The Three Codes of Commercial AI Creation


A recent New York Times article dived deep into IBM’s creation of Watson – a variation of ‘AI for business’ known for its natural language processing abilities.

It’s a compelling read, one that explains the harmful effects of presenting any AI algorithm as a ‘one size fits all’ commercial solution, much like Watson’s deployment across healthcare, finance, law, and academia.

As a product manager by training, and CEO by title, this excerpt really resonated with me:

“The company’s top management…was dominated until recently by executives with backgrounds in services and sales, rather than technology product experts. Product people might have better understood that Watson had been custom-built for a quiz show, [it was] a powerful but limited technology.”

– Steve Lohr, NY Times (What Ever Happened to IBM’s Watson?)

It resonated because IBM’s approach to AI product development was the exact opposite of creation processes I led at Salesforce and Zone7, where comprehensive AI tools such as Einstein are effectively deployed to steer users toward action rather than the burden of more analysis.

Throughout this journey, I’ve seen and heard the inherent skepticism around AI and its true capabilities. This stems partly from predecessors who have naively overpromised, mislabeled, or underdelivered on AI products. As Zone7 engages sports organizations considering whether to adopt our platform, this is precisely why I explain in great detail what our underlying algorithm is, what it isn’t, and what it needs to provide the best outcomes.

Three core principles have been essential to our successful creation and commercialization of AI products. I believe that without following them, no amount of data or capital can optimize legitimate algorithms.

1) Identify – And Stay True To – A Specific Use Case

Repurposing AI is not as simple as repurposing cloud storage or mainframe computing. Each algorithm is designed for precise needs and applications. You cannot boil the ocean with ‘one AI for all’ thinking.

In the early days of development at Salesforce, we spent considerable time and effort identifying which ‘use cases’ to tackle first. Consistent with product management best practices, this meant finding the sweet spot between what users cared about most and what our AI experts could realistically build. Experimentation is crucial, mistakes are made along the way, but tracking metrics and results over time was key to our ‘fine tuning’ process.

At Zone7, we started targeting professional sports organizations because of (a) the high costs that injuries carry and (b) the significant amounts of untapped health and performance data collected each day. We identified professional soccer as our core market and that decision framed all aspects of our AI’s development, from data ingestion, to modelling risk and structuring feedback.

Over time, Zone7 has garnered interest from teams across rugby, basketball, American Football, cycling and other sports, each looking to unlock the value of their own datasets to better protect their players.

To lift Zone7’s AI as-is and place it in the context of a whole new sport would be a recipe for poor performance and disappointment. Instead, we recalibrated our algorithm to deliver new and alternative versions of the product, which are custom-built for each sport and the unique injury risk factors they present to those athletes.

Similarly, we do not envisage using Zone7’s AI in sport beyond the use case of preventing injuries. Our expansion, over time, will lead into new industries but only with the same fundamental objective of ensuring wellbeing. This is the purpose behind our product – to address health and burnout issues affecting high-performing individuals in their environments, whether that be surgeons working night shifts, or military personnel pushed to their limits.

2) Enlist Practitioners in The Algorithm Building Process

Effective AI cannot be built without extensive input from its future adopters.

When IBM technologists applied its Jeopardy! winning AI to the high-stakes world of oncology medicine – without extensive testing or engagement with healthcare professionals – it was no surprise that end users were left frustrated by “the complexity, messiness and gaps in genetic data” across cancer centers.

One of Salesforce’s key competitive advantages during my time there was the ample access our team had to the real-life operators our commercial AI was designed for. The ability to collaborate with, and secure swift feedback from, our target audience was invaluable. Fast feedback, fast adjustments, faster progress.

At Zone7, we created an algorithm to help coaches, physios and other sports leaders make better-informed decisions. To do this, we had to understand how these professionals make decisions as well as what methods, tools, and data analysis processes they use. Productivity-driven AI cannot be created in a vacuum, it requires a willing partner who brings a real-life lens. Working closely with early adopters in football/soccer – such as Getafe CF, Rangers FC, Hull City, and Champions Leagues clubs who wish to remain anonymous – has allowed us to “incubate” the algorithm. We now have similar methodologies and projects taking place in professional Rugby, cycling, and the NFL.

With this context baked into the algorithm we improve pattern recognition – using an athlete’s past performance and medical history to determine the best course of action for mitigating injury risk, creating personalized intervention plans, and optimizing output.

3) Keep Scientists In The Driver’s Seat

An AI is only as competent as the professionals driving its deployment. In IBM’s case, the commercialization of Watson was a process driven by ‘sales and service’ experts, rather than the combination of (a) data scientists with an intimate knowledge of its workings and weaknesses and (b) product operators who would incorporate AI advice into their daily routines.

AI product management demands vision and patience. It requires the ability to incubate an algorithm for a very specific problem, and permission to carefully refine it over time. Just because an algorithm can ‘learn’ and self-tune does not mean it should be taken out of its original context, or away from its data scientist creators.

By handing off AI to those without the requisite skills, IBM created a disconnect that ultimately hurt product performance. Now, with a new CEO at the helm – one with computer science credentials – IBM has self-corrected its positioning of Watson products by pivoting to a more basic and pared-down commercial strategy.

In contrast, Zone7’s executive team, including my Co-Founder Eyal Eliakim and our Performance Director Rich Buchanan, is well-versed in data science and its specific application to professional sports. We understand the nuances of customizing AI for different clients and risk tolerances, as well as the need to hone them over time. There are no internal silos impeding our product’s swift deployment and ongoing refinement. That is a conscious decision and the results, client case studies, and positive sentiment from performance managers speak for themselves.

By abiding by these three codes and combining them with a talented team, flat management structure, and extensive pool of valuable reference data, our AI is delivering effective injury risk forecasting models that translate to actionable guidance, athlete availability, and improved performance.