I eagerly devoured the HBR article – 8 Ways Machine Learning is Improving Companies’ Work Processes, when it appeared in their newsletter.  While the article includes some really good points, I disagree completely with the recommendation of creating AI Centers of Excellence. To their credit, the authors do caution against islands of innovation that are disconnected from each other.

The bigger issue I have is that machine learning is too critical to be sequestered within an innovation lab!  Frank Chen provides the best analogy of how machine learning will be embedded in products.  According to Chen, machine learning is analogous to relational databases – which have been core to all current digital products and experiences – from e-commerce to cat videos.  Relational databases are superb at storing and sorting data.

Machine learning applications will similarly be foundational to the experiences of the future.  The difference is that with ML, the unit of measure is prediction.  Going back to the article, an AI COE would be as ridiculous as a Relational Database COE.

Instead, I suggest an ML First approach.  So – how should this work in practice?

Types of Innovation

There are basically three types of innovation:

1. Incremental Innovation – This is the Kaizen method of steady tweaking and improvement.
2. Disruptive incrementalism – This is innovation in the interstitial spaces that results in “hockey stick” results. Think – launch of the Apple’s App Store.
3. Radical Innovation – This is extremely rare in business and almost never seen in nature. Think – creation of Uber and the rise of the sharing economy.

For an excellent analysis of these types of innovation, watch this Google Talk by Google’s ex-CIO, Doug Merrill.  It’s fairly dated, but still very relevant.

While Machine Learning will create some radical innovation, I think the more interesting opportunity is in creating incremental changes that have disruptive effects.

How do you go about identifying opportunities for disruptive incrementalism?  A great first start is to begin with your customer (external or internal) and follow the following steps:

• Map out the customer’s journey
• Identify white spaces in the journey, where customers are peeling off or not being served in some critical way
• Are you collecting valid data? For example, are you collecting the customer’s demographic data, current expressed intent signals, value of those signals to the company etc
• Is there a way to build a model to deliver a unique message that raises the chances of customer conversion?

All the above steps are incremental changes, but collectively (and at scale), they can deliver disproportionately big results. That, to me, is the larger opportunity of Machine Learning – not sequestered in a COE somewhere, but on the front-lines of delighting your customers.