“We just need the robot to pick the product.”

“We just need to automate the unloading.”

“We just need to connect the systems.”

If you’ve spent enough time implementing automation, you’ve probably heard some version of these statements.

And you know what usually follows.

This week, The Wall Street Journal examined Walmart’s multibillion-dollar effort to automate its distribution network.

The company has made substantial progress. But after years of investment, some seemingly simple tasks remain surprisingly difficult to automate.

Handling irregular products.

Unpacking merchandise.

Accommodating existing buildings and infrastructure.

Even the packaging used to transport products can become a constraint.

The lesson isn’t that Walmart’s automation strategy is failing.

It’s that the complexity of automation rarely lives where we initially expect it.

The Robot Is Usually the Easy Part

Consider something as ordinary as unloading a trailer.

A person can walk into a trailer, recognize different pallet conditions, adapt to shifting loads, handle damaged packaging, and respond to unexpected obstacles.

We do these things almost instinctively.

For an automated system, every one of those conditions introduces another variable.

Pallet positions change.

Trailers aren’t perfectly aligned.

Loads shift during transportation.

Stretch wrap hangs loose.

Pallets arrive damaged.

Floor conditions vary.

The technology must either accommodate those variations or require the operation to eliminate them.

This week, Gideon announced the commercial availability of TREY, an autonomous forklift designed to load and unload trailers from the first pallet to the last.

The company says the system is already deployed in automotive manufacturing, food and beverage distribution, and third-party logistics operations.

That is meaningful progress.

But the important technical challenge isn’t simply navigating a forklift into a trailer.

It’s handling the variability that makes trailer operations difficult to standardize.

The task is simple to describe. The operating conditions are not.

Complexity Doesn’t Disappear. It Moves.

Traditional automation has succeeded largely by controlling the environment.

Standardize the load.

Standardize the container.

Standardize the presentation.

Standardize the movement.

Reduce variability until the machine can perform reliably.

That approach has delivered enormous productivity improvements across manufacturing and logistics.

But it also creates a trade-off.

The more rigid the automation, the more the surrounding operation may need to adapt.

Sometimes that means changing packaging specifications.

Sometimes redesigning material flows.

Sometimes introducing additional handling steps.

Sometimes requiring upstream suppliers to comply with new standards.

And sometimes investing in infrastructure that costs as much as—or more than—the equipment performing the task.

Walmart’s experience illustrates the challenge.

According to the Journal, the retailer has encountered difficulties with certain product categories, existing infrastructure, and the packaging requirements of its automated systems.

These are not necessarily failures of robotics.

They are reminders that automation is an operating-system transformation, not simply an equipment installation.

And that distinction matters enormously when evaluating investments.

The Variability Tax

There is a cost in automation that rarely appears as a separate line in the initial business case.

I call it the variability tax.

It is the additional cost of making automation work when reality differs from the conditions assumed during design.

It appears in different forms:

• Capital: Additional equipment, infrastructure, sensors, buffers, or handling stations.
• Labor: Operators managing exceptions, recovering failed transactions, or performing tasks the automation cannot handle.
• Performance: Micro-stops, reduced throughput, blocked flows, and slower recovery during peak demand.
• Complexity: More software logic, interfaces, maintenance requirements, and operating procedures.
• Flexibility: The cost of adapting when product profiles, volumes, or customer requirements change.

Some variability can be eliminated through better process design.

Some can be absorbed through buffers and operational flexibility.

And some requires more capable automation.

But none of it should be ignored.

Consider a system designed for 1,000 units per hour.

It achieves that rate under controlled conditions.

In production, however, exceptions, upstream interruptions, and downstream congestion reduce sustained throughput to 800 units per hour.

The equipment might still report excellent uptime.

But the operation has effectively lost 20% of its designed capacity.

If the business case assumed 1,000 units per hour, the financial model and operating reality are now describing two different systems.

This is why the cost of variability should be evaluated before the equipment is purchased, not discovered after commissioning.

Design for Reality, Not the Demonstration

There is a practical way to improve this.

Before finalizing an automation concept, I would ask the engineering and operations teams to perform a structured variability assessment.

Not another generic risk register.

A detailed examination of what actually happens inside the operation.

Start with five questions:

1. What percentage of the workload is truly standard?

Analyze real transaction data, SKU profiles, dimensions, packaging conditions, and operating exceptions.

2. What happens outside the standard operating envelope?

Does the system recover automatically, slow down, stop, reject the transaction, or require human intervention?

3. How does variability affect sustained throughput?

Test the actual product mix, peak-hour conditions, replenishment requirements, and upstream and downstream constraints.

4. What does exception recovery cost?

Include labor, lost capacity, maintenance, rework, and any additional infrastructure required.

5. What happens when the business changes?

New SKUs, different packaging, higher volumes, new customers, or a different order profile can fundamentally change the operating assumptions.

The objective is not to eliminate every possible exception.

That would be economically unrealistic.

The objective is to understand which variations matter, what they cost, and whether the system can absorb them without destroying the investment case.

The Next Competitive Advantage

Physical AI could fundamentally change this equation.

As robots become better at perception, manipulation, and adapting to unfamiliar conditions, the amount of environmental standardization required may decline.

That could make previously uneconomical workflows attractive.

But more intelligence does not eliminate the need for engineering discipline.

It changes the trade-offs.

The best automation architecture will not necessarily be the one with the most sophisticated robots.

It will be the one that delivers the required performance at the lowest total cost while accommodating the variability the business actually experiences.

The goal isn’t to build a system that works perfectly when everything goes right.

It’s to build one that performs economically when things don’t.

3 Signals Worth Watching

First Supply and Exotec turn automation into a customer promise

First Supply reports that its Exotec Skypod deployment increased picking speed tenfold and supports next-morning delivery for orders placed by 5 p.m., with room to quadruple capacity within the existing footprint. The interesting metric isn’t robot speed. It’s the ability to translate warehouse performance into a differentiated customer service commitment.

CEVA Logistics and Zalando bring AI-powered robotics into fashion returns

The companies have begun deploying robotic systems from Sereact to process fashion returns in Germany. Returns are particularly interesting because product presentation and condition can vary significantly. This is a useful test of whether AI-enabled manipulation can expand automation beyond highly standardized outbound workflows.

Volvo Autonomous Solutions and Waabi move autonomous trucking into commercial freight operations

The companies launched customer freight operations on the Houston–Dallas corridor with logistics provider Warp, with a human observer still onboard. The larger question is how autonomous transportation will transition from controlled operating domains to broader networks with greater variability.

The next time someone says, “We just need to automate that process,” ask what happens when the process doesn’t behave as expected.

Because the difficult part of automation is rarely the task we designed for.

It’s everything we assumed wouldn’t happen.