AI in the Lab and the Plant

“AI in a context not related to food can work independently, be self-trained. But when it comes to food, the nuances, the emotions, the palate, the culture are so very intertwined that AI changes from being artificial intelligence to augmented intelligence – human intelligence plus artificial intelligence,” he says.

The current fascination with AI is probably no different than what must have been some awestruck 1970s reports at what computers someday could do. Or a wide-eyed 1990s look at this newfangled World Wide Web. But alone those couldn’t create Joe Biden’s voice suggesting you not vote for him or naked photos of Taylor Swift or Alexa or Siri ordering products indiscriminately.

So, like certain trees growing in the Garden of Eden, choose the appropriate fruits and avoid the bad apples. Investigate AI with a goal to implement it in a responsible, smart and effective way. And rather than depending on the Tree of Knowledge, you better have enough wisdom of your own beforehand to get a useful result.

Following are just a couple snippets of how AI already is helping food & beverage processors.

Hastening product development

Product developers should be developing products, creating and testing formulations, not doing internet searches on available ingredients. Leave the latter to AI. AI helps accelerate the final outcome.

Before co-founding BCD iLabs, Indraganti was global vice president of digital transformation and strategy at Ingredion Inc., focusing on both R&D and manufacturing transformation. He gives a hypothetical example:

“I want to create a protein bar with 15g of protein, less than 10g of added sugar, it’s got to be non-GMO and it will be sold in these countries. And don’t forget cost.”

Instead of a product developer doing the numerous web searches to come up with options for that formula, an AI program can do the grunt work. The R&D people can take the search results and immediately start experimenting.

“The R&D person can choose among the protein sources suggested, the sugars or sugar substitutes and other parameters and start doing experiments,” he says. “In the first iteration you’ll get an outcome, good or bad. You give the program a more desired outcome – maybe a certain kind of mouthfeel, even a desired emotional response – and it will fine-tune the suggestions.”

Like the R&D process itself, the AI process will be hit or miss at first, but eventually the hits will become apparent. As the program becomes more familiar with your company’s products, ingredients used and preferences, it will become better at suggesting formulations.

“It’s like having a new employee, a trainee in the lab,” Indraganti continues. “In the beginning you have to train him or her, and that can take a lot of time. Sometimes you wonder if it’s worth the effort. But at some point the AI system will have learned so much that you see an exponential reduction in your effort and increase in its effectiveness.”

Alexia Ciarfella was a senior scientist at Mondelez International. Now, after a brief stint at Capital One, she’s in the process of co-founding an AI food product development start-up called FlavorMind.

While at Mondelez, she helped develop an artificial intelligence tool to optimize the recipe development process of a well-known cookie. Her team’s work on this generative AI tool led Mondelez to win the “Best of Business AI” award from Microsoft in 2021, as one of the top 10 global organizations expertly using AI as a catalyst for business innovation.

“My role involved leveraging my extensive knowledge of technical challenges and business needs in product development to identify key opportunities to apply AI solutions,” she says. “Our data scientist then used these insights to create computational models tailored to these specific challenges, while our sensory scientist provided the essential, structured data required to train and refine these models.

“This close collaboration ensured that our AI initiatives were well-aligned with the company's business objectives, technically sound and supported by accurate data, leading to meaningful results and practical applications within the company,” she relates.

She offers some points – not just for product developers – for considering AI for your next project:
• How do you formulate your problem in a way that artificial intelligence can solve it?
• Is your problem specific and measurable?
• Can you describe your problem quantitatively?
• What do you currently measure?
• What else can you measure?
• Are there any boundaries or constraints you need to set?
• How will you measure success?
• Do you need AI to solve your problem?
• Can AI solve your problem?

Ultimately, she suggests three steps along an AI spectrum:
1. Leverage off-the-shelf AI tools from a vendor.
2. Partner with a vendor to co-create a customized AI solution.
3. Build and maintain in-house, bespoke AI models.

What do consumers want?

Perhaps the most often mentioned use of AI is in assessing changing consumer demands and suggesting a starting point to respond to them with product development.

“AI can dramatically decrease product innovation cycles so PepsiCo teams can respond to consumer demand in shortened timelines,” says a web posting from Athina Kanioura, a PepsiCo executive vice president and chief strategy and transformation officer.

“Insights revealed people were discussing, searching for and ordering seaweed products online thanks to an AI tool that analyzed millions of social posts, recipes and menus,” she continues. “That led R&D to develop Off The Eaten Path seaweed snacks in less than 12 months. Similarly, AI insights showed that consumers were interested in immunity. Six months later, Propel with immunity-boosting ingredients was on store shelves, ready to sip.”

“When generative AI is linked to consumer insights – and is tailored to the F&B industry specifically – it has the ability to impact organizational operations in a more meaningful way,” says the founder of one enterprise insights platform.

Tastewise is an Israeli firm whose AI tools provide that link between consumer insights and product development. The company touts “real-time marketing insights and consumer intelligence.”

The Tastewise platform scours a number of sources that indicate what people are currently eating – consumers’ in- and out-of-home purchasing, restaurant menus, even recipes and discussions about favorite dishes on Instagram and TikTok. “Traditionally, by the time you get some of those insights and start to act on them, consumers have moved on,” says Lee Brymer, marketing communications manager at Tastewise.

“What are the most popular soups with Millennials?” he proposes. “We can get you the answer in seconds.”

On its website, Tastewise has case histories with titles like “How Campbell’s redesigned their innovation processes with an efficient, data-driven approach” and “How Treehouse used Tastewise to put the spotlight back on their pickles sales.” Both of those projects were driven by consumer data picked up by Tastewise’s AI program.

It’s already in the plant

When it comes to the plant operations side of the food & beverage industry, AI might be considered a shapeshifter. Implementation at this point in time seems based predominantly on a processor’s comfort level with being a trailblazer of new concepts or waiting to see success in others’ efforts.

In some plants, AI already has begun to make a mark, refining some processes and data analyses in certain applications. For others, AI remains a technology with untapped potential to revolutionize the way the plants operate — once they get around to implementing it.

Count the poultry processing industry as one group embracing AI on the operations side, says Juan DeVillena, senior vice president of quality assurance and food safety at Wayne-Sanderson Farms. From hatcheries to packaging lines, he says the industry stepped up its implementation of AI over the past several years after a slow start.

“As chicken parts move down the line to be cut and portioned, every piece is simultaneously [image] captured from multiple perspectives and developed into a 3D image, enabling the computer to direct the DSI [portioning system] cuts to get the best yields, meeting dimensions criteria,” DeVillena explains.

Similar technology is used in beef-processing applications, says Brett Erickson, director of prepared and packaged solutions for Certified Angus Beef, which works with numerous plants to process its beef products.

“The machinery literally looks at each piece of meat and determines how to slice it to the most efficient level,” he says. “AI also can determine how to package the sliced product best and store the information, learning as it goes how to most efficiently produce that product and get it in the package in its final state.”

Potential remains a big draw for processors considering AI implementation. Konrad Ahlin, senior research engineer at the Georgia Tech Research Institute, says the technology reaches across so many areas that it can be difficult to pinpoint specific impact areas, but industry continues to work to find the lowest-hanging fruit for successful implementation.

“Every bird is different, and AI gives us a tool to address these differences in our process rather than attempting to plan around them,” Ahlin explains. “We will no longer have to automate for an ‘average’ and accept the losses that come from assuming that every product is the same size and shape.”

Food safety is another area that has felt an impact, DeVillena relays. AI and hyperspectral imaging scans give processors increased confidence that product does not contain foreign material, he says, when compared to traditional hand/visual inspections and older X-ray and metal detector technologies.

Hill’s Pet Nutrition, a Colgate-Palmolive division, is taking advantage of AI as a food safety tool. In opening a new pet food processing plant in Tonganoxie, Kan., Hill's specifically called out how the technology will play a role. Along with a serious dose of automation and robotics, the facility will use AI to drive its digital food safety vigilance system.

“Hill’s will utilize technology to work alongside Hill’s staff and a new state-of-the-art Mission Control Center to provide unprecedented visibility and monitoring through every aspect of pet food-making, from ingredient intake to final packaging,” Chad Sharp, director of manufacturing for the Tonganoxie Plant, said when the plant opened in October 2023.

Of course, there can be hurdles to jump and challenges to solve, says Geoff Coltman, vice president of Catena Solutions. One stumbling block for some early adopters of AI has been that initial investments didn’t always take the human element into account.

“There are companies that have spent millions — $20 million, $50 million, $250 million — on these automated processes, machines and technologies, but they never taught their people how to use it,” he says. “They bought a Corvette, but nobody knows how to fix the Corvette; so they’re going to run it until it runs out of gas, and then what?”

Coltman says that train of thought is why there’s a big focus to get food & beverage maintenance and operations workers caught up on these systems to ensure they succeed. As such, he says, “the ROI has yet to be realized in these investments; more investment needs to happen in order to get there” for processors.

In addition, some companies continue to believe the fantasy that AI will eliminate the need for human intervention, particularly on the data analysis side. AI finds the exceptions in the data and attempts to fix them, but if it doesn’t learn the proper solution through human intervention, the system will continue to kick out exceptions.

“AI, to this point, is not replacing people,” Coltman says. “AI is just a different way to work with the machines — but the machine is only learning based on what is being put into it. Humans are still there in order to make the decisions.”

He adds that change management, as well as education, development and the elevation of employees — or finding the right talent that understands the technology — need to be in the crosshairs during AI implementation.

DeVillena agrees that processors looking to AI to solve challenges cannot forget the people portion of the equation. “As an industry, ensuring the efficiency and safety of evolving processing technology is key, along with investing in the development of skilled individuals to operate and maintain these systems,” he concludes.

There have been a few warnings citing that old adage that “the leading edge is the bleeding edge.” Diving into a new market or technology early can be risky. But most observers think AI’s time has come, and most companies should at least start investigating.

“I think we’re just scratching the surface,” PepsiCo’s Bagavathi says. “We’re going to keep seeing more ways AI will simplify our lives and make us more productive.”

As always, the human element is a critical piece of the puzzle. An AI system is only as good as the information fed into it … by humans. And all the information that program churns out is useless if people are unwilling to learn alongside the technology. Companies need a plan that addresses how teams will use AI long term.