Walk a trade show floor this year and the crowds gather around humanoids. They pick things up. They walk. They look like the future.
Meanwhile the technology most likely to improve your operation next quarter is a camera system that tells you where your bottlenecks actually are, or a retrofit kit that makes your existing racking work harder.
Will O’Donnell, managing partner of Prologis Ventures, and JP Keating, founder and managing partner of Prochain Ventures, made this argument at a recent event on logistics robotics. Keating put it in one line. The robot is not the product. The productivity improvement is the product.
For anyone signing off on an automation business case, that distinction is worth several million dollars.
The bottleneck just moves
A machine can be technically impressive and still fail to improve throughput, reduce errors or lower operating cost. Worse, automating one step without examining the rest of the process simply relocates the constraint.
O’Donnell describes the failure mode directly. Bring in a robot to increase picking and packing without addressing the throughput going out the door, and you create a jam at the dock that breaks the warehouse.
Anyone who has run a distribution center recognizes this. Faster picking produces pallets waiting on the floor. More pallets on the floor consume staging space. Less staging space slows loading. The automation delivered exactly what it promised at the pick face and nothing at all at the P&L.
The lesson is not to automate less. It is to measure the system rather than the station.
Where the quieter opportunities sit
Four categories deserve more attention than they currently get from operators.
Modularity has improved materially. Traditional fixed systems, conveyor installations in particular, resist adaptation when a customer’s needs or a site changes. O’Donnell points to a shift over the past four years driven by AI, where automation can serve multiple purposes. A mobile robot can be pulled out of one warehouse and deployed in another.
That matters for two reasons. Equipment that can be moved or repurposed has a longer useful life, and it has resale value. Both reduce the risk of stranded capital, which is the quiet terror of every automation investment made against a three-year customer contract.
Simulation has cut the cost of being wrong. AI-enabled simulation lets operators run iterations before anything is installed. O’Donnell notes that testing, piloting and visualization now happen faster, where previously a physical robot was needed in a real environment to run the test at all.
For buyers, this changes what you should demand during evaluation. Ask the vendor to simulate your actual flow profile, with your SKU dimensions and your peak curve, before you commit.
Operational visibility may beat the robot entirely. Keating describes computer vision tools that map movement across a warehouse or yard, identifying bottlenecks and underutilized assets without the year of human observation such a study used to require.
This is the least theatrical technology in the category and possibly the highest return. It tells you where automation is genuinely needed, and more usefully, where a process change would deliver more than another machine.
The orchestration layer is the real gap. Most warehouses now contain machines from several automation suppliers alongside existing warehouse, transport and yard management software. Making those systems work together, and showing managers the upstream and downstream consequences of a change, is where O’Donnell sees significant opportunity.
His framing of the daily reality is accurate. A manager running a promotion this week, with ten percent of the workforce out sick and a forklift down, needs to make decisions now. He points to the 2020 chassis shortage at the Port of Los Angeles as the macro version of a problem that happens at small scale in warehouses every day. Adjust one link and the effects travel both directions, and nobody can see them until the change is made.
The economics that actually decide it
Margins in logistics and warehousing are thin, which makes both the capital outlay and the speed of commissioning decisive.
O’Donnell lays out the comparison plainly. A $40m automation system carries a long and heavy payback, perhaps five years. Alternatives that lower implementation cost, run closer to plug and play, or come as robotics-as-a-service exist specifically to remove that upfront friction and accelerate the return.
His sharpest point concerns integration time. If integrating a robot into your IT systems takes six months, good luck running a pilot at all.
Keating supplies the clearest example of the principle. One company is developing ways to retrofit existing racking rather than replace it with a proprietary system. The capital outlay drops from roughly $100m to $5m, which, as he notes, gets attention. He also points to autonomous rail cars as a way to make underused freight assets more flexible.
Neither of those is a robot in the sense the press covers robots. Both do more for a network than a machine that impresses visitors.
Cobots now, humanoids conditionally
Collaborative robots are already practical. They absorb repetitive, physically demanding or hazardous tasks while people handle decisions requiring judgment. The case is not removing people from a process. It is removing strain and waste from it.
Humanoids offer something different, versatility across tasks and environments designed around people. The form alone does not create value. In a flat warehouse, wheels beat legs. Dexterous gripping remains hard technically and expensive commercially.
Keating is measured rather than dismissive. For heavy industry, high throughput and high reliability, humanoids are a tough sell at the moment, and the answer comes down to the specific job that has to be solved.
For now, the test is task-specific. A purpose-built machine with a clear return beats a general-purpose one promising broader capability it has not yet demonstrated at acceptable cost.
Two takeaways
First, write the throughput number into the contract, not the robot specification. Define the measurable gain you are buying, in units per hour at the dock door rather than picks per hour at the station, and make the vendor commit to it in your environment. This single reframing kills most of the proposals that would have moved your bottleneck rather than removed it.
Second, demand evidence of scale-out before you treat a pilot as proof. A successful pilot in one building with one customer and one SKU profile is not adoption. Ask for reference sites across multiple facilities and operating models, and ask what broke during the second and third deployments. The gap between a working pilot and a replicable network standard is where most automation value disappears.
The thesis is not that every warehouse needs more robots. It is that logistics operators need systems that adapt, connect and keep goods moving with less friction. The most valuable innovation in this category may be the least photogenic.
What did your last automation project actually change at the dock door, as opposed to at the workstation? And have you ever mapped your bottlenecks with vision tools before buying a machine, or did the machine come first?
Continue the discussion on Chain.NET, the global supply chain community, https://www.chain.net/c/ask.
Ask your questions, join events and debates, and see the full member benefits at https://gscc.co/benefits.