Warehouse Automation Has Raised the Technical Ceiling. What About the Brownfield Floor?

Before a warehouse introduces its first robots, it needs to establish how the operation currently works and what makes sense to automate. An apparent picking bottleneck might originate in replenishment, misplaced stock or exceptions that never appear properly in the data. By the time the operator has identified a suitable application, checked its systems and building and planned installation around live orders, considerable work may already have taken place.

Very little of that preparation is visible in a photograph of the finished installation. Yet it may explain why the second investment feels quite different from the first.

I think of this as a progress bar, starting with a largely manual warehouse and extending towards lights-out fulfilment. The bar represents the business becoming ready to automate more of its operation, with much of the early progress taking place before equipment arrives. The first successful deployment might sit somewhere around the middle, reflecting how much has had to be established for the first time. That position is illustrative, rather than a claim that the warehouse is 50% autonomous. The next investment may benefit from preparation the business no longer has to repeat, even where the technical challenges are different or more difficult.

Steve Banker’s Forbes article, “When Will Dark Warehouses Become Common?”, prompted me to think about this. As more of the work inside a warehouse becomes technically possible to automate, I wanted to look at what would make that capability accessible to an existing operation approaching its first substantial investment.

There is still considerable room for adoption. Prologis Research estimates that 30% of modern logistics space incorporated at least one form of automation in 2025, up from 20% to 25% five years earlier. This measures space using a broad range of technologies, rather than the number of warehouses approaching full autonomy, but it puts the discussion about lights-out facilities alongside a market with substantial scope for first investments.

Before the first installation

Companies introduce automation for a number of reasons, from increasing capacity and improving accuracy to addressing labour shortages or reducing physically demanding work. For an existing warehouse that still relies largely on manual processes, establishing how to achieve those improvements can involve examining decisions made over many years. A layout adapted for successive customers may bear little resemblance to the original design, while the people keeping the operation running have developed workarounds that are familiar to their colleagues but poorly documented.

A supervisor may know exactly which locations cause problems and when replenishment falls behind. Turning that knowledge into requirements a supplier can use takes time, particularly when the same people are still responsible for getting the day’s orders out.

Preparation may involve trialling pick-to-light alongside reviewing pick paths and product locations, giving employees an opportunity to explain where instructions are awkward or why a workaround exists. Where a trial introduces closer monitoring of individual performance, people may feel scrutinised while learning or worry about how the information will affect their jobs. Explaining what is being measured and why, allowing time to adapt and involving employees in adjustments are part of the change management needed to establish the process.

Physical conditions, right down to the concrete flooring, also need attention. Depending on the equipment, floor smoothness and flatness can affect an installation’s suitability, with Prologis’s research including flooring among the relevant building features. Any remedial work needs a cost, an owner and a place in the operating schedule.

Integration adds another substantial demand on the business. In Interact Analysis’s July 2026 research on material transportation, 45% of respondents identified integration complexity as a barrier to further adoption, compared with 41% citing high upfront costs and 38% citing budget limitations. The research also found companies phasing rollouts and starting with processes requiring less integration with their warehouse management system.

Someone must establish how instructions will pass between systems and who investigates when each supplier believes its own part is working correctly. This requires time from IT teams and supervisors alongside the physical preparation and employee support, all of which contribute to a commitment the equipment’s purchase price alone cannot capture.

The impact of a first successful deployment

After a successful deployment, the business has experience of how the technology performs in its own operation. It knows which assumptions needed changing, how long employees took to become comfortable with the process and what support was required after go-live. Those findings can inform the next proposal, provided someone has recorded them and distinguished what can carry across to another application from what needs to be assessed afresh.

Radial offers an example of what that learning can look like. In a published account of an interview with David Welsh, identified at the time as its VP of Fulfilment Services Delivery, Locus describes how the operator reconsidered robot numbers and allocation after its first rollout with the supplier.

The second deployment extended the robots’ use into replenishment, putaway and returns, with associates overseeing several workflows within a zone. Welsh described shorter onboarding through the interface and the continuing need to align labour planning with robot utilisation. Coordination with technology partners over waypoints and queueing processes was still required, but the team could make those adjustments with a better understanding of where the robots were useful.

This supplier-published account shows learning within an existing robotics programme, although it provides no comparable cost or implementation-time figures to quantify the benefit. The broader question is how much of that learning transfers when the next investment involves a different system or supplier.

A method for evaluating suppliers, involving employees in trials or recording support requirements can remain useful even when the equipment changes. Technical connections and training materials may need to be developed again, but the business can draw on experience of managing the introduction and judging the result.

Zalando illustrates how distinct applications can sit within one network: Nomagic’s picking systems, including shoebox handling, and Sereact’s Cortex-powered dual-arm robots processing returns at CEVA-operated facilities. These announcements demonstrate investments in different processes; they do not establish that one deployment made the other easier.

When the first experience creates more work

A poor deployment can leave the business further away from being ready to invest again, particularly if the same employees who would lead the next project are still trying to make the existing one work.

Interact Analysis’s order fulfilment research describes an automotive and retail 3PL whose warehouse structure had shifted, leaving beam levels uneven and creating problems for a stacker crane that required precise calibration. The system had to be dismantled, the floor rebuilt and the automation reinstalled in another warehouse. The respondent reported that the project failed to deliver its expected return. The account does not establish that this was a first deployment or describe what happened to later investment decisions, but it demonstrates how a physical condition that has not been adequately understood can undermine the economics of an entire project.

The consequences can also last beyond the installation. Managers who supported an investment may have to explain why it fell short, while employees who spent months compensating for its shortcomings have reason to question the next proposal. Even where the appetite to automate remains, the budget and people available for further improvements may be absorbed by remedial work.

Representative trials can expose problems while there is still room to change the proposal. McKinsey’s guidance on warehouse automation recommends representative pilots, phased investment and contingency planning to protect business continuity. A manageable initial scope gives the operator room to learn, but excluding every difficult product or busy period postpones important learning until after a larger commitment.

Making the first commitment manageable

Robotics-as-a-Service can help an operator get started by reducing the initial equipment outlay and bringing support into the agreement. Where deployment, maintenance and training are included, a business without an internal robotics team can access capabilities that would otherwise need to be recruited or contracted separately. Some arrangements also allow additional robots during busy periods, reducing the need to own a fleet that is sized only for a short seasonal peak.

Formic’s three-month palletiser trial, offered to manufacturers and co-packers, is one example. It includes installation, programming, training, monitoring and maintenance, after which the customer can continue through a service arrangement, purchase the system or end the trial.

The trial needs to encounter the conditions under which the equipment will be expected to perform, while rights to add robots, reduce a fleet, relocate equipment or leave an agreement need to be understood individually. A 3PL with changing customer requirements may value those options differently from an operator with stable volumes. Site preparation and internal project time remain costs, so the comparison needs to cover the full commitment and what happens if demand develops differently from the forecast.

RaaS also creates a continuing reliance on the provider’s ability to maintain and support the fleet.

Understanding who handles faults, how quickly help is available and what maintenance is included establishes how much responsibility the supplier takes on and what the warehouse team still needs to manage.

Knowing whether it worked

The data needed to select a system and the data needed to judge its success are closely connected. Item dimensions and order profiles help establish what equipment is suitable; inventory and location records affect whether it can perform as intended. Before deployment, the business also needs a baseline it can trust, otherwise it may finish with functioning equipment and no convincing answer to the question that will be asked when another investment is proposed.

Has the operation actually improved enough to justify the cost?

The measures should reflect the original reason for investing, whether that was reducing physical strain, improving accuracy or increasing capacity. For a picking deployment assessed against order lines per paid labour hour, the result depends on whose labour is included and whether the order mix has changed. If employees have moved into replenishment or exception handling, counting only the remaining pickers gives an incomplete picture. Higher output also needs to be considered alongside dispatch performance. Agreeing these definitions can take considerable work during a first project, but leaves the business better able to assess subsequent proposals consistently.

The next proposal can account for support work omitted from the original forecast and distinguish additional capacity from savings actually realised. Experience can also inform training estimates, although a different system may place new demands on employees.

Keeping those records accurate requires someone to remain responsible for them. Receiving errors and unrecorded stock movements can quickly undermine preparation that took months to complete.

Why warehouse intelligence belongs in this discussion

Warehouse intelligence can support those decisions by giving an operator a more reliable account of what is happening inside the building. Systems that observe physical stock and reconcile it with digital records can address inventory discrepancies while producing information relevant to later investment. For a business struggling to locate stock, that may be a useful first application in its own right.

Recent funding shows the interest this area is attracting. Gather AI announced a $40 million Series B in February 2026, reporting 250% bookings growth and a doubling of its operational footprint over the preceding year. Dexory announced $165 million in financing in October 2025, including a $100 million Series C and an expanded debt facility. These company figures indicate momentum, although they do not establish a category-wide growth rate or explain how much demand is linked to further automation plans.

The operational case is easier to examine at customer level. In a Dexory case study of DCL Logistics, the supplier reports 16 labour hours saved per day through automated inventory checks, while DCL president Dave Tu describes improved accuracy supporting faster operations.

That improvement can be worthwhile without another robotics project following it. If the operator does automate further, reliable inventory records can inform its assessment of where to invest. Within existing installations, AutoStore’s CubeVerse and AutoStore Intelligence announcement describes analytics intended to improve operation and increase throughput without additional hardware. In either case, someone needs to act on the findings: a recurring discrepancy needs investigation, and a bottleneck requires a decision about the process causing it.

What Walmart shows about the work that remains

Walmart’s experience shows how much adjustment can sit behind an automation programme. Recent Wall Street Journal reporting describes changes to box dimensions and an additional taping step to accommodate automated handling, while oversized products such as large bags of dog food remain difficult to handle. After earlier sorting approaches failed to perform adequately or proved too expensive, Walmart developed small wheeled robots. All of this has had to happen while its warehouses continue operating.

Even the dimensions of the box had to be reconsidered.

That detail matters when considering what an operator is buying into. These systems enter buildings where packaging, layouts and working practices have developed around people, sometimes over decades. Introducing equipment can therefore require changes beyond the task it is intended to perform, with additional work for employees and alternative arrangements for products the system cannot handle. Those adjustments need to be reflected in the business case, including the time required to identify and implement them.

Walmart qualifies the progress bar: a business can become better at introducing automation while still encountering difficult new applications. Experience may help it recognise problems earlier and budget more realistically, even when the technical solution needs to be developed afresh.

Carrying the work forward

Preparation has lasting value when the knowledge remains available to the next team. Recording commissioning changes, exceptions and actual support requirements reduces the risk of repeating work when an employee leaves or a different supplier becomes involved.

None of this requires lights-out fulfilment to be the eventual destination. A smaller set of well-chosen investments may deliver the capacity, accuracy or working conditions an operator needs.

What deserves more attention is how the first deployment changes the choices available afterwards. A warehouse with reliable records and experience of introducing technology can bring that preparation to assessing a different process or system. As the technical ceiling rises, helping more operators make that first investment successfully, and retain what they learn from it, will influence how much of the industry’s capability they can use.