So you want to generate more value from AI. One approach is to hire more data scientists, but that doesn’t scale because they’re in high demand, difficult to attract, have low productivity the first six months as they learn your business, and leave in under three years. Another approach is to outsource it but then you end up with fragile bespoke apps, low agility, and little progress toward becoming a data-driven organization.
At Dataiku, it’s our belief that the future of work is one where the use of data, machine learning, and AI is so ubiquitous that it becomes a part of our everyday lives, in the same way that we turn on a light switch without thinking about the long history of invention and the vast technical infrastructure that enables the electricity to flow reliably and safely to the LEDs in the bulb. We believe that it’s the people who are the stars of the show — the ones who leverage this technology for productive and, sometimes, transformational purposes and democratize it throughout their organizations.
Hiring and outsourcing won’t meet your goals. The approach that we’ve seen work in many companies and industries is to upskill your analysts, provide them with a common self-service AI / machine learning (ML) platform (like Dataiku), and nurture those who perform best. That approach wasn’t practical just a few years ago because AI/ML platforms were difficult to learn, inefficient, and unforgiving of mistakes. That has changed. With the promise of a new future of work, organizations can embed the productive and valuable use of data throughout their business.
Spreadsheets increased the number of people doing financial modeling 100 fold. Windows interfaces increased computer users 100x. WYSIWYG editors increased typesetting 100x. Tableau increased business intelligence report developers 100x. No-code, low-code, AutoML, and interdisciplinary team collaboration platforms are doing the same for AI today. However, achieving 100x and having it stick is harder than just downloading the latest desktop tools and pointing analysts to YouTube lectures.
Best Practices for AI Upskilling
At a high level, best practices we’ve learned for large scale AI upskilling (as it relates to the future of work) are:
1. Evolve your AI organizational structure from Centralized Center of Excellence to Hub and Spoke to Center for Acceleration to fully embedded.
2. Boost productivity three- to five-fold from the onset by implementing a highly usable AI development and operations platform.
3. Fund an adoption program to get data scientists and business users onboard.
4. Fund an upskilling program to get frontline domain experts developing AI/ML products.
70% of companies understand how AI can generate value for them but only 10% capture it. The blocker isn’t computation. Tom Siebel compared cloud computing costs to the value of AI by saying that “computation and storage is basically free.” The cost of computing is a 16,000th of what it was 20 years ago, and the cost of training an image classifier is down 64% since 2018. The bottleneck is talent.
Andy Grove was a co-founder and CEO of Intel. The company’s market capitalization grew 50x under his leadership, from $4 billion to $197 billion. An east coast business reporter flew out to Santa Clara in the early days of the PC revolution to figure out what Intel did and was having a hard time with it. They made machines smaller than your fingernail out of sand? Frustrated, the reporter resorted to first principles and asked, “What’s the most important input to your product?” Andy thought for a moment and replied, “Smart 25-year olds.”
A few years later, another east coast reporter visited suburban Seattle to ask Bill Gates, co-founder and CEO of Microsoft, about his company’s book value. They didn’t have any factories, warehouses, trucks, or planes so what was it worth? Bill said that, “The inventory, the value of my company, walks out the door every evening.” Andy and Bill’s insights foreshadowed a trend that’s affecting AI adoption today: the battle for talent.