AI & ML Fintech

Why Brilliant Data Science and AI Ideas Fail (Beaumont Vance Head, posted on

So you have created a machine learning algorithm that has uncovered some very powerful insights. It has been a year and it is still not in production. Why doesn’t anyone get this??? Why do 80% of AI and data science ideas wither on the vine?

So goes the lament of the data scientist, the innovator, and the inventor. 

The problem is not the rightness of the formula, nor the brilliance of the insight. It is that the inventor has only solved about 1% of the problem, but thinks they have solved it all. 

Businesses are systems with other systems within systems. There are multiple organizations, data systems, web systems, IT systems, Q&A, production, DevOps, sales, marketing, legal, compliance……all operating within a political system at the top that operates within a broad economy with shareholders and vendors, investors, etc. 

An insight or invention can be brilliant, but it needs to integrate across those systems….and guess what, everyone operating them is already plenty busy and might not care to re-configure everything to accommodate your brilliant idea. In fact, as Clayton Christenson pointed out in The Innovators Dilemma, companies are made of stable systems that by definition and design resist perturbations (aka change)! 

If a company wishes to see any good idea become a reality, it must plan beyond the exploration phase and have a plan for the entire path to production including funding resources, people, roles & responsibilities, project management, system integration, change management, communication, staffing, DevOps, Q&A, support… other words, all of the elements that support all the existing, accepted products and platforms.

The problem is that inventing is fun. Everyone loves the rush of discovering breakthroughs and ideating. Spending a year on operational challenges, fighting pushback and engaging in change management though….not quite as much fun, and with fewer volunteers.

A wise COO once told me, “If you published all of these top secret discoveries we make on a billboard on the interstate, no one would ever steal them.” At the time I thought he was crazy; but now that I have done the work to build AI platforms, I see that he was right.

There is no shortcut. Building AI and data science products that operate at scale in production is hard work and requires investment beyond the smart people in the lab. Its really not a mystery. Many companies simply fail to staff and fund for all the work required to actually deliver the solutions.

I often think of Henry Ford. He did not invent anything new in terms of the car. what he invented was just the method of putting new technology into production. He invented the delivery process for invention.

Data Science and AI are essentially at the same spot as the car was in 1907. before we can unleash the immense power of this new industrial revolution, we need to move away from what Tom Davenport calls “artisanal analytics” by doing what Ford did. Its about operations.

Beaumont Vance is the Head of AI, Advanced Analytics and Emerging Technology DevOps at TD Ameritrade

Leave a Reply