CEO Perspective: Your Enterprise and the Future of AI
Arijit Sengupta, founder and CEO of Aible, explains how AI is changing and why a single AI model is no longer smart business.
- By James E. Powell
- April 14, 2020
There’s lots of buzz about artificial intelligence, but as Arijit Sengupta, founder and CEO of Aible, points out, “Everyone has heard a lot about AI, but the AI we’ve been hearing about is not the AI that delivers business impact.” Where is AI headed? Why is a single AI model no longer the right approach? How can your enterprise make the most of this technology?
Upside: What technology or methodology must be part of an enterprise’s data or analytics strategy if it wants to be competitive today? Why?
Arijit Sengupta: AI needs to deliver context-specific recommendations at the moment a business user is making a decision. We’ve moved away from traditional analytics and BI, which looks backwards, to a forward-looking technology. That’s a fundamental shift.
What one emerging technology are you most excited about and think has the greatest potential? What’s so special about this technology?
Context-specific AI has the greatest potential to change business for the better. The first generation of AI was completely divorced from the context of the business. It didn’t take into account the unique cost-benefit tradeoffs and capacity constraints of an enterprise. Traditional AI assumed that all costs and benefits were equal, but in business, the benefit of a correct prediction is almost never equal to the cost of a wrong prediction.
For example, what if the benefit of winning a deal is 100 times the cost of unnecessarily pursuing a deal? You might be willing to pursue and lose 99 deals for a single win. An AI that only finds 1 win in 100 tries would be very inaccurate based on model metrics, although it would boost your net revenue. That’s what you want from AI.
The second generation of AI has a laser focus on the specific business reality of a company. As Forrester and other analysts have pointed out, AI that focuses on data science metrics such as model accuracy often doesn’t deliver business impact.
What is the single biggest challenge enterprises face today? How do most enterprises respond (and is it working)?
Solving the last-mile problem of AI is the single biggest business challenge facing companies today. Right now, most business managers don’t have a way to understand how a predictive model would impact their business. That’s a fundamentally different question than finding out what the AI has learned.
Just because I tell you how a car works doesn’t mean you know how to drive a car. In fact, in order to drive a car, you often don’t need to know all of the details about how a car works. In the first generation of AI, we obsessed over explaining how the car works in great detail. That’s what was considered “explainable AI.”
What we are shifting to now is the ability for businesses to understand how the car affects their lives. Enterprises need to know how the AI affects their business outcomes under different business scenarios. Without this knowledge, you can’t get AI adopted because you’re asking business owners to play Russian roulette. You’re not giving them the information they need to understand how a given AI model will affect their KPI. You’re just giving them a few models and telling them to hope for the best.
Is there a new technology in data or analytics that is creating more challenges than most people realize? How should enterprises adjust their approach to it?
Traditional AI built on model accuracy can actually be incredibly harmful to a business. AI that’s trained to optimize model accuracy is often very conservative, and that can put a business on a death spiral. A conservative model will tell you to go after fewer and fewer customers so you’re assured of closing almost every deal you pursue, but many times that means you end up leaving a lot of money on the table and slowly destroying your business. AI that maximizes accuracy at the expense of business impact is worse than useless -- it destroys value.
What initiative is your organization spending the most time/resources on today? In other words, what internal project(s) is your enterprise focused on so that your company (not your customers) benefit from your own data or business analytics?
We’re an early-stage startup with a relatively small volume of data, but we believe in getting started with AI quickly rather than waiting to get a ton of data. What we first started doing is using AI to predict which customers were likely to go from a first contact to a first meeting and which were likely to click on an email.
Over time, we’ve collected more data and been able to optimize our marketing spending across different channels and figure out exactly which customers to focus on. If we had waited until we had a lot of data to get started, we wouldn’t have progressed as far as we have. By getting started with AI quickly, we were able to improve our AI process much faster.
Where do you see analytics and data management headed in 2020 and beyond? What’s just over the horizon that we haven’t heard much about yet?
Everyone has heard a lot about AI, but the AI we’ve been hearing about is not the AI that delivers business impact. The AI we’ve been hearing about is the AI of labs that’s abstracted from business realities.
What’s just over the horizon that people are beginning to wake up to is that to get business impact, you have to have a very different kind of AI. Creating a single AI model doesn’t make any sense because business realities constantly change. What you need to do is create a portfolio of AI models that are tuned to different business realities. You need a different model if your cost to pursue a customer goes up 10 percent or if your average deal size goes up 20 percent. If you create a portfolio of AI models, your business will be much more resilient to change -- and the only thing you can count on in business is change.
Describe your solution and the problem it solves for enterprises.
Aible’s AI platform ensures business adoption by giving users tools tailored to their existing skills and needs. Aible overcomes the last-mile problem by enabling end users to customize models and see how they affect the business. Aible lets you get started quickly with the data you have by fully automating the machine learning process; team members can contribute their unique business insights to AI projects. Uniquely, Aible delivers dynamically balanced AI models so you always deploy the right model at the right time. Aible ensures data security by running in your secure AWS or Azure account or on premises and never sees your data or trained models.
James E. Powell is the editorial director of TDWI, including research reports, the Business Intelligence Journal, and Upside newsletter. You can contact him
via email here.