What does a pilot deployment need to prove?
That it can fail. Decide before it starts what result would stop the rollout, and if no result would, then you are running a rehearsal rather than a pilot. Rehearsals only ever prove that the rehearsal itself went well, which is not a finding.
It has to be able to fail
Write down, before anything is installed, what result would make you stop or change direction. A tap rate below something, a failure rate above something, a fallback used more than the tag.
If you cannot name such a result, the pilot has no decision attached to it and will be interpreted favourably whatever happens.
Share that criterion with whoever is sponsoring the project, so the interpretation is agreed in advance rather than negotiated afterwards.
This is the single thing that separates a pilot from a demonstration, and it costs nothing but honesty.
Pick the worst conditions
The worst location rather than the most convenient one: the metal doorframe, the corner nobody passes, the darkest aisle.
The worst phone in the audience, which means an older device with a case on it rather than the newest one in the office.
The worst moment, which is the busiest hour rather than a quiet afternoon.
A pilot in good conditions tells you the good conditions work, which you already assumed.
Watch rather than measure
Stand where you can see people and say nothing. Where they hesitate is the finding.
Numbers tell you that something is wrong. Watching tells you what, and it usually turns out to be the sign, the height or the gesture rather than the technology.
Ask three people afterwards what they thought the tag did. If they cannot say, the signage is the problem.
And count how many people walked past without noticing anything, which no arrival figure will ever show you.
How long and how many
Long enough to include a quiet period as well as a busy one, so the result describes normal rather than a launch.
Enough tags to cover genuinely different locations, since comparing locations is what isolates placement.
Few enough that changing your mind is cheap, which is the whole point.
And run it with the process you intend to use at scale, including the writing and verification, because the process is what the second stage of a pilot actually tests.
Frequently asked questions
How many tags in a pilot?
Enough to cover different locations and few enough that changing your mind is cheap. Ten to a hundred covers most cases.
How long should it run?
Long enough to include a quiet period. A launch week describes a launch rather than normal operation.
What if the numbers are low?
Compare locations against each other rather than against a target you invented, and watch real people before concluding anything.
What makes a pilot useless?
Having no result that would stop the rollout. Without that, it is a rehearsal and it will be read as a success.