Turn the campaign brief into one sentence that names the eligible Android user, delivery environment, destination and accepted outcome. For example: buy Android mobile-web traffic in selected GEOs for a compatible checkout, then optimize sources against confirmed paid orders after the normal approval delay. A clear sentence exposes missing decisions before money moves.
Build an eligibility matrix from the product requirements rather than from every targeting option in the interface. List the oldest supported OS version, required screen or browser behaviour, store or app availability, languages, payment methods and excluded markets. Your campaign should never purchase a route the product cannot complete.
Choose the customer route deliberately. Website traffic can move directly to a responsive page; an app acquisition flow can use a store destination; an installed-user journey may use a verified App Link or another tested deep link. Give unsupported or failed link handling a useful web fallback instead of allowing customers to reach an error.
Use names that make the hypothesis visible: GEO, environment, format, destination version and accepted event. Clear names let our source reports and your analytics records meet without a separate decoding exercise, and they reduce the risk that later changes blend incompatible evidence.
Prepare creative for the physical viewing context. Small screens need one recognizable promise, a legible primary action and assets that remain understandable after resizing. If a format can crop or truncate content, the central proposition should survive without relying on the last line or edge of the image.
Treat the destination as campaign infrastructure. Check load behaviour on representative low- and mid-range Android phones, mobile keyboards, orientation changes, consent choices, redirects and form validation. A technically delivered visit cannot create value if the person cannot finish the intended action.
Preserve source and campaign identifiers through every supported redirect and into the advertiser's business record. When an app or store route breaks the chain, document the attribution method and unmatched share instead of silently assigning all unattributed outcomes to the campaign.
Define acceptance before reviewing performance. An install can be provisional; a lead may be rejected; an order may be cancelled. Compare sources on the mature event that the advertiser values, and label earlier signals as diagnostics rather than substitutes for the final business rule.
Give new sources a bounded discovery allowance. The allowance should be large enough to observe the selected event but smaller than the loss the team is willing to accept. A promising source can move to a focused probation cell; a proven source must repeat under more than one fragile condition.
Read rates beside denominators and delays. One accepted event from a handful of visits is a useful clue, not a scale signal. A larger sample with no accepted outcome is stronger negative evidence, provided the destination and tracking were healthy for the same period.
Expand one material dimension at a time—budget, bid, GEO, source set or creative volume—so the cause of a change remains explainable. Evaluate the newly added spend rather than letting a strong historical average conceal weaker marginal traffic.
Keep a campaign decision log with the operator, time, reason, expected effect and rollback state. This small discipline stops repeated tests, helps agencies explain client changes and separates an inventory shift from a modification made by your own team.
