"Indoor positioning" describes systems whose accuracy varies by two orders of magnitude. The distinction is not academic: each tier answers a different commercial question, and buying the wrong one means paying for data that cannot support the decision you bought it for.
Tells you someone is in a hall or a large section of one. Usually derived from network association or sparse detection points. Cheap, easy to deploy, often available from infrastructure a venue already owns.
Answers: how busy is this hall, how does occupancy change through the day, are we approaching a capacity limit.
Cannot answer: anything about a specific stand, exhibit or product. At this resolution a person is somewhere within an area containing dozens of commercial actors, all of whom would like the credit.
Resolves to a defined area: a stage, a catering point, a large stand. This is where most commercially available indoor analytics sits today.
Answers: which zones draw crowds, rough dwell distribution, broad flow between areas.
Cannot answer: which of two adjacent 9 m² stands someone actually visited, or whether they were engaged with an exhibit or waiting for a colleague beside it. At trade-show booth sizes, a 3 m error is the difference between one exhibitor's lead and their competitor's.
Resolves position tightly enough to attribute a visit to a specific object, stand or section within a stand.
Answers: which stand, which part of it, in what order, for how long, and whether they came back. This is the tier at which movement becomes behavior rather than occupancy.
Cost: denser infrastructure and meaningfully harder calibration, particularly in crowds. Bodies absorb and reflect signal, and a hall at peak behaves very differently from the same hall during setup.
Accuracy is not a specification you compare on a datasheet. It sets a ceiling on what can be built on top.
This is why "we also do indoor positioning" is an ambiguous claim. Two systems can both be accurate descriptions of that sentence and be incapable of answering the same questions.
Three factors matter alongside it, and they are frequently undersold:
Coverage. High accuracy across 20% of a venue is usually worse than moderate accuracy everywhere. Journeys break at coverage boundaries, and a broken journey cannot be stitched back together honestly. Partial coverage tends to produce confidently wrong conclusions about where people went.
Continuity. A position every few seconds describes a path. A position at three checkpoints describes three checkpoints. Continuous tracking is what makes hesitation, return and bounce observable, and those are the behaviors that carry intent.
Identity linkage. Accuracy without a consented identity layer gives you excellent anonymous geometry. That is genuinely useful for operations and entirely useless for lead generation. Conversely, identity without accuracy gives you a name attached to a vague region.
Questions worth asking, in roughly this order:
Question five separates two categories that are often presented as equivalent. A system dependent on app installation typically observes a minority of attendees, and that minority is not a random sample. It skews toward the engaged, which flatters every metric derived from it.
TRAKKER deploys at sub-metre precision, with full-venue coverage and continuous capture rather than checkpoints, because the product we are building is attribution: connecting behavior to a specific stand and, on explicit opt-in, a specific person.
Deliberately, the model is sensor-agnostic. The accuracy tier is a property of deployment, not of a hardware commitment, so we can run on what we install, ingest signal a venue already has, or combine both. As positioning hardware continues to improve, the intelligence layer does not need rebuilding to benefit.
