AI is Shaping the Future of Snack Production

“It is easier for snack processors to use AI today in areas where they couldn't do it 12 months ago,” says Eamonn Cullen, our global category manager. “Obviously snack producers are not data scientists, they're not machine learning experts. So what's really changed? It’s the availability of pre-trained AI services, easy access to foundation models, increase in cloud-based storage capabilities and data silos which ensure data is isolated and secured."

Today, big cloud providers such as AWS, GoogleCloud, Microsoft Azure, Snowflake and others will not only host mind-boggling volumes of data securely in a data lake or warehouse, but they also provide adaptable models and tools to help make that data work for their customers.

“People like AWS have developed what are called foundation models [Large-scale machine learning model, which is trained on vast amounts of data and designed to perform a wide range of tasks]. And they have customer support teams who can help assess the snack processor data so it can be input into those foundation models and used as a basis to develop a unique model for the processor.

That's really important, because the processors don't have the bandwidth or the expertise to develop their own models from scratch. So, by using services which are now available from those big-name companies to leverage the AI models already developed, snack processors can now easily implement AI solutions into their business without having to recruit huge teams of data scientists.”

“I think that's a real change that we're seeing in the past 12 months and it's going to evolve even further over the next three to four years,” says Cullen. “I’m really excited about it.”

So what might this mean for snack producers?

In terms of production facilities, AI and machine learning have already carved out one of the strongest cases for investment so far by enabling predictive maintenance. By applying AI to plant data from production machinery, companies can better predict and prevent machine failure, reducing downtime and optimising maintenance resources.

However, a broader application of AI can enable a whole new level of process and all-round business optimisation, according to Cullen: “When you talk about AI, the whole focus is on asking, what does your customer need and when do they need it? And also what is the food quality specification you're trying to achieve and how can we maximise the utilisation of every piece of raw material entering the food production process to achieve that specification?”

Take the example of a private label snack maker looking to create products with different specifications for different brands or customers. “Typically, they’ll have some information about raw material coming in today and they already have an expectation of what that’s going to mean in terms of the finished goods going out,” he says. “But with AI, what processors are now trying to do is to optimise the workflows and to map the entire process, based on large volumes of data, AI predictions of raw material quality and customer buying behavour.”

The EU’s recent decision to withdraw approval for the most widely used smoke flavours for safety reasons provides another illustration: “So imagine smoke flavoured chips are out, but you have historical data on who was buying those - the market data, the customer buyer data and the macro economic data from the actual target markets - and you can bring that into your capacity planning. For example, a UK snack processor might hypothetically want to produce less prawn cocktail versus cheese and onion. Or maybe the European football tournament means you need more larger packaged chips in Germany. That data is actually now a critical part of the production process flow,” explains Cullen.

One of the issues that threatens to slow progress is concern that AI will take human expertise out of the equation. But Cullen believes that worries around job security are misplaced: “Process experts should not be afraid that their jobs will be gone. They should understand that now you have more data and predictive AI models, you've got more decisions to make. So the jobs of process experts are still there but their roles are evolving.”

Head of R&D Judith Lammer, believes that equipment suppliers such as INSORT are going to be vital to manufacturers as sources of data. “We're collecting massive amounts of data. In the past snack processors lacked the bandwidth to analyse all the data collected and make sense out of it. Now, automation can assess that data, alerting them to trends, clusters and connections that they may never have considered.”

While suppliers of many kinds of production equipment are gearing up to offer more data and process insights, sorting technology provides an especially good illustration of why AI is set to be such a gamechanger.

“With our eyes, we can see only certain colours, but imagine if you could see an additional 240 colours in the infrared range simultaneously. And now, consider what you are seeing is chemical information, not just colours. This completely changes what is possible for sorting food, allowing us to detect properties like bitterness or rancidity”, explains Lammer.

What really excites me is that it is not just possible to see the invisible, but see it in milliseconds,” says Lammer. “We now have the capability to identify tiny defects such as small spot defects and insect bites. When you combine that with the capacity to analyse the data properly, you streamline the process into a simple sorting program and target specification. INSORT customers are already leveraging this data to develop their own AI models”, she says.

It will enable processors to be much more dynamic about sourcing and how they  pay suppliers, for example, as quality varies subtly over time.