Heavy.AI CEO: Use AI to solve business problems, but not all problems

Artificial intelligence is best used when it is targeted to where it will provide the most benefit

AI use in the supply chain

Use of AI in the e-commerce supply chain needs to be targeted or the issue of too much data will overwhelm businesses. (Photo: Jim Allen/FreightWaves)

Rising inflation and booming consumer demand have CFOs feeling good about their companies’ financial health, but there is a toll that is being extracted. A September 2021 Deloitte survey found that over 60% of CFOs expected that year’s sales to be reduced because of the supply chain disruptions that occurred.

And it has only gotten worse since then. That same survey indicated that 40% of CFOs said supply chain shortages or delays had raised costs by 5% or more.

As a result, more companies are turning to artificial intelligence. An April 2021 McKinsey & Company report found that early adopters of AI saw a 15% decline in logistics costs, improved inventory levels 35% of the time, and improved service levels 65% of the time.

But not all AI is equal, and to Jon Kondo, CEO of HEAVY.AI (formerly OmniSci), that is a key part of implementing the technology.


“We have to have better visibility and better analytics so I can find that one thing that will be an issue,” Kondo told Modern Shipper. The old way was to aggregate data and create an exception, but “you can’t do that because when one thing matters, everything matters.”

An example of that has been the semiconductor shortage that has driven up the price of new and used cars. Kondo said if a manufacturer is looking at most of the automotive supply chain data, but failed to account for the semiconductor, assembly lines shut down. And that is what happened.

“You need to be able to have flexibility, have speed of thought and interactivity with the data,” he said. “The old paradigm of how you process the data … there is so much data, how do you do this? So the old way was you would pick one out of 10 items.”


Watch: AI in the last mile


While AI can help in this area, pulling data from all 10 items, Kondo said it is important for humans to remain involved. He noted a situation in which the rule set may say never pay more than X for a particular component, but if that item is the semiconductor that is preventing completion of that car, the company may be willing to go above that preset price. Human involvement with the data is able to override AI in these situations and ensure a good outcome.


“What has to happen is you have to bring the business problem together with the technical problem,” Kondo said. “Where companies have failed in the past is they bring together all this data but they fail to look at the business problem.”

Retail, and especially e-commerce, supply chains have become ever more complicated. It’s common for a single item to be transported across multiple modes of transport — ocean, air and truck — and visit several distribution centers even before it eventually ends up on a van to reach the end consumer.

Companies may feel the need to implement an AI solution that analyzes all these touch points, but Kondo noted that companies don’t necessarily need overarching AI solutions that solve all these problems. A focused approach is often more valuable to the organization.

“I think you have to think about the particular thing you are trying to solve because that is what you know will have the biggest impact on your business,” Kondo said. “If I can fill 10% more orders on a timely basis, that will optimize our business rather than going out and collecting all the data in the supply chain.”

Trying to do too much at once may be part of the reason so many respondents in the McKinsey survey found implementation of AI was often too slow and too expensive. The company found that more than 60% of AI projects were delivered late or over budget — or both. In addition, fewer than 33% of companies had conducted a “value diagnostics” review to identify how AI would best help the organization, and even more failed to conduct a vendor and solution review to ensure the solution would meet the organization’s needs.

“As a first step, companies need to identify and prioritize all pockets of value creation across all functions, from procurement and manufacturing to logistics and, ultimately, commercial. Less than one-third of companies perform an independent diagnostic at the outset, but this exercise can ensure companies have an accurate list of all the value-creation opportunities,” McKinsey wrote.

Kondo also pointed out that doing too much can overwhelm systems, including AI systems.

“That’s a really hard problem to do from a data perspective, but we are getting there,” he said, noting that the proliferation of sensors and other data points throughout the supply chain is generating more data than ever before.


“Think about if I could see this or get an answer to this, then I can make a better decision for my business,” Kondo said. “Then work backwards and ask what are the components that help me make a decision. I can almost guarantee that there is data being collected on all of these components.”

 Click for more articles by Brian Straight.

You may also like:

Drones are flying into weather data deserts. Can they be stopped?

Navigating COVID-19 shipping chaos: Finding capacity and servicing the customer

Need a warehouse? You may have to wait 9 months

Exit mobile version