How to Turn One Repetitive Factory Task Into a Pilot Service

A support plan for testing robots on the factory floor has been announced

The Ministry of SMEs and Startups (중소벤처기업부) and the Ministry of Science and ICT (과학기술정보통신부) announced on September 7, 2026, that they would jointly support small and medium-sized manufacturers adopting physical AI. Physical AI means artificial intelligence that uses cameras and sensors to perceive its surroundings, then moves robots or machines to perform real-world work.

The support is not designed to transform an entire factory at once. In 2026, autonomous mobile robots and unmanned transport vehicles will be applied to individual logistics routes. In 2027, the scope will expand to factory-wide logistics. From 2028 onward, the plan is to manage production and logistics together.

The Physical AI Alliance will include a separate collaboration group for small and medium-sized manufacturers, intended to connect technology companies with manufacturers. Its core role is to find on-site pilot projects and help companies use the Ministry of SMEs and Startups’ smart-factory support alongside the Ministry of Science and ICT’s factory-operations support.

Earlier, KRW 80 billion was allocated to the 2026 Manufacturing AI-Specialized Smart Factory programme, which recruited about 400 projects. Factory deployment received up to KRW 200 million per project, while data collection and validation received up to KRW 50 million. This is a level of public support that shares testing costs which small manufacturers may struggle to carry alone.

But the starting point is low. A survey by the Korea Small Business Institute (중소벤처기업연구원) found that the AI adoption rate among small and medium-sized manufacturers was 0.1%, or roughly one in 1,000. Among companies collecting manufacturing data, 75.7% used manual entry, and 99.2% had no dedicated organisation for this work. If equipment is supported while factory records and operating staff remain unchanged, use may end when the pilot ends.

What might change first in a factory with 18 employees

Consider a hypothetical metal-parts machining company with 18 employees. After machining, workers move boxes of parts to the washing area, then send them to inspection. They push carts or call a forklift for the job.

Machining completion is reported verbally or through messaging, while movement records are written on paper and entered again in the office. When boxes pile up, it is hard to know which machine they waited in front of. Even when the company asks a robot supplier, it cannot immediately answer how many moves happen each day or how long the waits are.

This factory is likely to fail if it tries to automate everything from the start. Empty carts and loaded carts have different weights. Materials may be left in the corridor. On urgent delivery days, workers may use routes that differ from normal ones. If those exceptions are excluded and only normal conditions are shown, a demonstration may succeed while actual operations stop.

Instead, choose only the route to the washing area. For one week, record the number of moves, the time for each move, reasons for waiting, and situations that required human intervention. Then send the same work description to several technology companies to request test methods and costs.

Once the test begins, do not compare only movement time before and after deployment. Record each day the share of tasks the robot completed on its own, the number of times people had to clear its path, production downtime, and whether parts were damaged. These records make it possible to decide whether to continue, revise and test again, or return the equipment.

Some work does not change. People still decide how to stack out-of-specification parts, check corridor safety, and change the production sequence when equipment fails. A good service should therefore show clearly when people need to intervene, rather than promise to remove people.

Post-deployment management matters as much as the test. The Korea Small Business Institute analysed that adapting and retraining systems for the site, as well as calibrating connections between sensors and existing equipment, usually takes 12 to 24 months. The gap, then, is not simply robot sales. It is a service that covers task selection, pilot records, safety checks, and post-deployment operation in one flow.

Other countries also use small pilots to find reasons not to proceed

Mittelstand-Digital Zentrum Magdeburg, a German support organisation for small and medium-sized businesses, worked with TinkerToys to build a test setup in which a collaborative robot removes finished items from a 3D printer. The robot body alone cost about KRW 31 million, excluding sensors, grippers, and replacement plates.

The technology was feasible, but the company concluded that the investment cost would need to fall further before actual deployment. It treated finding a reason not to buy as a valid result, rather than defining pilot success as purchasing the equipment.

Made Smarter Adoption in the United Kingdom supports diagnosis and roadmap development before helping companies introduce equipment. Agricultural machinery manufacturer Storth Engineering identified a shortage of skilled welders and production delays as its core problems, then introduced a robotic welder and additional automation equipment in sequence.

The company reported a combined 20% productivity increase from several technology deployments. The figure comes from the company’s own announcement, but the sequence is still useful: it put on-site diagnosis and workforce reassignment before choosing equipment.

Singapore’s Certact Engineering used government support to connect robots and conveyors for loading and unloading materials from metalworking machines. Certact Engineering collected shop-floor data through the Internet of Things, or IoT, which connects equipment to collect and exchange data.

In a government-published case, the company said it saved seven hours per employee each month and about KRW 640,000 in labour costs per machine each month. Rather than using a broad productivity slogan, it calculated effects for one machine and one worker at a time, making it possible to decide whether to introduce the next piece of equipment.

Four services that can be built from this gap

1. Repetitive-task collector. A service that uses a phone to record a task name, daily frequency, time per task, and reasons for waiting for one week, then identifies one automation candidate. It is for a plant manager at a metal-processing factory with 10 to 30 employees who is considering robots but does not know which process to start with.

Because public support also starts with individual logistics routes, a record of one route is needed before a broad plan. The first screen should contain only an Add today’s repeated task button and four fields: task name, count, time taken, and reason for exception.

2. Robot proposal comparison inbox. A service that lets a manufacturer send one task description to several technology companies and compare equipment price, installation cost, safety equipment, training cost, and monthly management fees in the same format. Its main user is a food-packaging factory with around 20 employees introducing transport robots for the first time.

A policy to connect suppliers and manufacturers is beginning, but proposals in different formats are difficult to compare. The first screen should include Select a task to test, along with fields for travel distance, load weight, corridor photos, and daily frequency.

3. On-site pilot logbook. A service that brings together daily records of time, successful runs, human intervention, causes of stoppage, and safety issues before and after a robot pilot, so users can decide whether to continue deployment. A robot supplier and a factory manager running a two-month pilot at an automotive-parts factory would use it together.

As support programmes expand, comparable on-site records will matter more than demonstration videos. The first screen should show four items in traffic-light colours: today’s success rate, number of human interventions, downtime, and safety exceptions.

4. Post-deployment operations desk. A service that receives alerts from robots and sensors made by several companies, categorises their causes, and follows through with repair requests, worker guidance, and monthly performance checks. Factories with fewer than 30 employees that have bought automation equipment but cannot employ a dedicated technician can share its cost.

Because on-site adaptation can take one to two years, the management gap after installation may last longer than expected. The first screen should list, in order of urgency, equipment currently stopped, the impact on production, and the first action to check on site.

Why this matters where you are

The specific support programmes in this article are Korean, but the underlying test is portable: identify one repetitive task, record its exceptions, and compare the operating result with the expected result. Public funding, labour costs, and available suppliers may differ where you are. You can still check whether local manufacturers have the records needed to choose one pilot route before they buy equipment.

What to check today

Within 30 minutes, call one owner of a manufacturing business with 10 to 30 employees and ask which movement, loading, or inspection task people repeat every day with almost the same sequence. If one task emerges within five minutes and the company is willing to record its frequency and time for a week, the repetitive-task collector is worth testing first. If the task cannot be identified, narrow the industry and process before gathering technology companies.