The most memorable part of a construction robot demonstration is often the movement of the machine. It covers a surface steadily, reaches a repeatable position or completes an application pass with an impressive rhythm. The observer can immediately imagine the potential. What is less visible is the work before and after that movement: positioning, protection, preparation, cleaning, inspection and the adjustments needed for the next location.
My position is that those surrounding activities belong in the initial comparison, even when the first purpose of a trial is to understand the application itself. Otherwise the programme can collect an attractive result that says little about the job a project team would actually buy. A machine's application time, the crew's person-minutes and the room's elapsed completion time are different measures. They can improve in different directions without any of the observations being wrong.
At Varelim, that distinction changed how we discussed a painting-robot trial. The question was not whether the robot looked capable during the application pass. It was whether the proposed method, under specified conditions, could make a useful contribution to site work. That required an account of the whole cycle and an honest description of which conditions had been observed. It also required resisting a single summary word such as faster when the measures disagreed.
Measure the work around the application
Our initial campaign began in May 2026, ran for eight weeks and cost S$360,000. It included six matched room pairs. One illustrative pair showed forty-two minutes of robotic application and sixty-eight minutes for setup, protection and cleanup, producing 110 elapsed minutes. The manual comparison used sixty-four minutes for application and thirty-five for the surrounding work, totalling ninety-nine elapsed minutes. The machine applied faster but the complete observed cycle took longer.
That pair was not the campaign average or a claim about every room. It was a concrete example of why the chosen measurement boundary mattered. If we had reported only application time, the reader would have received an incomplete impression. If we had reported only total elapsed time, the reader would have missed a potentially useful difference in labour demand. Both parts of the account were needed to understand why the method deserved further investigation without yet deserving a general deployment claim.
The setup contribution suggested an important question about geometry and continuity. Repeated positioning and protection may have a different relationship to productive work on longer uninterrupted wall areas than in smaller rooms. That was a hypothesis to investigate, not an explanation we could declare proven from the selected pair. It gave the next trial a purpose: examine whether the surrounding work could be spread over a more suitable application without creating other problems.
I wanted operations involved in selecting that next setting. A technology team can easily find a demonstration area that flatters its equipment while avoiding the practical conditions that make adoption difficult. Operations could identify where continuous surfaces existed in plausible work packages and which adjacent activities would still have to be accommodated. The investigation needed a setting that taught us something about a possible use, rather than merely producing a better photograph or a shorter application time.
Explain what the crew numbers do and do not mean
Two workers were assigned throughout the 110-minute robotic cycle in the illustrative pair, giving 220 person-minutes. Three were assigned throughout the ninety-nine-minute manual cycle, giving 297 person-minutes. The lower labour requirement and the longer elapsed duration coexist. There is no arithmetic contradiction. They answer different questions: how much assigned labour the activity used, and how long the observed activity occupied its work sequence.
Neither measure established a cash saving. Rental, support, transport, preparation requirements, quality outcomes and the organisation of the wider crew still needed commercial assessment. A reduction in person-minutes does not necessarily remove a paid shift or allow a person to be productively redeployed. Conversely, a method may have value in a labour-constrained setting even if the immediate payroll is unchanged. Those are questions for the actual operating arrangement, not conclusions that follow automatically from the multiplication.
Quality also needs a deliberate place in the comparison. A method that completes an initial pass quickly may create later correction work. A different method may produce a finish that is easier to inspect or more consistent in a particular setting. Our selected time account did not settle those questions. The consolidated engineering and commercial review had to consider the available quality observations before a wider recommendation could be made.
The leadership challenge is to make this complexity usable without flattening it into a sales claim. I would present the few measures that alter the decision, state the observed conditions and identify the remaining work. A sponsor should be able to see why we are interested and why we have not yet recommended broad adoption. Colleagues do not need every raw observation in the opening discussion, but they do need the distinctions that prevent a mistaken interpretation.
Design the next trial around the uncertainty
In July we authorised a further six weeks at S$260,000, focused on continuous wall areas. By September, the observations had been collected, while the consolidated engineering and commercial review remained unfinished. That was the position when this account was written. The follow-on campaign was not a portfolio rollout, and its authorisation did not mean that a saving had been established. It was a decision to investigate a more specific operating possibility.
I would now agree the principal measures with operations and finance before the first campaign begins. The application team should still collect the technical detail it needs, but the sponsor should know how the result will be interpreted in the work cycle and the commercial review. Doing this early helps avoid the familiar situation in which each function arrives at the final meeting with a different definition of success.
It also makes a decision to stop more credible. If the method cannot provide sufficient value in its plausible use conditions, the team should be able to say so without appearing to have failed a promotional commitment. The purpose of a trial is to reduce uncertainty about a choice. A well-founded rejection can be useful, just as a carefully limited continuation can be useful. Neither needs to be disguised as a completed transformation.
Construction automation will involve many different machines, tasks and site arrangements. The lesson I take from this campaign is therefore deliberately modest. Watch the machine, but also watch the people and activities that make its work possible. Keep labour, elapsed time and money distinct until the operating account connects them. That gives a promising demonstration a fair opportunity to become a useful construction method, while keeping the decision grounded in the job the project must actually complete.