An AI project should start with a question, not an insight

Before launching an artificial intelligence project, Avis' former director of business intelligence and analytics advised that companies first make sure they have a strong analytics practice in place.

Ahead of his keynote talk at EGG NYC 2018, hosted by software company Dataiku, Ashok Kumar sat down with us to talk about why it's important to start an AI project with a question, rather than an insight.

He also weighed in on the difficulty of quantifying AI's impact at the car rental company and how he's seen the business intelligence and analytics landscape change in the last decade.

You're giving a talk on how a 'holistic, end-to-end' analytics process can be an accelerator for an AI project. Why?

Ashok Kumar: One of the things I would say is you need to ask the right questions. Sometimes, we don't do that, and then we blame the machine learning. So, ask the right question. Make sure the question makes sense and that there's data to support the question you want to answer.

What does the wrong question look like?

Ashok Kumar

Kumar: I had one executive tell me, 'Hey, listen, we've got a whole bunch of data here. Let me just give you all the data, and you tell me what you can tell me.' This was one of the things IBM Watson was promising. It doesn't work that way. We have to spend time coming up with the question we want to ask, why we want to ask it and what we're going to do with the insight.

One consultant I spoke to advises a problem-solution approach as a best practice for getting started with AI, but that service-centric companies tend to use a solution-problem approach -- and they're having more trouble proving the value of AI than asset-centric companies.

Kumar: That's an interesting observation. In our case, we have hard assets -- [500,000 or] 600,000 vehicles -- and 20 basis points in utilization adds up to a lot of money. So there, we could say, we've seen a half-percent utilization better because of AI or machine learning capabilities. You could easily quantify that.

In other areas, it's not so easy. Customer marketing is a good example. We've tried different ways of marketing using AI and machine learning, and it's hard to quantify. But you still need to do marketing, because the cost of not doing it could be disastrous.

Avis is battling new competitors thanks to digitization. Who is going to emerge as the winner?

Kumar: The winner is going to be the organization that can execute and pivot. That's going to be the key to being No. 1 -- to be ready to experiment and fail. And when we see opportunity, we execute.

You've been the director of BI and analytics at Avis for more than six years now. What has surprised you the most about the changes you've seen take place?

Kumar: The advances in cloud technologies and that a lot of these capabilities have become commodified, to some degree, make it easy to execute and learn from something without a lot upfront costs. So, we can't use the infrastructure as an excuse.

That [has surprised me] and that analytics has been accepted at a very high level. Now, we have top executives talking about analytics as an enabler, as opposed to a few years ago when we had to convince them. The key here to make sure we execute and that we're not just using big technology words.

What advice can you give to CIOs who are planning to launch an AI project?

Kumar: You need to get some successes to get more successes. In order for you to do that, you need to fully understand what are the major pain points for the business.