Notes from people who had to explain a pin drop to their own board.
These are not composite quotes. Attribution styles differ because the people who write to us differ — some want a full name, some a city, some an industry only.
Kanya’s session on attribution windows finally gave us language for “the user was probably at the branch”. We now show a confidence band next to store visit counts. Finance hated it for a week and then started using it.
The Indoor Mall Telemetry short course assumed probe data we simply did not collect. The GPS-only workaround was documented, but I would not call that week a highlight. The consent module, however, changed our OS prompt copy within a sprint.
★★★★☆Dense-grid geofence rules from the Journal essay were applied in class to our Sukhumvit polygons. Visit volume fell 18% and suddenly matched receipt data. Painful, then relieving.
I came in sceptical of yet another analytics classroom. What I kept was the doubt column in the event schema. Our privacy counsel can finally see which location facts are inferred.
Path Reconstruction Workshop is short and slightly brutal. You walk a soi with a notebook, then try to rebuild the trace from the SDK dump. I still draw corridors on paper before I open Looker.
Rama IX mall visits without double counting
A fashion retailer in the Map Room Cohort arrived with three overlapping geofences: the mall footprint, the wing containing their store, and a “VIP lounge” circle that spilled into the food court. Weekly “visits” were heroic. Receipts were not.
Working through the overlap module, the team wrote a fence-priority rule: lounge only if already inside the store wing; store wing only if already inside the mall; mall never counted alone as a conversion precursor. They also split dwell at the food court into a separate event so lunch queues stopped looking like browsing.
After four weeks the visit-to-receipt ratio stopped oscillating every time a neighbouring brand redrew a polygon. The PM called it “boring in the best way”.
Permission-off as a segment, not a hole
A ride-adjacent consumer app treated missing location as dropped rows. Activation looked healthy among users who had already granted “always” permission and disastrous everywhere else — which was most of Thailand’s first-week cohort.
In District Signal Studio they added consent-state as a first-class dimension and rebuilt the first-session funnel around the OS dialogue, not the first ping. The “always allow” group still converted; the story for product became the prompt timing, not a phantom acquisition leak.
Limitation they still live with: iOS precise-location toggles remain noisy in their SDK version. The studio did not pretend otherwise.
