Will AI inspire a new M&M? How artificial intelligence is reshaping Mars

Could AI text-to-image generators like DALL-E inspire new designs for iconic candies like M&Ms or Skittles? 

As a candy-packed Halloween approaches, it seemed like an obvious question to ask the head of AI and machine learning at Mars Inc. — a company which over the past century has overseen a slew of popular confectionery brands from M&Ms to Milky Way and Snickers; grown into a CPG behemoth that includes brands such as Dove, Pedigree and Whiskas; and now claims to care for half the world’s pets through nutrition, health and services businesses including Banfield Pet Hospitals and Anicura. 

While Shubham Mehrish, global vice president of digital strategy at Mars Inc., wouldn’t say whether an AI-designed M&M was on the horizon, he did sound bullish on DALL-E and other AI art tools for idea generation at Mars. 

DALL-E will augment creativity

“The DALL-E team has been stingy in giving access, but we have a few of our AI scientists already playing with it,” he told VentureBeat. “I think DALL-E is not going to replace the creative endeavor, it’s going to augment it — we are going to use DALL-E for inspiration and we’ve started to at least play with its capabilities within Mars.” 

Mehrish, who oversees Mars’ global digital, data and analytics teams, including in AI and machine learning, says efforts with DALL-E are just a tiny part of the massive AI-focused digital transformation journey Mars has been on over the past five years. 

“We are about midway through, so we are still on that journey,” said Mehrish, who joined Mars in 2018 after a two-decade career in banking, technology, consulting, strategy and data science. The effort behind both AI and digital transformation at Mars, he explained, came from from outgoing CEO Grant Reid with a focus on agile development, speed and scale. 

“It’s a passion project that our new CEO, Poul Weihrauch, is going to continue and build on, so it’s really a top-down mandate and it’s always been a very high priority for the company,” Mehrish said.

At this point, it’s not a question about when Mars is using AI, he added: “It’s a question of finding a process where we are not using AI and machine learning today.”

Artificial intelligence boosts pet care

One example of Mars Inc.’s strong AI focus is in the company’s pet business.

“We are the largest veterinary hospital chain in the U.S. now and the largest pet care company in the world,” said Mehrish, who explained that Mars uses AI in its pet food business to forecast commodity prices, track inflation, and optimize pricing decisions, promotions and assortments. 

In 2016, Mars spent $117 million to acquire Whistle, a smart collar startup known as a “Fitbit for dogs.” This past May, Whistle launched Whistle Health, an AI-enabled, data-driven smart device for dogs aimed at preventive care, which a press release said can “translate every dog’s movements into a personalized, holistic wellness index,” including health behaviors like eating, drinking, scratching, licking, sleeping and fitness. 

“These are built on huge training sets with machine learning and AI built in that gives you a score that allows you to track your dog’s health and also ties in with a vet,” said Mehrish. 

A sweet use of AI for products and supply chain

In Mars’ confectionery business, AI is combined with sensors on the manufacturing line to boost speed and accuracy while lowering costs. In 2021, it announced a digital twin initiative with Microsoft to develop virtual clones of its physical supply chain operations to help simulate scenarios that would be too difficult to test with physical assets.

“We can do preventative maintenance now using AI methods to see where problems are occurring on the line so that we can improve capacity and slowly automate certain components of that process,” he explained, adding that AI is even used to determine candy defects. 

“Imagine an M&M going through on a conveyor belt where we can see shape deformities through image recognition and augment human workers who are doing quality assurance to look at M&Ms and decide whether they are good enough to be packed or not,” Mehrish said.  

In addition, the company also uses AI in it’s upstream supply chain. “We can look at weather patterns and see how our crops are doing, actually predicting cases of fungus that impacts our raw material in India and Africa,” he said. The Mars Advanced Research Institute (MARI) and the University of Tennessee’s National Institute for Computational Sciences recently created the Mars Advanced Research Virtual Environment Lab (MARVEL) to carry out next-generation data analytics and better understand the science behind Mars products and services. 

For example, Aflatoxin is a poisonous but little-known natural product made by certain fungi. Specialists from the University of Cambridge and Mars Digital Technologies are using MARVEL to predict the likelihood of aflatoxin in maize, by analyzing data on things such as humidity levels, temperature, and rainfall, all of which affect whether aflatoxin can grow. 

MARI also recently announced a multi-year agreement with AI company PIPA to speed the discovery of new plant-based ingredients. PIPA’s AI platform LEAP combines AI, knowledge graphs and bioinformatics, to highlight links between food, compounds, microbes, and health states. 

Evolving data science teams at Mars

Mehrish detailed the evolution of Mars’ AI teams — saying the company realized quickly that multiple small pockets of excellence would not work. Instead, it needed a centrally driven route for the first wave of transformation. 

“We hired data scientists and data engineers at the center of the company – the center of marketing, the corporate center,” he said.

Over the last two years, the company has moved to what it calls a ‘federated center of growth,’ with business segments adding their own data scientists and engineers. 

“That’s the second wave, as we go down more and more into the organization and upskill our associates from a bottom-up perspective,” Mehrish said. 

This move will continue over time, he added, moving to local markets with their own AI teams with central oversight and governance around best practices, knowledge, sharing and artifacts. 

Increasingly, Mars has also reconfigured teams into products and platforms.

“For example, I oversee the entire data lake platform for the company and the infrastructure – we decide on the tooling that the businesses will run on, decide on the governance and then we let them work within that framework incorporating privacy and security,” he said. “And then each business has product teams focused on certain use cases, like strategic revenue management.” 

Mars goes from AI laggard to leader

Mars was a bit of a laggard when it comes to AI in the consumer packaged goods (CPG) space, Mehrish admits. Now, the company is a CPG leader, according to its external benchmarking. 

“But my intention is not to compare us with CPG companies, but to the Googles and the Amazons of the world,” he said. “That’s really the inspiration I’m trying to drive within our teams.” 

These days, Mehrish, who comes from a financial services background rather than CPG, finds consumer packaged goods a satisfying industry to be in — and not just candy-wise. 

“You deal with products that consumers touch and feel and interact with every day, and consume in our case,” he said. “So the fact that people’s eyes light up when you say where you work, it’s just a different level of satisfaction.”