What to Add: Where Experienced Warehouse Operators are Placing their Bets

Guest blog by MHI member Logistics Reply

Conference season has a predictable rhythm. Every session promises the next technology that will transform your operation. Every demo looks flawless. And then you go back to the warehouse on Monday, and the real problems are still there.
The add recommendations from this panel discussion were a different kind of conversation. Three veterans close to a century of combined warehouse experience were not talking about what is exciting. They were talking about what they are actively piloting, what is showing real results, and what they think the industry cannot avoid by 2028. This is the third post in the series. Part one and part two covered what to keep and what to kill. This one covers what to build toward.

Start With the Problem, Not the Technology

Before getting to specific recommendations, the warehouse systems director made a point that framed everything that followed. The biggest mistake in technology adoption, she argued, is starting with a solution and working backward to the problem it solves. Her team’s approach is the opposite: start with a deep understanding of the operation, identify what is actually breaking down, and then ask what kind of tool addresses that specific failure.

This sounds obvious. In practice, it runs directly against the way most technology decisions get made, with executives attending demos, agreeing that a platform looks right, and then deploying it into operations that were never fully understood in the first place.

Her other foundational point was about growing with complexity rather than buying for a future state that has not arrived yet. Build the foundation first. Make it repeatable. Then layer in capability as the operation requires. Systems that were designed with the maximum use case in mind from day one are almost always overbuilt for where a company actually is today.

AI in the Warehouse: Hype Versus Reality

The moderator raised the topic that had been circling the conversation: AI. At conferences, AI dominates. On the floor, the picture is more complicated. She asked the panel to be specific about where AI is delivering practical operational value today, and where it is not.

The consultant offered one of the clearest framings of the session. AI gets credit for more than it is doing in many warehouse contexts, he said. Automation systems have had sophisticated control logic for decades, and much of what gets labeled AI today is an evolution of that capability, not a discontinuous leap. The genuine value of AI is in processing large volumes of data at high speed to surface patterns a human analyst would miss or catch too late.

Order management is one example he cited. The error rate on large-scale order management tends to concentrate in small details: a ship-to address that looks correct but differs in one field from the last 150 orders. AI can catch that. It can check at a volume and speed that no human team can match. But to do that reliably, it has to be trained on your data, in your environment, with your operators involved in teaching it what matters.

“AI will out-process you every day of the week on large volumes at high quality. But someone has to know how to tell it what to do, and someone has to own what it hands back.”

The warehouse systems director echoed the theme from her own experience with ASRS. Automation and AI work reliably in repeatable, high-volume processes with clean inputs. They struggle in exceptions, in edge cases, and in situations where the algorithm was not built for the specific variant in front of it. Her ASRS story from earlier in the discussion was the same lesson: the technology was not the problem. The absence of a human interception point for the cases the algorithm could not handle was.

The CPG veteran’s take on AI was forward-looking. He sees the tasker role, the warehouse office staff who manage wave planning, order drops, and operational coordination, as the most likely near-term target for AI displacement. Not because AI will replace warehouse workers broadly, but because that specific role involves high-volume, rule-based decision-making at scale: exactly where AI performs best.

Practical Add: Forklift Driver Assist

The CPG veteran offered a recommendation that was specific, low-cost relative to full automation, and immediately actionable. For operations that cannot yet justify AGVs, forklift driver assist technology provides a meaningful safety and efficiency improvement at a fraction of the cost.

The concept is similar to lane assist and collision warning in modern cars. The system keeps the operator within defined travel paths, warns when two forklifts are converging, and can intervene to slow or stop before a collision occurs. It does not replace the operator. It gives them better information and a safety net when attention lapses.

His framing was practical: you do not have to solve the full automation problem to make meaningful progress on safety and consistency. There are intermediate steps, and this is one of them.

Practical Add: Integrated Planning Over Point Solutions

The warehouse systems director described the failure mode she wants to move away from: isolated improvements that optimize one part of the operation while creating problems somewhere else downstream. Increase pick efficiency here, and you create a bottleneck in staging. Automate the staging lane, and the trailer loading becomes the constraint.

Her add recommendation was integrated process design, building toward solutions that account for the full flow from planning through pick, pack, and ship, rather than solving each step in isolation. The goal is a baseline design that holds across the full operation, with enhancements applied at specific points where the data shows the actual constraint lives.

She gave a concrete example from her own work: trailer loading optimization. The loading problem does not start at the dock. It starts when the pallet is first pulled from production or storage. Weight distribution, axle loading, dunnage placement, the order in which product moves to staging: all of those decisions need to be made at the beginning of the wave, not at the point of loading when it is too late to fix them. Her team is exploring smart technology that pushes those decisions upstream where they can still be acted on.

“The problem doesn’t happen in the bottleneck. It happens upstream and compounds into that bottleneck. You have to fix the starting point.”

Practical Add: Digital Twins for Simulation Before Investment

The consultant made a point that surprised some people in the room. Digital twin software, tools that let you simulate a shop floor, a warehouse layout, or a logistics network before committing to physical investment, used to cost a million dollars or more to implement. Today, he said, capable simulation tools are available for roughly $100 per user.

This changes the calculus on experimentation significantly. The old model was: build it, run it, find out what does not work. The new model is: simulate it first, surface the failure modes, involve the operators who will actually run it in reviewing the simulation, and then build with much higher confidence.

He was emphatic about the operator involvement piece. Innovation that gets designed in the back office and handed to the floor tends to fail. Innovation that gets designed with the people who will run it tends to stick. Digital twins make that collaboration possible in a way that was not practically accessible to most operations until recently.

His broader argument was that AI gets credit for things that are actually digital twin and simulation capability. The tools are not the same and conflating them leads to poor investment decisions. Know what you are buying.

Managing the Vendor Relationship Differently

The consultant raised a point that does not make it onto many conference agendas but belongs on this one. The relationship between an organization and its technology vendors has to change as AI becomes embedded in operational systems.

Legacy ERP and WMS implementations had a relatively stable profile once they were in. The system ran. You managed bugs. You planned upgrades on a predictable cycle. AI-embedded systems do not work that way. The model outputs can drift over time. The behavior can change as the model learns, sometimes in ways that are not visible until something breaks. Most AI vendors cannot fully explain what their models are doing or why. They know the outcome. The mechanism is opaque.

That opacity demands a different kind of vendor engagement, one that involves ongoing testing, close collaboration on how the AI is trained within your specific environment, and a clear process for catching drift before it causes an operational failure. Procurement as traditionally practiced is not equipped to manage that relationship. It requires people who understand the operation deeply enough to know what the AI is supposed to do and to recognize when it is starting to do something else.

Warehouse Operations in 2028: What Experienced Operators Are Predicting

The consultant was direct. The labor supply is contracting, and the pace of contraction is going to accelerate. This is not primarily about wages. Something structural is shifting in who is available to do physical warehouse work and at what scale. Organizations that are not actively building toward autonomous operations today will not be ready when the pressure arrives in earnest. The answer, he argued, is not to hand the problem to a consulting team and wait for a roadmap. It is to build toward autonomy incrementally, simulation by simulation, pilot by pilot, with the operators who will actually run those systems involved in shaping them from the beginning. He referenced electric vehicle manufacturing facilities he has worked with overseas, where output at scale is achieved with a fraction of the traditional labor count. That model is coming to North American distribution, on a timeline that is shorter than most planning cycles currently assume.

The warehouse systems director’s prediction was more focused. She sees the repeatable, high-volume tasks within the warehouse moving steadily toward automation: full pallet picking, wave planning, and order drops. What remains is exception management. The human role in the warehouse of 2028 is not gone. It is concentrated in the places where the algorithm cannot go, the unexpected, the edge cases, the judgment calls. Her name for the role that survives is not the tasker, but the exception manager.

The CPG veteran landed on the same note he started with. Investment in technology is one decision. Investment in the people who will operate, maintain, and improve that technology is a different decision, and it has to be made alongside the first one. Whatever the warehouse looks like in 2028, there will be people in it who need to understand what the systems are doing and why. Building that competency now is not optional.

“The repeatable options are going to be automated. We will be handling the exceptions. The human element in the warehouse isn’t going away. It’s just going to a different place.”

A Final Note

Across the entire panel discussion, a few themes kept returning: the importance of planning over purchasing, the cost of underinvesting in people, the danger of complexity for its own sake, and the gap between what technology promises in a demo and what it delivers in a live operation.

These are not new observations. But hearing them from people who have spent decades inside these problems, and who are still in the middle of solving them, gives them weight that a vendor presentation cannot replicate.

Logistics Reply hosts conversations like this because we believe the most useful thing we can do is bring real operational experience into the room and let it speak.