To get your AI initiatives off the ground, you don’t need to use the fanciest algorithms and machine learning techniques available. You don’t need to find that one, perfect use case. You don’t even need the cleanest and highest-quality datasets (at least, not to start). What you need to succeed with AI is, first and foremost, people.
Connect Data & Doers
Companies that are not able to connect data with doers — that is, get data in front of the people who are working every day in every part of the business to make things run, whether that’s physical high-value equipment, marketing campaigns, supply chains, the back office, or customer support — won’t survive. The future is one where AI is part of the fabric of an organization and, without enabling people, efforts will inevitably fall short.
“Attaining significant return on investment for discrete AI or a small collection of business area use cases is difficult because of the tactical focus of AI design and the accumulation of technical debt.”
— Gartner, What Is the True Return on AI Investment?, February 17, 2022
I hear you already: but what about investing in the right technology? What about having good processes in place for AI projects? Surely, you need technology, processes, and people to succeed. Or at the very least, shouldn’t the one thing you need be… data? Or even better: high-quality data?
Without You, It's Just Data
Sure, once you have people, you also need the right technology and processes. And of course you need (good) data for people to work with. But data is just that — it can’t, in and of itself:
Identify & Evaluate Risk
Remember when Amazon scrapped their recruiting tool because it was biased against women? Or when Google’s photo app was labeling people with darker skin as gorillas? Examples abound of PR crises caused by AI systems abound. Though it’s true tools for explainable and responsible AI are only becoming more sophisticated and commonplace (Dataiku, for example, provides reports on feature importance, partial dependence plots, subpopulation analysis, and individual prediction explanations), they will never completely replace the need to have humans in the loop. Because only humans can …
Reflect on Decisions
No technology, no matter how advanced, will ever be able to think critically about what your organization needs and make a decision about what has to be done. AI can be an essential decision-making partner for people, but it cannot — and should not — be definitively making decisions that affect critical business functions (at least not with human-in-the-loop checks in place). For example, you might build an AI system that can help customer service representatives recommend products or services, but only a person can decide when the best moment is to make that recommendation, or if that recommendation makes sense at all. For a more serious example, consider an AI system that grants or refuses loans — you still need people to ensure the system is working as expected and not putting your business in jeopardy.
Tell Stories
At the end of the day, it’s humans that design AI systems. People are behind every AI use case and every AI success story (or failure). It takes a soul behind the machine to dig into data and figure out how to use it in new and creative ways — no amount of data (yes, even really, really, high-quality data!) can do that by itself.
For example, Dataiku was working with a truck manufacturer that wanted to do a proof of concept (POC) on a relatively classic (but nonetheless challenging) use case: leveraging internet of things (IoT) sensors for advanced predictive maintenance. In working on the project, the team put together data from different teams across the organization that they had never blended before, including classic sources like truck movement but also less obvious data such as warranty information.
Ultimately, in putting these disparate data sources together, they found some oddities: in particular, that there were some trucks that were supposed to be out of service, but instead, they were traveling around. In investigating further with business teams, they uncovered cases of warranty fraud — in other words, people were sending parts for repair for trucks under warranty, but actually using those parts in other trucks not under warranty. Ultimately, the POC was a success because the teams uncovered a problem that was even more important to solve than their initial business need. It took the creativity of people putting their heads together and diving into the data to get there.
These are all intrinsically human qualities, and no amount of technology, no amount of investment in AI, can remove the need for humans to take on these critical tasks.