Recency bias in Thai subscription apps

Recency feels scientific because it decays. In a Thai subscription app it often just describes the calendar. A seven-day window that includes Songkran, a payday Friday, or a World Cup match will inflate “highly engaged” and then scold the same users three weeks later for cooling off.

We see this most clearly in news and video products that still copy a Silicon Valley recency default. The product is used in bursts that follow Thai holidays, salary cycles, and school terms. The score pretends those bursts are personal loyalty.

Three distortions we annotate

Festival lift. Songkran and the long New Year close change both time spent and content mix. If your recency weight is steep, a user who only reads during holidays will look newly loyal every April and newly lost every May. Annotate the window. Do not “correct” it by adding a secret multiplier you cannot explain.

Payday reading. Several media chairs have shown us that paid-article opens cluster after salary dates. That is cashflow, not a new content habit. If your engagement definition includes paid consumption, recency alone will over-reward the first week of the month.

Campaign weather. A blast that reactivates dormant accounts will look like a health recovery if recency dominates. In Threshold & Bandcraft we ask teams to park campaign-induced opens in a diagnostic track so the operating score does not congratulate marketing for waking people up with a coupon.

What we do instead of pretending time is flat

In the Score Atlas, recency is a documented choice. Some products should weight it heavily — a ride-hailing app, for example, where yesterday’s trip is genuinely more informative than last quarter’s. A monthly magazine app probably should not.

When we keep recency, we pair it with a seasonal footnote: a short table of known Thai calendar effects that the drift-watch meeting must review. It is unglamorous. It stops a PM from declaring a “retention miracle” every Songkran.

If your score still uses a seven-day half-life copied from a blog post, try a month of reporting both the recency-heavy number and a recency-light twin. The gap is the bias. Bring that gap to a studio if you want help naming it.