One Rung at a Time: How Food and Beverage Companies Actually Get AI Into Production

By Katherine Parr, Senior Food and Beverage Solutions Consultant, Aptean

Key takeaways:

• Most food and beverage companies aren’t lacking appetite for AI. They’re trying to adopt all five levels of maturity at once instead of starting with one step: basic visibility into operational data.
• Aptean’s survey of 300 F&B decision-makers (with Vanson Bourne) found that, for 85% of respondents, integrating AI with core business systems is the real challenge, and only 23% have made AI essential to daily operations despite 83% believing they’ll fall behind without it.
• Automate one workflow with a single human checkpoint, such as invoice matching or shelf-life-aware demand forecasting, prove it out, then connect it to the next workflow instead of building the whole system at once.

The company asking me about AI agents is often the same company still running on spreadsheets or wanting to move to a new ERP because they need to get their business processes in order.

That’s not a criticism. I’d say most of the companies I sit with aren’t choosing between AI vendors at all. They’re trying to get settled into an ERP, whether that means moving onto a new ERP or using theirs more fully. Then they’re curious what AI could do on top of that. Which is why something in our own research confused me at first.

We surveyed 300 food and beverage decision makers with Vanson Bourne this year, and 84% told us they’re leaning into AI because they’re genuinely motivated by what it can do, not because a competitor forced their hand. More than four in five (83%) believe their organization will go obsolete without it.

Yet only 23% have actually made AI essential to how they work.

So if it isn’t appetite and it isn’t budget, then something else is holding everybody else back. Here’s what I think it is. They’re scoping an entire five-year program when what they actually need is the first step.

The AI maturity ladder

There’s a rough progression of AI maturity most operations move through, and it goes something like this: AI that assists, then recommends, then optimizes across more than one process at a time, then executes inside guardrails you’ve set, and eventually coordinates whole workflows end to end. Five levels. Written out like that it reads like a five-year program, which is exactly why so many companies stall out in the exploring phase. They price out the whole ladder and decide to revisit it next year. They get overwhelmed with moving from no AI to full AI coordination.

And honestly, the hard part isn’t even the AI itself. Eighty-five percent told us that integrating it with their core business systems is the bigger challenge. Core processes need to be well documented and defined in order to get the AI workflows in place. But that’s a plumbing problem, and plumbing you need to do in a specific order.

The first rung: operational data 

If you’re mostly manual today, your first rung isn’t an agent at all. It’s read access to real operational data. A lot of organizations fall out right here, and it’s because the visibility itself is still a human process where somebody walks the floor or somebody opens four spreadsheets. People have to understand how their ERP functions before they can really understand what AI can do for them. Everybody wants to be more right tomorrow than they were today, right? None of that works without a data backbone underneath it.

The second rung: low risk decisions 

After that, hand the system exactly one low-risk decision and let it own that decision end to end. Reorder points, maybe. Not pricing. Definitely nothing customer-facing. And what you’re really doing at this stage isn’t automating anything, it’s finding out whether you trust what the thing tells you somewhere that being wrong is survivable.

The next rung: a single workflow  

This rung is where I’d say most people get stuck. Take a single workflow and automate it with one human checkpoint. One. Not five layered approvals. The example I talk about almost every day is sales and purchase automation, because when somebody in procurement is hand-matching purchase invoices against receipts, I start doing the math in hours. Ten hours a week, maybe. Forty over a month. What are we paying that person to do? Clean matches can close themselves. Real mismatches go to a person. That’s your checkpoint.

That one’s back office, and it’s where I see time come back fastest. Further up the ladder the stakes stop being administrative and that’s really where you start seeing returns on your efforts.

Say you have a batch of yogurt sitting in the back of the cooler with five days of shelf life on it. FEFO (First Expired, First Out) logic says move it first, but one of your customers requires at least three days of remaining shelf life on arrival and the lead time to ship to their DC is two days. That leaves a single day of margin, not enough buffer for a delay at the dock. A forecasting agent that understands lot-level shelf life, customer-specific requirements, and ship to lead times can catch that the batch technically qualifies for FEFO but doesn’t actually match the customer’s spec. The agent could reroute to a shorter lead-time customer or flag it for markdown before anyone finds out the hard way at a receiving dock. It will ask for confirmation before it touches the ERP, but you’ve taken what would have been five steps and turned it into one confirmation step.

Once you’re doing that, you can pull in things that have nothing to do with your own four walls, which is where I get carried away, because I work with a lot of produce companies and this one is near and dear to me — USDA pricing for the week you’re in, historical weather patterns, tariffs coming down the line on anything you import. Maybe your vendor says you’re getting a certain amount. Everything else says you might come up short. The example extends of course to all sub-industries in F&B, but you can see where I’m going with this. 

Anyway, somewhere else, a bad forecast gets you an ugly spreadsheet and an awkward meeting. In our industry, it gets you product in the dumpster. It’s waste you can see, and in the case of yogurt, smell.

There’s a second benefit here nobody ever puts in the business case. Right now you’re probably relying on one person in purchasing to carry every vendor quirk and workaround in their head. Which supplier always runs three days behind. Which PO never matches the delivery note. Who to actually call when a load shows up short. Every company I walk into has a Cindy. And when Cindy goes on vacation, or retires, all of that goes out the door with her. Honestly, that’s the thing I hear the most anxiety about when we start talking transformation. Not the technology, but the people who drive the most value suddenly not being around to handle all the manual tasks. Clearly defining business processes and then building AI automation into that allows for repeatable tasks that drive positive outcomes and lets your people spend time on more value-add activities.

Never the whole ladder all at once

Once that first workflow is running well, connect it to the one downstream of it. That’s really where this starts to compound. And it’s where IT stops being the bottleneck, right? They still hold the governance and the audit trail and the control points. They’re just not the ones hand-building every single workflow anymore.

I want to be really clear, because it’s where I see good intentions go wrong. It’s the next rung. Not the whole ladder all at once. And I’d bet the one in four who now call AI essential got there exactly this way, one rung at a time, never skipping ahead to look impressive or move faster than anybody was ready for.

So. What rung are you standing on today? 

Katherine Parr is a Senior Food and Beverage Solutions Consultant at Aptean, blending an engineering background with seven years of deep, hands-on experience with Aptean’s purpose-built food and beverage solutions. She specializes in fresh produce and manufacturing operations and has built a reputation for turning complex operational challenges into clear, actionable paths forward — helping food and beverage businesses harness AI-powered technology to modernize and future-proof their operations from the inside out.

Read the full research referenced in her piece here.