A video campaign can report an attractive acquisition cost while the app's measurement partner credits far fewer users to it. Before increasing or cutting spend, establish what each report counted. A click, an impression, an engaged view, a matched event and an incremental customer are different pieces of evidence.
Click-through attribution assigns conversion credit after an eligible click; view-through attribution assigns credit after an eligible impression without a qualifying click under the reporting system's rules. Engaged-view attribution applies a separate engagement requirement. None of these labels, by itself, establishes that advertising caused the conversion. The useful comparison records interaction type, attribution window, event definition, matching method and reporting scope before drawing a budget conclusion.
This guide is for app founders, heads of growth, UA leads and measurement owners reviewing an agency report, a video channel or a proposed attribution-setting change. Its purpose is to decide which differences require correction, which require explanation and which require stronger evidence.
View-through vs. click-through attribution: what changes?
The comparison below describes the questions a buyer should ask. It does not prescribe one platform's settings for every network.
Comparison criterion Click-through attribution View-through attribution Engaged-view attribution --- --- --- --- Eligible interaction A qualifying click before conversion A qualifying impression without a winning click under that system's rules A qualifying video engagement before conversion Evidence to request Click definition, timestamp, window and eligible event Impression definition, window, matching and credit priority Format-specific engagement definition, window and eligible event Common comparison error Treating every reported click as the same action Treating every impression as a watched video Treating engaged views as ordinary impressions or clicks What the label cannot establish That the user would not have converted without advertising That the impression caused the outcome That attention produced incremental commercial value Appropriate initial use Understand credited response under documented rules Understand credited non-click response under documented rules Understand credited response following defined engagement
Google's App campaign documentation distinguishes click-through, engaged-view, view-through and post-install conversion windows. The post-install window concerns the time from installation to an in-app action; it is not another name for the click lookback window (Google Ads: App campaign conversion windows).
A trial start within a post-install window and an install within a click window are therefore different reporting objects. Write down both when comparing acquisition cost with downstream trial or purchase performance.
Separate interaction type, evidence method and causality
View-through is an interaction-credit category. Modeled is a description of how evidence is estimated. They are not interchangeable labels. A report can distinguish click versus view while also distinguishing matched, modeled, aggregated or otherwise restricted evidence. Request those distinctions where the provider exposes them; record unavailable detail instead of filling the gap with assumptions.
AppsFlyer's documentation describes view-through attribution using both device-ID matching and probabilistic modeling, subject to supported partners. It also notes that some partners restrict user-level data. That is provider-specific documentation, not permission to obtain identifiers or use a matching technique in every environment (AppsFlyer: Measure view-through engagements).
A click-only report does not solve the causal problem. Someone can click an ad while already intending to install. Conversely, a person can encounter useful advertising and later install without clicking. Attribution establishes credit under a rule; deciding what would have happened without advertising needs a counterfactual.
Google describes Conversion Lift as a controlled comparison between groups exposed and not exposed to advertising. Eligibility and study feasibility must still be checked for the proposed account and intervention (Google Ads: About Conversion Lift). For design choices and uncertainty, use the separate incrementality-testing guide.
Build a credit register before comparing totals
Create one row per report and conversion object. A practical register contains:
Decision: the spend, reporting or partner decision the comparison will inform.
Population: app, operating system, territory, acquisition versus re-engagement, and user eligibility.
Outcome: first launch, activation, trial, purchase or another precisely defined event.
Interaction: click, view, engaged view or a documented combined total.
Window: interaction-to-conversion and any post-install event window.
Method and coverage: matching, modeling, aggregation and unavailable detail.
Credit scope: one network's report, an MMP's cross-network allocation or another projection.
Time basis: time zone, interaction date, conversion date, cohort date and extraction time.
Use: reporting comparison, optimization input, contractual fee calculation or commercial evaluation.
Keep the original exports and configuration screenshots. Record the extraction time because delayed reporting can make a later export different even when nobody changes a setting.
The register should make an important limitation visible: two rows can share a window while covering different events or populations. Matching a day count is useful, but does not establish equivalence. The broader mobile app attribution architecture explains which system should own each business question.
Verify platform scope without copying a default table
Google: distinguish video engagement from an impression
Google's engaged-view documentation defines engagement differently across video formats. It also explains that engaged-view conversions can appear in the Conversions and All conversions columns, and describes segmentation by ad event type for comparison with click-through conversions (Google Ads: Engaged-view conversions).
Request the actual conversion action, campaign type, format and report breakdown. Do not take a total labeled conversions and assume it is click-only. Do not apply one video's engagement threshold to every format.
TikTok: Ads Manager and the MMP are separate configurations
TikTok's Self-Attributing Network window guidance says its Ads Manager attribution setting controls TikTok conversion reporting and does not change MMP reporting. It recommends consistency to reduce discrepancies (TikTok: Attribution window on Ads Manager).
A separate, newer TikTok help page lists broader ad-group window options and distinguishes click-through, view-through and engaged view-through categories (TikTok: Attribution windows at the ad-group level). The pages have different stated contexts. They should not be flattened into a universal list of options available to every app advertiser.
For a live decision, retain the controls actually available in the relevant account, objective and ad group. Ask support to resolve a material documentation/configuration conflict before changing settings. This is more defensible than selecting whichever help-page range makes the report look strongest.
MMP: record priority and the integration boundary
AppsFlyer documents priority for eligible clicks over impressions in its view-through attribution flow. It also states that disabling view-through attribution in AppsFlyer does not disable it on a self-reporting network. A changed MMP total is therefore not evidence that the network's reporting setting changed (AppsFlyer: View-through engagements).
Its lookback documentation separately recommends matching the MMP and network windows to reduce discrepancies, while describing different rules for attribution methods and integrations (AppsFlyer: Set up lookback windows). These are AppsFlyer rules; evaluate another provider from its own documentation.
For Meta or another network, obtain the current app-campaign definitions and settings from the applicable official documentation and account. This article does not prescribe Meta defaults: public help-page access did not provide enough verifiable detail for that claim. Website-conversion announcements should not be silently generalized to app campaigns.
A fictional CPA comparison: the denominator changes first
Consider a deliberately simplified example. These are invented numbers, not client evidence or a benchmark.
A team spends $12,000 on one app campaign. An export contains 120 credited qualified activations. For this example only, the report owner has confirmed mutually exclusive interaction categories:
Credited activation category Activations --- ---: Click-through 60 Engaged-view 20 View-through 40 Combined total 120
The reported activation CPA using all three categories is $100: $12,000 divided by 120. A click-only reporting view produces $200: the same $12,000 divided by 60.
No user disappeared when the analyst changed the reporting lens. No additional spend occurred. The two calculations answer different attribution questions. Neither proves that $100 or $200 is the incremental cost of acquiring an activation.
The category sum is valid only because the example explicitly assumes exclusive buckets. In a real export, determine whether columns are exclusive, nested, overlapping, assisted or separate window comparisons before adding them. A one-day and a seven-day click column can describe overlapping populations; they are not automatically two groups of customers.
Now suppose an MMP credits 75 activations to that network. Do not label the missing 45 as fake conversions. First reconcile events, eligibility, timing, credit priority and integration coverage. The platform and MMP may assign a real event differently. A remaining unexplained gap is a question for investigation, not a license to choose the more favorable denominator.
The CPI versus CPA framework covers denominator choice. The mobile app ROAS guide addresses the additional value and spend reconciliation needed before attaching a return claim.
Use a report-only sensitivity bridge
Before touching live settings, build a comparison from available reporting views. Preserve the same spend, event, population and maturity wherever the tools permit. Show the credited outcome under each documented lens and clearly mark dimensions that cannot be aligned.
The output should answer three questions:
How much does the reported result depend on non-click credit?
Does the proposed budget decision change under the available reporting lenses?
Which unresolved difference could reverse that decision?
This is a sensitivity analysis, not a historical replay of how the platform would have optimized under another setting. The original campaign delivered under its original configuration. Filtering its report afterwards does not simulate a new campaign.
Use both counts and economics. A large percentage change on a small denominator can be commercially minor; a modest change on a substantial commitment can matter. Define the escalation threshold from the actual decision exposure rather than an invented industry discrepancy tolerance.
If the same prudent action holds across the credible views, avoid turning every difference into an engineering project. Document the limit and proceed within the agreed risk boundary. If the views support opposite allocation decisions, constrain the commitment while obtaining stronger evidence.
Triage discrepancies in a useful order
First, verify the event. Confirm that both reports refer to the same production action. A download, first launch and qualified activation should not be reconciled as identical outcomes. Check app/version scope and known event failures with the implementation owner.
Second, align eligibility and time. Separate acquisition from re-engagement, territories and operating systems. Record date conventions and allow the relevant conversion/reporting window to mature before interpreting a partial cohort as a final result.
Third, inspect credit rules. Compare interaction categories, lookback windows, priority, assists and self-reporting network boundaries. Retain settings from the period being analyzed, not merely today's configuration.
Fourth, inspect coverage. A missing breakdown may reflect restricted reporting or unsupported integration detail. Mark it unavailable. Do not construct an individual-level join that the source does not expose or that privacy requirements do not permit.
Finally, identify the remaining decision. An instrumentation fault needs correction. A reporting-definition difference needs a documented bridge. A question about incremental demand needs a feasible experiment. Buying another tool is justified only if it can supply the missing decision evidence; the Firebase versus MMP comparison addresses that procurement question.
Treat a live setting change as a separate intervention
Changing a report filter and changing a campaign or integration setting are different actions. Before a live change, the accountable platform owner should document whether it affects reporting, optimization signals, eligibility, historical comparisons or partner billing. If that boundary is unclear, resolve it first.
A useful acceptance packet contains the original exports, setting values, intended change, affected campaigns, effective timestamp, implementation owner, expected observable result and a rollback plan where the platform supports one. State what evidence would justify keeping or reversing the change.
Avoid changing attribution, creative, bid strategy, event definitions and audience scope together unless the intervention explicitly requires that bundle. Otherwise, a later performance movement cannot be usefully associated with the reporting change.
Historical comparability also matters. Preserve the old definition in the reporting record. A chart that joins two attribution regimes without annotation can manufacture apparent growth or deterioration. Describe the break and compare suitable cohorts rather than silently rewriting the previous month's business narrative.
Choose among four next actions
Evidence state Proportionate next action --- --- Definitions differ, but the bounded budget decision remains stable Keep delivery configuration; document comparable reporting views and limitations Event receipt, mapping or scope is wrong Assign a correction owner and acceptance check before using the metric A proposed setting change has a clear mechanism and measurable acceptance condition Approve one bounded change with preserved baseline and review point Allocation depends on whether the advertising caused additional outcomes Evaluate a feasible incrementality study; do not substitute click-only attribution for it
These are editorial decision criteria, not a claim about a proprietary Sharply Labs process. There is no universal winner between click and view credit, and no defensible rule that every view-through conversion should always count or always be discarded.
What to ask an agency before accepting the report
Ask the partner to reproduce one material result from its source export. Request the credit register, window settings, event definition and time basis. Ask which totals can be added, what remains restricted and whether the reported result would support the same decision under a click-only lens.
Then ask what evidence would change the recommendation. A sound answer can identify an event repair, a reporting clarification, a bounded setting test or a causal study. It should not require treating the partner's preferred dashboard as unquestionable.
Keep contractual calculations explicit too. If fees depend on attributed outcomes, identify the report, windows, eligible events and treatment of later corrections in the agreement. Reporting documentation does not settle commercial terms or replace legal review.
Bring the disagreement to the right conversation
For app teams comparing materially different acquisition reports, Sharply Labs' mobile app growth service and paid-acquisition work provide an appropriate starting point for scoping the evidence problem.
Bring redacted platform and MMP exports, the relevant settings and the budget decision you need to make to the initial 15-minute conversation. The starting output is a prioritized agenda: which definitions need alignment, which evidence gaps matter and who should own the next check. Detailed reconciliation, account access, implementation and experiments require a separately agreed scope.
The conversation does not promise complete attribution, a channel winner, lower CPA, improved ROAS or a positive lift result. Its purpose is to make the next decision explicit before another report total becomes permission to spend.