Short answer: for most mobile apps, last-click is the right starting model. It matches short install journeys, it is what ad network postbacks expect, and it is easy to explain. Move to a data-driven multi-touch model once you have high conversion volume, several channels competing for the same users, and someone who can interpret the output. Match the model to your vertical's real journey, then prove it with a holdout test.
You're tracking 50k installs a month across Meta, Google and TikTok. CPI looks healthy. But when you ask which campaigns drive paying users, you get three different answers depending on which report you pull. That isn't a data accuracy problem. It's a model selection problem.
Most teams pick a model once during MMP setup and never revisit it. That matters more than it sounds: a ₹15k-a-month AppsFlyer setup with the wrong model will give you worse decisions than a far cheaper per-install platform with the right one.
What is an attribution model?
An attribution model is the rule that decides which marketing touchpoint gets credit when a user installs and converts. Users rarely see one ad and install: they see a Meta ad on Monday, a Google ad on Wednesday, tap an influencer link on Thursday and install on Friday after searching your brand name.
Two settings work together:
- The attribution window decides which touchpoints are eligible, for example clicks in the last 7 days.
- The model decides how credit is split between those eligible touchpoints.
The most common model in mobile is last-touch attribution, usually called last-click. Multi-touch attribution is the umbrella term for every model that shares credit across several touchpoints.
The six attribution models in brief
- Last-click: 100% of the credit to the last click before install. Simple, fast and aligned with how ad networks receive postbacks.
- First-click: 100% to the first touchpoint. Rewards the channels that create awareness rather than those that capture it.
- Linear: equal credit to every touchpoint in the window.
- Time-decay: more credit to touchpoints closer to the install. For example, the last day might earn half the credit and earlier days progressively less.
- Position-based: 40% to the first touch, 40% to the last, and 20% split across the middle.
- Data-driven: compares the paths of users who converted with those who did not, then assigns credit by each touchpoint's measured contribution. A Markov "removal effect" model, for example, asks how many conversions would be lost if a channel disappeared.
Attribution model by vertical
The right model follows how long and how complex your users' journey is. Use this as a starting point, then validate it against your own data.
| Vertical | Typical journey | Model to start with | Check before you trust it |
|---|---|---|---|
| Gaming | Minutes to hours, often one playable ad | Last-click | Most installs (70% or more) land within 6 hours of the last click |
| Fintech | Days to weeks of research and trust-building | First-click, read alongside last-click | Users average four or more ad touches before install |
| Subscription (fitness, meditation, productivity) | Several touches before a paid commitment | Linear | Converters have far more touches than non-converters |
| Travel and mobility | Early awareness, intense comparison in the final days | Time-decay | Conversion probability rises sharply close to booking |
| eCommerce and fashion | Discovery, browsing, then a promotion prompt | Position-based | First and last touches come from distinct channels |
| High-volume D2C | Many channels and complex, overlapping paths | Data-driven | At least 10k conversions a month for stable results |
Two warnings apply across the table:
- Gaming teams should not add complexity they cannot use. If the real journey is one touchpoint, last-click is accurate, and it keeps creative testing cycles to a day or two.
- Data-driven models need volume, not spend. Below about 10k monthly conversions, week-to-week credit swings are mostly noise.
For vertical metrics beyond attribution, see the Metrics That Matter editions for gaming, fintech, travel, eCommerce and D2C brands.
How iOS privacy changes the picture
Any multi-touch model needs a user-level path of touchpoints. On iOS, that path is often not available:
- ATT: users who decline Apple's App Tracking Transparency prompt share no IDFA, so their touchpoints cannot be joined across apps and networks.
- SKAN and AdAttributionKit: Apple's privacy frameworks return aggregated, delayed postbacks that credit a single winning ad interaction per install. AdAttributionKit is now the primary path over raw SKAN from iOS 18.4. Neither exposes the full journey.
- What that leaves: multi-touch models run on deterministic, consented data, such as Android traffic, iOS users who opted in, and logged-in first-party journeys. Treat non-consented iOS reporting as campaign-level, last-touch style data.
Our guide to AdAttributionKit vs SKAN covers what each framework reports.
When simple models beat complex ones
- Data quality comes first. A multi-touch model splitting credit across five touchpoints amplifies tracking errors. If click tracking is unreliable, last-click at least points in the right direction.
- Team capacity. A two-person growth team running 15 campaigns needs clear signals about what to scale and what to cut. Last-click gives them that.
- Budget threshold. Below roughly ₹10 lakh a month in ad spend, a sophisticated model often costs more in analysis time than it saves. Simple holdout tests teach you more.
- Channel separation. If prospecting and branded search reach different users, model choice matters less. Models matter most when channels compete for credit on the same users.
For the full argument, read why last-click is fine for most growing apps and our comparison of multi-touch vs single-touch attribution.
How to check your model reflects reality
- Holdout test. Exclude 10% of your audience from your best-looking channel for two weeks. If Meta remarketing claims 30% of conversions but total conversions fall only 10% without it, the model is giving it about three times the credit it earns.
- Time-lag analysis. If 90% of installs happen within three days of the first click, a model that rewards touchpoints from three weeks earlier is crediting visibility, not influence.
- Financial reconciliation. Compare CAC from the model's channel split with total spend divided by total installs. A gap above about 15% points to a tracking hole or double counting, so fix that before trusting the model.
What changes when you switch models
- Reported performance moves overnight. Moving from last-click to first-click typically makes branded search look much more expensive and prospecting cheaper. Neither campaign changed; only the credit did.
- Wait before reallocating. Give the new numbers at least two weeks to settle before moving budget.
- History rarely recalculates. Most MMPs apply a new model only from the switch date, so export historical reports first.
- Postbacks can change. If your model changes which conversions are sent to ad networks, their algorithms may need a week or two to relearn.
The MMP migration playbook covers how to keep a clean baseline through changes like this.
What to ask your MMP
- Which models are actually available, and can you see exactly how credit is assigned?
- Can windows be set per network, rather than one window for everything?
- Is multi-touch an upgrade? Some contracts include last-click only and charge more for advanced models, which keeps teams on the wrong model.
Platforms like Linkrunner include last-click by default and a data-driven multi-touch model on request, with no extra fee. Click-to-install attribution windows can be set per network from 1 to 30 days (30 days for Google and Meta and 7 days for other networks by default). The multi-touch model is a Markov removal-effect model with last click, platform-reported, multi-touch and hybrid views side by side.
For more vendor questions, see 15 questions to ask in MMP demos.
Frequently asked questions
What is the best attribution model for mobile apps? For most mobile apps, last-click is the best starting point because install journeys are short and ad network postbacks are built around it. Move to a data-driven multi-touch model once you have roughly 10k or more monthly conversions and several channels competing for the same users. Long-journey verticals such as fintech and travel benefit from reading a first-click or time-decay view alongside last-click.
Last-click vs multi-touch for apps: which should you use? Last-click is simple, fast to act on and matches how Meta and Google receive conversions, but it over-credits bottom-funnel touchpoints such as branded search and retargeting. Multi-touch shows assists and shares credit across the journey, but it needs volume and deterministic, consented data, so it cannot see non-consented iOS traffic. Most growing apps run last-click for daily decisions and check a multi-touch view monthly.
How do attribution windows interact with attribution models? The window decides which clicks are eligible, and the model decides how credit is split between them. A 7-day window with last-click gives all the credit to the most recent click in the last 7 days. A 30-day window with first-click gives it to the earliest click in the last 30 days.
Do SKAN and AdAttributionKit support multi-touch attribution? No. Both return aggregated postbacks that credit one winning ad interaction per install, without a user-level journey. Multi-touch analysis is limited to Android, consented iOS users and logged-in first-party data.
Do I need data-driven attribution if I'm spending ₹50 lakh a month? Not necessarily. Data-driven models need conversion volume, not spend. Spending ₹50 lakh across three channels for 5k monthly conversions is too little data, while the same spend producing 50k conversions across eight channels can support it.
If you are spending more time reconciling reports than acting on them, the model is only half the problem. Linkrunner shows attribution in real-time dashboards from ₹1.20 per attributed install, with every feature on every plan and multi-touch attribution available on request. Book a demo to see it on your own data.