Select sites against the design assumptions
The rollout depends on assumptions about available power, cabling, equipment space, connectivity and the systems being replaced. It also depends on people: access contacts, local acceptance and remote support. The pilot selection should identify which of those assumptions need evidence.
A useful sample can include materially different site types, such as a constrained equipment room, an older cabling plant or a location with a different WAN arrangement. Geographic spread alone does not guarantee that these conditions are represented. Several distant sites can still share the same modern design.
The objective is not to include every possible variation in the first wave. It is to select the variations most likely to alter the deployment decision. The resulting pilot should be explainable in terms of the questions each site helps answer.
A pilot needs enough operational exposure
An installation completed during a quiet period can miss behavior that appears when the site returns to normal work. Employee sign-in, a delivery cycle, shift change or an application batch may exercise dependencies that a short installation test does not reach.
Google's canary-release guidance emphasizes a population and duration representative of the behavior being evaluated. The application to physical rollouts is an inference: a site pilot also needs exposure to the work that could reveal a defect. A successful installation appointment alone is a narrower result.
That does not mean every pilot must remain open indefinitely. The project should identify the relevant operating cycle and the signals that will inform the next decision. Duration follows the question being tested, rather than an arbitrary waiting period.
Separate a shared defect from a local exception
When a pilot encounters a problem, the response should determine whether the cause belongs to the standard design, a defined site variant or a one-off condition. These categories affect the next wave differently.
A missing standard adapter may require a kit change for every remaining location. A particular mounting condition may require a second approved kit. An isolated damaged outlet may require local remediation without changing the national design. Treating every issue as a universal defect can overcomplicate the rollout; treating every issue as local can repeat a shared failure.
The pilot record should therefore connect the symptom with the affected assumption and the evidence for the proposed correction. That is more useful than a general lessons-learned list whose entries do not identify which future sites are affected.
Update the actual delivery materials before expanding
A lesson only improves the rollout if it changes the work that the next technician receives. Revised kits, configuration, instructions and acceptance criteria need to reach the delivery process before the next wave begins.
Version control matters operationally here. The team should know which locations received which kit and procedure so that later incidents can be related to the deployed variant. An updated document in a project folder is insufficient if dispatched work orders still contain the earlier instructions.
The pilot should also test support capacity. A design that requires lengthy remote intervention at every site may work technically while exceeding the staffing available for a national wave. That constraint should influence the rollout rate or the preparation process.
Expansion should follow resolved questions
The next wave is justified when the pilot has answered the material questions, the corrections have reached the delivery process and remaining exceptions have a defined treatment. The number of completed pilot sites is useful context, but it is not the central result. The result is a deployment method that has been tested against the conditions it will actually encounter.