Journal · 12 June 2026

Dwell time is not intent

If your dashboard treats a long stay inside a polygon as interest, you are measuring friction as often as desire. Location-Based App Analytics has to name the difference.

Walk the edge of any Bangkok mall at lunch and you will find three kinds of stillness that look identical in a dwell histogram. Someone is waiting for a friend. Someone is stuck behind a payment queue. Someone is actually looking at a window display. Collapse those into “engaged visitor” and growth will happily spend against a queue.

What dwell actually records

Dwell is a clock on a geometry. It knows nothing about posture, gaze, or whether the person wanted to be there. SDKs that emit dwell_ms without a companion reason code force the analyst to invent a story. The story is usually flattering.

In District Signal Studio we ask students to label a sample of long-dwell events by hand after a site walk: transit wait, service queue, social wait, browse, unknown. The unknown bucket is rarely small. That alone should stop anyone putting dwell on a board slide without a footnote.

A schema that keeps you honest

Pair every dwell event with context you can actually observe: time of day, whether the polygon is a concourse or a store wing, whether the user had an open ticket or cart, whether the next ping is on a BTS alignment. None of these prove intent. Together they stop you treating a bus stop as a showroom.

If you cannot collect those companions, publish dwell as an operational metric — “time spent in zone” — and refuse the word intent in the same sentence. Product copy is not a substitute for a field you did not instrument.

Where this bites in Thailand

Food-court seating attached to retail fences is the classic contamination. So is the motorbike queue outside a convenience store. Both generate beautiful dwell. Neither predicts a SKU-level conversion. Corridor models, not blobs, make the contamination visible.

Further reading in this Journal: geofence overlap and indoor session cuts. To practise the labelling exercise, see District Signal Studio.