Field notes · 18 Jun 2026

Giving the ad SDK a latency budget before it steals sessions

Module five of the Revenue Signal Lab is the one students did not expect. They arrived to argue about floors. They left arguing about milliseconds. An ad SDK that wins the auction after the tap target has moved is not a monetization success. It is a product incident with a revenue column attached.

Budget, not vibe

Pick a number with engineering, not with a vendor slide. For rewarded placements we often start with a budget the session can feel: if the reward prompt is visible, the ad has this many milliseconds to be ready or the prompt should decline gracefully. Interstitials need a different budget because the user did not ask. Banner refresh has yet another. One “SDK latency” KPI for the whole app is how the argument stays vague.

Join revenue to crash-free sessions

A week of rising eCPM next to a dip in crash-free sessions is not a mystery. It is a clue you are allowed to bring to stand-up. We ask students to put both on the same brief. Finance does not need the stack trace. They do need to know you are not proposing a floor change while the client is falling over.

Thailand networks are not a footnote

Latency that looks acceptable on a US office Wi-Fi is not the measurement. Operators shipping from Bangkok already know this; dashboards built elsewhere often do not. Fill Rate Diagnostics and the Signal Lab both keep hour-of-day views for a reason. Evening mobile networks in the region will expose a budget you thought was generous.

What we will not do

We will not tune your SDK for you. That is engineering. The desk’s job is to make the budget visible in the same conversation as fill and price, so monetization stops arriving as a surprise to the people who own taps. If your current review never mentions latency, start there before you buy another mediation feature.

Seats and lab dates live on the catalog. The Bang Kapi office will argue the number with you if you write first.