Every CPA offer owner eventually asks the same question: how do I know the conversions are real? This page explains how traffic quality is verified in practice — what bots, duplicates and incentivized traffic look like in data, how they affect your approve rate, and how Mobadly filters them on the buying side.
You submit the offer with vertical, GEOs, payout model and KPI definition.
Both sides confirm in writing what is billable, what is capped and what is prohibited.
A capped pilot on agreed sources with postback verification.
Sources and creatives are filtered by your approval data, not just conversion counts.
Volume grows only on cohorts that meet the agreed KPI.
In CPA, quality is not a vibe — it is the share of delivered conversions that survive your validation and behave like real customers afterward. A conversion is high-quality if it passes your approval (the deposit clears, the order confirms, the subscription activates) and its cohort shows real downstream behavior (retention, redeposits, activity). This definition matters because it aligns both sides: the buyer is paid only for what validates, so the buyer's incentive is to filter aggressively before delivery. Quality failures in CPA are almost always one of three things: automated traffic, duplicated conversions, or incentivized users — each with its own signature and its own defense.
Automated traffic betrays itself statistically. The classic signatures: click-to-conversion times no human produces; conversion time distributions that are flat instead of following daily human rhythms; device and browser fingerprints repeating at impossible rates; GEO mismatches between click, device settings and conversion; and conversions that never show any downstream behavior. No single signal convicts — real users sometimes look odd — but sources showing several signals at once are cut from rotation. Detection runs before delivery where possible (placement-level filtering) and after delivery through reconciliation against your system of record within the hold window, so bots are caught before payment rather than disputed after.
Duplicates are the quietest leak: the same user or order converted twice through different paths — organic plus paid, or two sources claiming the same user. Deduplication rules agreed upfront — first valid postback wins, defined attribution windows — close most of it; the rest is caught in reconciliation. Incentivized traffic is the more dangerous category: real humans, worthless intent — users paid to register, install or deposit the minimum. It passes naive validation and fails every downstream metric. The defense is cohort analysis: incentivized sources show approval followed by zero retention, zero redeposits, zero activity. When you share cohort data per source, incentivized supply is identified and cut — which is why the approval-and-behavior feedback loop matters more than any single fraud filter.
Your approve rate is the scoreboard where traffic quality shows up — but read it correctly. A falling approve rate can mean degrading sources, but it can also mean your call center slowed down, your delivery coverage slipped, or your funnel changed. This is why we reconcile approval data per source with you: sources with systematically low approval are filtered out, while offer-side causes are flagged back to you with data. The honest division of labor: we answer for what the sources deliver; you answer for what your funnel does with it; the reconciliation loop tells us both which is which. Offers that run this loop weekly keep approve rates stable; offers that review quality monthly discover problems after they have cost real money.
Four practices make any CPA supply cleaner. First, validate fast: slow approval windows hide bad sources and punish good ones — approve within the shortest window your operations allow. Second, share rejection reasons: a rejection without a reason cannot be optimized against. Third, watch cohort behavior per source, not just approval: retention and redeposit data catch what validation misses. Fourth, keep caps during validation: capped volume limits what a bad source can cost while it is being identified. These practices are built into how we work with offer owners — the KPI and payout terms page describes the written mechanics.
Reports, postbacks and disputed conversions are handled against written KPI terms and logs on both sides. Payout models, schedules and hold periods are agreed individually per offer and confirmed in writing before traffic starts.
Through layered verification: behavioral signals at the source level (timing, fingerprints, GEO consistency), deduplication rules, and reconciliation of conversions against your system of record within the hold window. What fails validation is not paid and its source is filtered.
Real users paid to complete your action — register, install, deposit the minimum. It passes naive validation and fails every downstream metric: no retention, no redeposits, no activity. Cohort analysis per source is what catches it, which is why we ask for approval-and-behavior feedback.
Two families of causes: source-side (degrading or incentivized supply) and offer-side (slower confirmation, delivery issues, funnel changes). We reconcile approval per source to tell them apart — the data usually shows which it is within days.
By deduplication rules agreed upfront — first valid postback wins, defined attribution windows — plus reconciliation against your system of record. Rules are written into the KPI definition and apply consistently.
Conversions with validation status, traceable to source attributes; the granularity of placement and sub-source detail follows the transparency agreement for your engagement. Quality questions always get a data answer.
As fast as your operations allow. Slow validation hides bad sources, delays optimization, and strains the partnership. The hold window is fixed in the agreement so both sides plan around it.
They are cut from rotation and their pending conversions reviewed against the written validation rules. Systematic quality failure ends the source's access to your offer — that feedback loop is what keeps supply clean.
Yes. Caps and pauses are normal quality-management tools, agreed in writing rather than discovered from rejections. Quality issues are worked through with data on both sides.
Short brief — we reply through your messenger.