‘Physical AI’ is transforming warehouse operations beyond traditional visibility

Artificial intelligence is not a new technology, but its rapid adoption in the past two years has garnered all the headlines. From chatbots to generative and now agentic AI, the technology is all anyone seems to want to deploy.

But in warehouses, there is another trend quietly taking place, and while AI is part of the story, it is not the full story.

“I don’t know what flipped last year,” said Sankalp Arora, CEO of Gather AI, “but customers have now started to get why they need digitization; why they need solutions like us. If you know what’s happening in your warehouse, that leads to better on-time in-full, better labor productivity, fewer damages, fewer overages, fewer shortages.”

The realization that connecting floor-level visibility directly to OTIF, labor efficiency, and shrink reduction appears to be accelerating adoption of what Arora calls “physical AI.”

Closing the physical-digital gap

For years, supply chains have invested heavily in digital planning systems, control towers, and predictive analytics. Yet a persistent gap remained between what systems believed was happening in a warehouse and what was actually happening.

“There has been a gap between physical and digital thus far,” Arora told Supply Chain Management Review in a recent interview. “What we are doing is we’ll get you the on-the-ground truth of what’s actually happening in the facility.”

Gather AI’s approach uses cameras affixed to assets. Initially that was drones, but advancements have allowed the cameras to attach to other assets such as forklifts. Rather than requiring workers to scan barcodes manually, the system digitizes inventory state passively as the drone, the forklift, and potentially even people, navigate the warehouses.

“The natural extension … is getting moving cameras to collect data,” Arora said. “Things that move in a warehouse are pallet jacks, forklifts … people and other equipment.”

The long-term goal is simple: provide end-to-end visibility from inbound dock to outbound shipment without the manual labor to do physical inventory checks.

Beyond inventory counts

While automated inventory accuracy is the entry point for many customers, Arora says the real value lies in orchestration.

“Once you start getting all of that data, then we know what’s actually happening in the facility better than a lot of tools that they have,” he said. “Then we are very well positioned to orchestrate the facility.”

In practice, Arora said customers typically see ROI through improved on-time, in-full performance; labor efficiency gains and a reduction in overages, shortages, damages and write-offs.

In one case study Arora cited, an inventory control and quality assurance (ICQA) department shrank from six employees to one, while inventory errors dropped 70%.

But Arora pushes back on the idea that this is about labor replacement.

“I don’t think anyone in the warehouse really is replacing labor today because everyone’s short on labor,” he said. “We let people focus on tasks they’re good at and we take away the ICQA part of it.”

AI before AI was cool

Part of the confusion surrounding warehouse AI stems from terminology. Arora is careful to distinguish between different types of AI technologies.

“Before large language models came about, there was deep learning, which was really good for computer vision,” he explained. “Now, once you have that data digitized, then you get into the Gen AI and large language model stuff.”

Computer vision models interpret images and Classical Bayesian AI (an AI system that utilizes Bayesian statistics and probability) governs movement and guarantees in dynamic systems like drones. Generative AI excels at extracting insights and enabling natural language interaction but is “not really good at reading images … [and] not really good at controlling robots,” Arora noted.

In other words, AI is not one thing as many think of it, but rather a stack of technology executing various tasks. Arora notes that the industry has rebranded similar technologies repeatedly: machine learning, big data, AI, generative AI.

“AI existed way before all the marketing dollars were thrown onto them,” he said.

In a market saturated with “AI-powered” claims—some substantive while others are cosmetic—it is the ability to understand the different tools available and choose the right ones for the job at hand.

Hardware agnostic, capital efficient

Another defining element of Gather AI’s model is its hardware-agnostic approach. The company, which in February announced a $40 million Series B funding round led by Smith Point Capital, made a conscious decision early on to build the vision system and not the drones themselves. The software is designed to work with off-the-shelf drones or other assets. It is a decision that Arora believes was in the best interest of the company and paves the way to the best opportunity of success.

“We realized that the tech stack that we have is uniquely positioned to make off-the-shelf hardware work,” Arora said. This allows customers to avoid heavy capital expenditure and Gather AI is able to avoid large capital investments to build factories.

It also allows deployment in extreme environments from -20°F cold storage to some of the hottest warehouses globally. “We work in cold storage at minus 20 Fahrenheit, and we work in the hottest warehouses in the world,” he said.

Cold storage accuracy carries outsized importance, Arora noted.

“If you make a mistake … it’s four times more expensive than an ambient mistake,” Arora said. “It’s more important to get that right without human intervention, because it’s not a human-friendly environment.”

Arora says the latest funding round will primarily support scaling customer operations and expanding modality coverage.

The company currently operates in the U.S. and Dubai, with plans to expand globally.

SC
MR