling. In addition to the impact on in-house data science operations, the implications for third-party analytics services that rely on human data science teams using R and Python are profoundly disruptive.

The podcast should be good. We'll promote it when it's done and ready.

Mark Stouse is CEO of Proof Analytics, a powerful new business impact analytics platform that helps you and your organization prove and improve your business impact. For more information, visit www.proofanalytics.ai.