Which is more accurate, the ad network or the MMP?

Neither is more accurate in general; they answer different questions. The network counts every install it touched by its own rules, while the MMP gives each install to one source across all networks.

Why shouldn't I add up installs from every network?

Because each network claims installs within its own windows and cannot see the others, so the same install is often counted more than once. Only a tool that sees every network can de-duplicate.

How big a gap between Meta and my MMP is acceptable?

No platform publishes a standard. Align windows, time zone and date basis first, then track the remaining gap per network and investigate when it changes suddenly.

Should I reconcile revenue or installs?

Installs and event counts first. Google discourages reconciling on values because refunds, currencies and single missing purchases distort revenue.

What should I use for iOS campaigns?

Use SKAN or AdAttributionKit cohorts and networks' aggregate reporting. User-level iOS numbers are limited by ATT, so they will run lower than aggregate ones.

What is mobile attribution?

Mobile attribution is the process of crediting each app install or in-app event to the ad, link or channel that drove it. A mobile measurement partner (MMP) does this across all your channels so each install is counted once.

Ad network vs MMP, which install number to trust

Short answer: use your MMP for cross-channel totals and budget splits, because it counts each install once. Use each ad network's own numbers for optimisation inside that network. On iOS, read paid results from SKAN, AdAttributionKit and network aggregates. No single tool is wrong; each answers a different question.

Why no two tools agree

Source of truth: the one report a team agrees to use for a given decision, so that numbers are not argued over each time.

Every tool measures installs with its own rules: its own windows, its own definition of an install, its own time zone and date basis, and its own view of which ads the user saw. Google says discrepancies between Google Ads and third-party reports are expected. Meta says MMP results may differ, typically because of attribution windows.

Which number should I use for budget decisions?

Use your MMP for budget splits between networks, because it is the only tool that sees every network and gives each install to one source. Network numbers overlap, so adding them together overstates installs.

What each tool is good for

Tool	Best used for	Why its number differs
MMP	Cross-channel totals, budget splits, cohort revenue	One winner per install, its own windows
Meta Ads Manager	Optimising inside Meta	Own windows, view-through, AEM modelling on iOS
Google Ads	Optimising inside Google	Play downloads, reinstalls, Every counting, interaction-day reporting
TikTok Ads Manager	Optimising inside TikTok	Own windows and fixed time zone, per its help centre
GA4 or Firebase	Product analytics, Google paid	Google-only paid crediting, reinstalls, 30-day first_open lookback
SKAN or AdAttributionKit	iOS paid results without ATT	Aggregate, delayed, privacy tiers

The reconciliation checklist

Before treating any gap as a fault, line up these five things:

Same date range and data age. Leave out the most recent two to three days; Google, Meta and Apple all document delays.
Same time zone. Meta defaults new ad accounts to Pacific time; Google uses the account's local time zone.
Same date basis. Use conversion time in Meta and the Conversions (by conv. time) column in Google.
Same windows. Compare each network's click and view windows with your MMP's.
Same definition. First opens, not downloads; reinstalls counted the same way; One, not Every.

Then compare counts before values. Google's help centre says reconciling on values is highly discouraged, because single missing or refunded purchases move revenue a lot.

How to read the remaining gap

Once the five items match, what is left usually comes from things that cannot be aligned:

Cross-network overlap. Each network claims installs it touched; the MMP picks one. See installs claimed by two networks.
View-through claims that a last-click MMP gives to a later click elsewhere.
iOS privacy limits. Without ATT permission the IDFA is all zeros, so user-level iOS numbers will be lower than aggregate ones. See iOS attribution lower than Android.
Modelled results. Meta says it may model iOS results where reporting is incomplete.

Track this remaining gap per network over time. A stable gap is a known difference; a sudden change is worth investigating with the Meta vs MMP troubleshooter or the Google installs troubleshooter.

A simple reporting policy

Weekly business review: MMP installs and revenue by channel.
Campaign optimisation: each network's own dashboard, inside that network.
iOS: SKAN or AdAttributionKit cohorts by install week, labelled as incomplete until all postbacks are due.
Product metrics: GA4 or Firebase.
Disputes: look up individual installs in the MMP and read their touchpoints.

MMPs such as Linkrunner, Tenjin or Kochava give each install a single source across all connected networks, which is what makes them the usual source of truth for budget splits.

Where Linkrunner fits: Best for mobile-first teams running paid UA on Meta, Google and TikTok that want attribution, deep linking and SKAN in one platform. It is deterministic-first with no fingerprinting, and pricing is pay-as-you-go from $0.010 per attributed install after 25k free installs.

Attribution windows explains the window settings behind most gaps.
Self-attributing networks explains why network totals overlap.