The next test for AI is not whether it can produce an impressive pilot. It is whether it can work reliably, every day, in the messy and high-stakes conditions of the physical world.
Companies are prepared to spend. BCG’s 2026 AI Radar found that corporations expect to more than double AI spending this year, from 0.8% to approximately 1.7% of revenue. But the value is not keeping pace. According to PwC’s 2026 AI Performance Study, 74% of AI’s economic value is being captured by just 20% of organizations.
The constraint is no longer ambition or budget. It is execution.
Latest Videos FromTechRadar
For physical AI, that challenge is especially acute. Unlike a tool that drafts an email or summarizes a document, physical AI operates where decisions affect safety, uptime, service delivery and cost. Moving it beyond the pilot phase requires more than a capable model.
Organizations need to choose a problem worth solving, give AI the operational context to understand it, design the technology around how work actually gets done and connect every insight to an appropriate action.
Start with the operational problem:
Much of the enterprise AI conversation has focused on large language models and office productivity. But some of the most consequential and costly business problems exist beyond the office: vehicle collisions, equipment failures, unplanned downtime and disruptions to the essential services people rely on.
Physical AI brings intelligence into these environments by using data from vehicles, equipment, cameras and sensors to understand what is happening and determine what needs attention.
In fleet safety, for example, AI tools can analyze information from connected vehicles and cameras alongside factors such as weather, road conditions and driving patterns. That context helps an organization distinguish an isolated event from a broader pattern and focus its attention where it can have the greatest impact.
Interest is already substantial, but deployment remains limited. In robotics, one of the most prominent applications of physical AI, Capgemini found that 79% of organizations are engaging with the technology, while only 27% are deploying or scaling it.
The right starting point is therefore not, “Where can we use AI?” It is, “What operational outcome do we need to change?” What is causing the problem? Is the necessary data available? What decision or workflow should the technology improve? A pilot without clear answers may demonstrate technical capability without proving business value.
Build the operational context AI needs:
Selecting the right problem is only the beginning. To move beyond a controlled pilot, AI must understand the context in which an organization operates.
Knowing a vehicle’s registration or model is not enough. AI may also need its maintenance history, current location, typical route, driver behavior and operating conditions. It needs to understand how those factors interact and what normal performance looks like across comparable situations.
When that information is fragmented across different systems, delayed or recorded manually, AI sees only part of the operation. Digitizing workflows and connecting operational data create the foundation that more advanced systems depend on.
The significance is not simply the volume of data. It is that the data is available in a shared operational system through which teams can manage vehicles, maintenance and customer commitments. Without that common foundation, even a sophisticated model risks becoming another disconnected tool.
Design for the operation, not the demonstration:
A pilot can be carefully managed. A scaled deployment cannot depend on constant intervention from a technical team.
Physical operations are distributed across vehicles, worksites, warehouses and frontline teams. Technology must be straightforward to install, reliable in changing conditions and connected to the systems employees already use. It must also work for the people expected to act on its output.
That makes change management part of the technical strategy. Employees need to understand what the technology is identifying, how it supports their work and what action is expected of them. Leaders need to establish ownership, measure whether workflows are changing and incorporate feedback from the frontline.
Governance must be designed into the deployment as well. Organizations should define which actions systems may take autonomously, when human review is required and how decisions can be understood or challenged. The level of oversight should reflect the potential consequences of the action.
Close the loop from insight to action:
An insight is not an outcome. Physical AI creates value when it shortens the distance between identifying a problem and doing something about it.
This can be understood through a sense, decide and act model. Connected vehicles, cameras and equipment sense what is happening in the physical environment. AI evaluates that information in context and decides what requires attention. A person or system then acts—whether that means alerting a driver, creating a maintenance work order or adding a damaged road to a repair queue.
A vehicle health warning, for example, can initiate a maintenance workflow and notify the relevant team before a fault becomes a breakdown. Pothole detection technology can use vehicles already travelling across a city to identify and assess road damage, giving public sector teams better information to prioritize repairs and plan maintenance routes.
The appropriate balance between automation and human involvement will vary. Routine, low-consequence actions may be automated within defined parameters. Decisions that could materially affect safety, employment or service delivery should carry proportionate human oversight. The objective is not autonomy for its own sake. It is faster, more consistent action with the right safeguards in place.
Beyond the pilot:
Organizations must close the gap between AI investment and value by treating AI as an operational capability rather than a standalone technology project.
That means starting with a measurable problem, building the contextual data foundation AI needs, designing deployment around the realities of frontline work and connecting every insight to action. When those conditions are in place, physical AI can move beyond isolated demonstrations and begin delivering safer, more efficient and more resilient operations at scale.
We’ve featured the best AI website builder.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit
