Glance at the agenda of any supply chain conference or leaf through any trade magazine and you’re sure to run across something about artificial intelligence (AI) and its vast potential to streamline logistics processes. And you’d be hard pressed to find a press release from a logistics tech vendor these days that doesn’t make mention of AI.
But you have to look a lot harder to find examples of users that have taken their supply chain AI projects beyond the pilot phase and achieved a substantial return on their investment (ROI).
One reason for that is the physical nature of logistics and its exceptionally low tolerance for errors. Logistics leaders and their partners need to move large volumes of tangible goods from source to destination, and any mistake in that process could disrupt the entire flow of goods, says Brett Webster, software product manager for advanced cloud and analytics platforms at systems integrator Dematic.
“Folks are still tiptoeing into warehouse uses of AI. It can be nerve wracking to hand off a workflow to something that may come up with new ideas. There is just so much on the line; operational performance is so critical to business performance and safety considerations,” Webster explains. “So [it’s no surprise that] businesses are not as willing to take risks in that area compared to, say, using AI to help rewrite an email.”
Another reason for the slow uptake of AI in logistics is that many users are unsure how to measure the success of their AI projects. But Webster says that doesn’t necessarily have to be a deterrent. He believes that logistics leaders can use many of the same key performance indicators (KPIs) they would use to monitor any other logistics workflow. Examples include cost per case handled, volume (in orders handled per day), inventory holding costs, warehouse labor costs, and the number of clicks or steps required to accomplish a task.
PICK YOUR BATTLES
So what will it take to overcome users’ resistance and persuade them to hop on board the AI train? As one tech exec sees it, the answer lies in finding the right use cases for the technology—in particular, the tasks that are most difficult and time-consuming for humans to do.
To illustrate his point, that executive—Lukas Kinigadner, CEO and founder of Anyline, a Vienna, Austria-based provider of mobile data capture technology—offers an example from his corner of the logistics world.
Kinigadner says his company recently released an upgrade to its barcode scanning software that combines AI and augmented reality (AR). Dubbed Barcode:AI, the software runs on standard smartphones and significantly increases scanning speed and accuracy, the developer says. But what really differentiates the system from traditional scanners is its ability to adapt, learn, and work in real-world conditions, Kinigadner explains. Among other capabilities, it can read damaged or poorly printed barcodes through use of contextual image processing, thereby reducing scan failures and the associated workflow disruptions. The tool also allows for multi-code capture and intelligent interpretation at scale, and it offers AI-enabled anomaly detection and automated data validation, the company says.
Thanks to those capabilities, the system generates a quick return on investment by enabling faster receiving, more accurate inventory checks, and better returns processing, according to Anyline.
From a broader perspective, Kinigadner says he believes the logistics sector is currently shifting from AI experimentation to strategic implementation by adopting targeted, high-impact solutions that deliver measurable ROI in weeks, not years. Examples include predictive maintenance tools to avoid equipment downtime as well as the use of autonomous mobile robots for picking, route optimization tools for deliveries, and advanced demand forecasting and inventory optimization tools for stocking and replenishment planning.
BETTER FORECASTING THROUGH AI
When it comes to companies that are successfully using AI for demand forecasting, Southern Glazer’s Wine and Spirits is a case in point. The Miami-based distributor of beverage alcohol operates more than 40 distribution centers and stocks more than 7,000 brands, so forecasting has traditionally been a headache and a half. In the past, the company relied on statistical modeling to create demand forecasts, but those forecasts weren’t always as accurate as the distributor would have liked. So about 20 months ago, it launched an initiative to incorporate AI and machine learning (ML) into its demand forecasting process.
As the effort gained traction, Southern Glazer’s hired two data scientists to manage and refine the new ML system—a task that involves measuring the results, soliciting feedback from demand planners and sales teams, and fine-tuning the “inputs” to make the AI system more accurate.
The system went live in May 2024 and is now used to manage forecasting for 57% of Southern Glazer’s’ total order volume, says Diego Fonseca, vice president, supply chain and logistics for the distributor. The company tracks the platform’s performance by using three metrics: forecast accuracy, the amount of ML “bias” in under- or over-forecasting, and the percentage of order volume managed by ML. According to Fonseca, the system is already outperforming the previous statistical modeling method and has continued to improve over time.
The ML platform is now in its fourth generation and includes 12 specialized models that cover different product clusters like beer, water, and wine.
Among other wins, the AI-powered demand forecasting tool has helped the distributor cut down on the amount of excess inventory it was carrying as a hedge against stockouts. “We capture a lot of revenue in inventory reduction, and that builds confidence in both lower-level workers and management, which opens up opportunities to apply AI in other areas,” Fonseca says.
That early success has won over skeptics who initially regarded AI as a kind of mysterious “black box” that would be excessively complicated to use and could even pose a threat to their jobs, he adds.
And that underscores the point that, much like fine wine, AI tools for logistics applications are still a niche product, understood mainly by a small cadre of experts. But also as with wine, users are discovering the magic that can happen when they pair the right AI variety with the right supply chain task. And since a hallmark of AI/ML is that its results improve with practice, the AI you choose to streamline your logistics operations will likely become more valuable over time … kind of like that fine wine in your cellar.
