What makes a robotics pilot worth running?

What should a robotics pilot decide?

A pilot should decide one specific next action: proceed to a larger assessment, change the proposed solution, collect a missing piece of evidence, or stop. It is not a miniature sales demonstration. If the team cannot state the decision before the work starts, the pilot is unlikely to settle the important question afterward.

Begin with a bounded physical task, the operating conditions and the outcome a responsible owner needs to see. For example: collect inspection evidence on a defined exterior route during a planned safe state; move a stated load between two nominated points; or place a defined part into a fixture at a stated quality and rate. Do not turn a broad ambition such as 'automate the warehouse' into a pilot scope.

What belongs in the pilot brief?

This creates an honest comparison point. A provider can explain what its configured system is intended to do; the operator can define what the site will accept; and neither side needs to imply that a trial result proves broader deployment readiness.

• The baseline process, the proposed robot-assisted task and the conditions that must remain unchanged during the test.
• The measurable result: coverage, cycle completion, quality, evidence quality, recovery rate, intervention requirement or another agreed operational measure.
• The configured system boundary: robot, tooling, sensors, software, safeguards, interfaces, charging or utilities, and the provider support included in the test.
• The location, operating window, people, hazards, site rules, access controls and accountable safety owner.
• The raw evidence to retain, the person who reviews it, known unknowns and the condition that would stop the work.
• The decision meeting, decision owner and the options available after the pilot ends.

Is this uncertainty worth paying to resolve?

Before commissioning a pilot, write down the decision the buyer would make today. Compare continuing the current process, the configured robot option and at least one plausible alternative such as a process change. Then identify the uncertain input that could reverse that choice. It may be staffed recoveries per shift, quality at the required rate, integration effort at a real interface, or the support needed after a stoppage. A pilot has little decision value if every plausible result leads to the same next action.

A document review, site survey, sample demonstration or simulation may answer the question at lower cost. Use a paid site pilot when those methods cannot produce credible evidence under the operating conditions that matter. This is RobotAtom's decision method, not a universal investment rule. Australian Department of Finance business-case guidance considers a base case, alternatives, costs, benefits, risks and whole-of-life effects; Infrastructure Australia's options guidance focuses sensitivity analysis on uncertain inputs that could change the ranking. Both publications address public investment, so a private buyer must set its own commercial limits.

Write the result that would change the decision

Agree a conditional decision before paying for the test: if this configured system completes the named task under agreed conditions, with interventions and installed and recurring costs inside buyer-approved bounds, the decision owner will request a proposal or wider validation. If it misses, the buyer will revise the solution, collect specified missing evidence or stop. Leave the numeric limits blank until the operator and finance owner supply them from their own baseline and requirements; a supplier's attractive target is not a buyer-approved threshold.

Define the baseline period, operating range, measurement method, treatment of retries, exception log, evidence owner and decision date. The U.S. Department of Energy's measurement-and-verification guidance illustrates why baselines, assumptions and calculation methods need to be explicit, but it concerns energy-savings contracts and does not establish a robotics test method. Record an inconclusive result as inconclusive. Do not turn a narrow observation into a multi-site performance or return-on-investment claim.

Carry unknown costs into the post-pilot decision

These are prompts, not a claim that every project incurs every cost. Give each unknown an owner and a range or a request for evidence. If a pivotal cost remains open, a successful task run may justify a costed proposal, not a purchase commitment. The U.S. GAO technology-readiness guide distinguishes evidence for a governance decision from knowledge-building and calls for evidence representative of the operational environment. It is federal acquisition guidance, not a certification or a private robotics approval.

For example, a hypothetical transfer task may look viable in a provider demonstration while the number of staffed recoveries per shift remains unknown. A short site pilot could measure those recoveries across agreed shifts and exceptions. The buyer still needs an installed and support scope before deciding whether the robot option is preferable. The useful result may be a no-go or a narrower question; that is better than buying on a demonstration that never tested the disputed input.

• One-time: engineering, tooling, interfaces, site preparation, commissioning and staff training.
• Recurring: software, support, supervision, utilities, spares, maintenance and recovery work.
• Unknown: the cost of an unresolved interface, changed work, additional safeguards or a support obligation that has not been priced.
• Learning cost: pilot equipment, site disruption, measurement, provider and operator time, and any follow-on test needed after an inconclusive result.

Why is the configured system the unit of evaluation?

A robot platform alone does not perform the work. End effectors, sensors, fixtures, route design, fleet controls, network conditions, operating procedures, safeguards and recovery arrangements can all change the outcome. Record the exact configuration tested and any deviation made during the pilot, otherwise a result cannot be reproduced or compared fairly.

ISO 10218-2:2025 addresses industrial robot applications and robot cells across design, integration, commissioning, operation, maintenance and decommissioning. Its public scope is a useful reminder that an industrial application is a complete system and lifecycle, not simply a robot model. The standard does not approve a specific pilot or replace the site-specific engineering and safety work that may be required.

How should safety and commissioning be treated?

Treat safety, access and commissioning as entry conditions, not metrics to trade against a performance result. Identify the accountable safety owner, the permitted operating mode, emergency and recovery procedures, exclusion or collaboration arrangements, and any specialist review before running the work. A pilot that gets a useful throughput number while relying on an uncontrolled condition has not produced a deployment decision.

Universal Robots' current commissioning guidance tells integrators to test connected safety inputs and outputs and determine the commissioning requirements of the application. This is manufacturer guidance for its own products, but the principle is broader: a demonstration is not a substitute for verifying the configured application and its safeguards.

What evidence makes a pilot credible?

Keep the pilot evidence close to the question. Retain the task definition, dated configuration, run conditions, raw captures or system logs where appropriate, exceptions, human interventions, failed attempts and the method used to calculate every reported measure. Label estimates, one-off observations and incomplete runs rather than smoothing them into an average.

A credible outcome can be negative. If the task is too variable, the site needs a different fixture, the route is inaccessible, the system needs an interface that is not available, or the safety boundary is impractical, write that down. A timely no-go can prevent a much larger mistake.

What happens after the pilot?

Hold the decision meeting against the pre-agreed criteria. The result may justify a site survey, a revised requirement, a commercial proposal, a longer validation, a different robot class or a stop. Do not silently expand scope because the demonstration looked promising.

RobotAtom can help turn a physical-work problem into a structured brief and organise the evidence needed for an assessment. It does not certify a site, approve a safety case, endorse a provider or guarantee that a pilot will succeed.