You're tracking 50,000 installs a month across Meta, Google, and TikTok. Your dashboard shows solid CPI. But when you try to understand which campaigns actually drive paying users, your attribution model gives you three different answers depending on which report you pull. This isn't a data accuracy problem. It's a model selection problem.

Most teams pick an attribution model once during MMP setup and never revisit it. They use whatever their platform defaults to, usually last-click, and assume it's showing them reality. Then six months later, they're burning ₹10 lakh a month on channels their model says are working, but their revenue data tells a different story.

The uncomfortable truth is this: your attribution model choice matters more than which MMP you use. A ₹15,000/month AppsFlyer setup with the wrong model will give you worse decisions than a ₹0.8/install platform with the right one. Here's how to match your model to your app's actual user journey, not your vendor's default settings.

Why Attribution Model Choice Matters More Than Your MMP

An attribution model is the rule set that decides which marketing touchpoint gets credit when a user installs your app and converts. That might sound straightforward, but users rarely see one ad and immediately install. They see your Meta ad on Monday, your Google ad on Wednesday, click an influencer link on Thursday, and install Friday morning after searching your brand name.

Which touchpoint deserves the credit? Which channel should you scale? Which creative should you duplicate? Your attribution model makes that decision automatically, thousands of times a day, directly influencing where you allocate your next ₹1 crore in ad spend.

The problem is that most models were designed for web e-commerce, not mobile apps. They assume certain user behaviours that don't exist in your vertical. A gaming app with 2-minute consideration windows behaves nothing like a fintech app where users research for 2 weeks before downloading. Using the same model for both guarantees one of them is getting bad data.

Here's what typically happens. Teams use last-click attribution because it's the default. It works fine initially. Then they start running brand campaigns, retargeting, and influencer partnerships. Suddenly, generic brand search gets massive credit because it's always the last click before install. They scale brand spend, CPI drops temporarily, but new user acquisition actually slows. The model was rewarding bottom-funnel touchpoints that captured demand rather than created it.

The fix isn't switching to a "better" model. It's matching your model to how users actually move through your funnel in your specific vertical. That match determines whether your attribution data helps you make good budget decisions or accidentally optimises you into a local maximum where CAC looks great but growth flatlines.

Last-Click Attribution: Best for Gaming Apps

Model Logic: Credits the last marketing touchpoint a user clicked before installing.

When This Works: Gaming apps with extremely short consideration windows, usually measured in hours not days.

Mobile games typically get discovered through social feeds. A user sees a playable ad, decides in 15 seconds whether the game looks fun, clicks, and installs immediately. If they don't install within an hour, they probably won't install at all. There's rarely a multi-day research phase or comparison shopping.

In this environment, last-click attribution reflects reality accurately. The ad a user clicked immediately before installing is genuinely the ad that drove the decision. Earlier touchpoints weren't building consideration, they were failed attempts that didn't convert.

Last-click also makes your creative testing cycles much faster. When you launch three new playable ads on Monday, you'll see by Tuesday which one is driving quality installs based on D1 retention. You can kill underperformers and duplicate winners within 48 hours. More complex models add lag that hurts your iteration speed.

Validation Check: Pull your click-to-install time distribution. If 70%+ of your installs happen within 6 hours of the last click, last-click attribution matches your reality. If users take 3-7 days to convert, you need a different model.

Common Mistake: Gaming marketers sometimes avoid last-click because it "ignores the customer journey". But if your actual customer journey is one touchpoint, ignoring non-existent complexity is the correct choice. Don't add sophistication your data can't support.

Dashboard Requirements: With last-click, you need fast creative-level visibility. You're making daily decisions about which ads to scale, so your MMP needs to surface click-through rate, install rate, and D1 retention by creative within 24 hours. Legacy platforms often batch this data overnight. Modern setups like Linkrunner show it in real-time, letting you pause underperforming creatives before burning another ₹50,000 on them.

For a detailed breakdown of gaming-specific metrics beyond attribution, see Metrics That Matter: Mobile Gaming Edition.

First-Click Attribution: Best for Fintech Apps

Model Logic: Credits the first marketing touchpoint in the user's journey, regardless of what came later.

When This Works: Fintech, banking, and high-consideration apps where brand trust and initial awareness drive the majority of conversion value.

Fintech users don't impulsively install payment apps or investment platforms. They research for days or weeks. They read reviews, compare features, check regulatory approvals, and often discuss with family before downloading. The first touchpoint that introduces them to your app and establishes initial trust is disproportionately valuable, even if they click three other ads before finally installing.

First-click attribution rewards the channels and campaigns creating new awareness rather than those capturing existing demand. This prevents the common scenario where you overfund generic search terms (which convert well because users already know your brand) and underfund the content and display campaigns that actually built that awareness.

Why This Matters for Budget Allocation: A typical fintech growth team might see their branded search campaigns showing a ₹200 CAC while their educational content campaigns show ₹800 CAC under last-click attribution. They naturally scale the cheap channel and cut the expensive one. Six months later, branded search volume drops because nobody's creating awareness anymore. First-click attribution would have shown that the educational campaigns were actually generating ₹300 CAC users, they just took longer to convert.

Validation Check: Look at your multi-touch data. If users average 4+ ad interactions before installing and your install rate from first-exposure is significantly higher than from repeat exposures, first-click is appropriate. If users install on first sight or you don't have significant repeat exposure, you need a different model.

Technical Consideration: First-click requires your MMP to track the first touchpoint per user accurately over weeks. Many platforms expire attribution windows at 7 days, which breaks first-click for high-consideration apps. Verify your MMP can maintain first-touch data for 30+ days and handle cross-device scenarios where users research on desktop and install on mobile.

For detailed fintech measurement beyond attribution models, see Metrics That Matter: Fintech Edition.

Linear Attribution: Best for Subscription Apps

Model Logic: Distributes credit equally across all touchpoints in the user journey.

When This Works: Subscription apps (meditation, fitness, productivity) where multiple touchpoints contribute roughly equally to conversion.

Subscription apps often require users to understand the value proposition before committing to recurring payments. A user might see your Instagram ad highlighting one feature, click through to learn more but not install. Later they see a YouTube pre-roll explaining your pricing model. Then a friend shares your app. Then they see a retargeting ad reminding them. Finally they install and subscribe.

Which touchpoint "caused" the subscription? In this scenario, removing any single touchpoint probably reduces conversion likelihood. The Instagram ad created awareness. The YouTube ad answered pricing objections. The friend recommendation added social proof. The retargeting ad provided the final prompt. Linear attribution acknowledges this reality by crediting all touchpoints.

When This Fails: If you have a clear hero channel that drives 70%+ of conversions, linear attribution will underweight it and overweight ancillary touchpoints. You'll see artificially inflated performance from channels that are actually just along for the ride.

Validation Check: Analyse users who converted versus those who didn't. If converters have significantly more touchpoints (4+ vs 1-2) and no single touchpoint is dramatically more common, linear attribution fits your funnel. If most conversions happen with 1-2 touchpoints, use last-click instead.

Budget Implications: Linear attribution makes scaling decisions more conservative. Every channel gets partial credit, so no channel looks dramatically better than others. This prevents you from over-concentrating spend, but it can also prevent you from aggressively scaling clear winners. You'll need to supplement attribution data with holdout tests to validate incrementality.

Dashboard Requirements: Linear attribution generates more complex reports since every campaign touches multiple users across multiple stages. You need cohort-level views showing how channel mix changes over time. Platforms like Linkrunner surface these multi-touch cohort cuts without requiring manual CSV exports and pivot tables.

Time-Decay Attribution: Best for Travel Apps

Model Logic: Credits touchpoints based on time proximity to conversion, with recent interactions weighted more heavily than older ones.

When This Works: Travel, mobility, and event-based apps where recency matters but early awareness still contributes.

Travel purchase behaviour follows a specific pattern. Users might see your hotel booking app months before their trip, establishing awareness. As travel dates approach, they start actively comparing options. In the final 48 hours before booking, they might click multiple ads and check prices several times. The ads seen in that final 48-hour window are clearly more influential than the awareness ads from three months ago, but those early touchpoints still mattered.

Time-decay attribution handles this by giving 50% credit to touchpoints in the last day, 30% to the previous 3 days, 15% to the previous week, and 5% to everything earlier. This prevents generic brand searches from getting full credit while still acknowledging that earlier touchpoints built the consideration that made those searches happen.

Decay Curve Configuration: The critical decision is how quickly credit decays. For hotel bookings with 1-2 week consideration windows, a decay half-life of 3 days works well. For flight bookings with potentially month-long planning periods, you might use 7-10 days. Your MMP should let you configure this curve rather than forcing you into fixed windows.

Validation Check: Plot your click-to-install time distribution. If it shows a clear pattern where conversion probability increases dramatically in the final days but earlier touchpoints still show conversion rates above baseline, time-decay is appropriate. If conversion probability is flat or only the last click matters, use a simpler model.

Reporting Complexity: Time-decay creates interpretation challenges. A campaign showing ₹500 attributed CAC might actually be delivering ₹300 CAC users who convert quickly or ₹800 CAC users who convert slowly. You need cohort analysis showing conversion timing alongside attribution data. Without this, you'll accidentally scale slow-converting campaigns thinking they're efficient.

For travel-specific measurement considerations, see Metrics That Matter: Travel & Hospitality Edition.

Position-Based Attribution: Best for eCommerce Apps

Model Logic: Assigns 40% credit to first touch, 40% to last touch, and splits remaining 20% across middle touchpoints.

When This Works: eCommerce apps where initial brand discovery and final conversion prompts are both critical, but middle touches matter less.

eCommerce user journeys often follow this pattern: A user discovers your brand through a Meta carousel ad (first touch matters—they didn't know you existed). They browse but don't buy. Over the next week, they see several mid-funnel ads that keep your brand in consideration but don't trigger immediate action. Finally, they see a promotion or cart abandonment reminder that drives the actual purchase (last touch matters—it provided the conversion prompt).

Position-based attribution acknowledges that beginning and end carry special importance. The first touch created the opportunity. The last touch closed the deal. Middle touches were maintenance, not drivers. This prevents you from over-investing in mid-funnel retargeting that looks efficient under last-click but isn't actually creating new demand.

When This Breaks Down: If your business relies heavily on sequential nurture (users need to see educational content, then product features, then social proof in that specific order), position-based attribution won't capture this correctly. You'll need a more sophisticated data-driven model that can detect sequential dependencies.

Category Fit: This model works particularly well for fashion, beauty, and lifestyle eCommerce where brand discovery matters, mid-funnel is primarily maintenance, and promotion timing drives conversion. It works poorly for commodity eCommerce where price comparison is the primary driver and brand matters less.

Technical Implementation: Position-based attribution requires your MMP to reliably identify first and last touches, which means maintaining user-level tracking over your full conversion window. If your window is 30 days but your MMP expires cookies at 14 days, your first-touch identification breaks and the model becomes unreliable.

For comprehensive eCommerce attribution considerations, see Metrics That Matter: eCommerce Edition.

Data-Driven Attribution: Best for High-Volume D2C

Model Logic: Uses machine learning to analyse which touchpoints actually increase conversion probability, then assigns credit proportionally.

When This Works: High-volume D2C apps with 50,000+ monthly conversions, diverse channel mix, and complex customer journeys.

Data-driven attribution (also called algorithmic or machine learning attribution) doesn't use predetermined rules. Instead, it compares users who converted against similar users who didn't, identifies which touchpoints meaningfully increased conversion probability, and assigns credit accordingly. If users who saw YouTube ads plus retargeting converted at 8% but users who only saw YouTube converted at 3%, the model infers that retargeting is highly valuable and weights it accordingly.

This approach can detect patterns that rule-based models miss. Maybe your Instagram ads don't drive many last-click conversions but users who see them convert 40% faster through other channels. Rule-based models would undervalue Instagram. Data-driven attribution would correctly weight it based on its acceleration effect.

Volume Requirements: Data-driven models need substantial data to produce stable results. Below 10,000 monthly conversions, the model can't reliably distinguish signal from noise. Your week-to-week results will swing wildly, making budget decisions impossible. Wait until you have the volume to support algorithmic learning before switching away from rule-based models.

Black Box Risk: The biggest downside is explainability. When your CFO asks why YouTube got 30% credit this month versus 22% last month, you can't point to a simple rule. The algorithm saw patterns in the data that shifted the weighting. This makes financial forecasting harder and can create trust issues if stakeholders don't understand how algorithmic models work.

Validation Process: Data-driven models require ongoing validation. Run monthly holdout tests where you deliberately exclude 10% of users from seeing specific channels, then verify the model's incrementality predictions match actual results. If the model says your branded search is worth ₹300 CAC but your holdout test shows it's closer to ₹600 CAC (because it's capturing organic demand), recalibrate.

Platform Requirements: Not all MMPs offer true data-driven attribution. Some offer "custom attribution" where you manually set weights, which isn't machine learning. Others offer data-driven models but require minimum contract values of $50,000+ annually. Platforms like Linkrunner provide algorithmic attribution at accessible price points, but verify the model's training data volume before trusting results.

For more on D2C-specific measurement approaches, see Metrics That Matter: D2C Brands Edition.

When Simple Models Beat Complex Ones

There's an industry assumption that more sophisticated attribution models produce better decisions. This is wrong surprisingly often.

Model Complexity vs Data Quality Trade-off: A sophisticated position-based model trying to assign partial credit to five touchpoints will produce worse results than a simple last-click model if your click tracking has 20% error rate. The complex model amplifies noise while the simple model at least gives you directionally correct signals. Focus on data quality before model sophistication.

Team Capacity Constraints: Multi-touch attribution requires significant analytical capacity to interpret correctly. If your growth team is two people running 15 campaigns across 5 channels, they don't have time to analyse nuanced attribution shifts. They need clear signals about what's working and what to cut. Last-click gives them that. Position-based gives them homework.

Budget Size Threshold: Below ₹10 lakh monthly ad spend, sophisticated attribution often costs more in analytical overhead than you gain in optimisation. You're better off running simple A/B tests with holdout groups than trying to perfect your attribution model. Above ₹50 lakh monthly spend, better attribution can easily save 10-15% of budget, making the investment worthwhile.

Channel Separation Principle: If your channels target completely different stages of the funnel with minimal overlap, attribution model choice matters less. Your Instagram prospecting campaigns and your branded Google search campaigns aren't competing for the same user. You can evaluate them independently using whatever model makes each one measurable. Attribution models matter most when channels compete for credit for the same users.

For a deeper exploration of when simple models work better, see Why Last-Click Attribution Is Actually Fine for Most Growing Apps.

Model Validation: How to Check if Your Model Reflects Reality

Your attribution model should make predictions you can test. Here's how to validate whether your current model reflects actual causation or just correlation.

Holdout Test Protocol: Pick your highest-spending channel showing great performance under your current attribution model. Exclude 10% of your audience from seeing ads on this channel for two weeks. If conversions from the excluded group drop proportionally to the channel's attributed share, your model is accurate. If conversions barely change, your model is giving credit to a channel that's capturing demand, not creating it.

Example: Your Meta remarketing shows ₹400 CAC under last-click attribution, representing 30% of attributed conversions. Run a holdout test. If overall conversions drop 30% when you exclude the holdout group from Meta remarketing, the model is correct. If conversions only drop 10%, Meta is getting triple credit for users who would have converted anyway through other channels.

Time Lag Analysis: Pull click-to-install timing for each channel. If your model assumes certain touchpoint priorities but the timing data contradicts this, your model is wrong. Specifically: if your model gives heavy credit to touchpoints that happen 20+ days before install but 90% of your installs happen within 3 days of first click, your attribution windows are too long and you're rewarding channels that happened to be visible, not influential.

Correlation vs Causation Check: Look at users who converted without ever seeing certain high-performing channels. If 40% of your converters never saw your "best-performing" Meta campaign, that campaign might be showing in front of users with high natural conversion intent rather than creating that intent. Your attribution model is rewarding correlation, not causation.

Sequential Pattern Analysis: For multi-touch models, check whether touchpoint sequence matters. If users who see Channel A then Channel B convert at the same rate as users who see Channel B then Channel A, your position-based or time-decay model is probably overcomplicating things. If sequence clearly matters (A-then-B converts at 8% vs B-then-A at 3%), your model needs to capture this, which most standard models don't.

Financial Reconciliation Test: Calculate CAC using your attribution model's channel breakdown, then compare to total spend divided by total installs. If there's more than 15% difference, you have either a tracking gap (clicks/installs not being captured) or a model problem (double-counting or missed conversions). Fix tracking before trusting model outputs.

Switching Models: What Changes in Your Dashboard

Changing attribution models doesn't change reality, but it dramatically changes which campaigns appear to be working. Here's what to expect.

Immediate Metric Shifts: When you switch from last-click to first-click attribution, your brand search campaigns will show immediate CAC increases (often 2-3x) because they lose credit they were previously getting for capturing demand. Your prospecting campaigns will show CAC decreases (often 30-50%) because they start getting credit for demand they created but didn't close. Neither campaign actually changed performance. You just changed which one gets credit.

Budget Reallocation Pressure: Your instinct will be to immediately shift budget away from campaigns showing worse performance under the new model. Resist this for at least two weeks. Some of what looks like performance change is just noise from small sample sizes stabilising under different calculation rules. Wait for statistical significance before making major changes.

Historical Data Breaks: Most MMPs don't retroactively recalculate historical data when you change models. Your January data uses last-click, your February data uses first-click, and your year-over-year comparisons are now meaningless. Before switching, export historical data under your current model so you maintain some comparison baseline.

Team Training Requirements: Different models require different analysis approaches. Last-click attribution lets you make quick daily optimisation calls. First-click attribution requires you to evaluate campaigns over weeks, not days. Time-decay requires you to analyse conversion lag alongside cost. Switching models means retraining your team on how to interpret the new data correctly.

API and Postback Implications: If you send conversion data to ad platforms via postbacks, changing your attribution model might change which conversions get sent. This affects platform algorithms. For example, switching from last-click to first-click might start sending conversions to your prospecting campaigns that previously went to retargeting. The ad platform will shift budget toward prospecting, potentially destabilising your account for 1-2 weeks while algorithms relearn.

For a comprehensive guide to managing these transitions, see The Complete MMP Migration Playbook.

Implementation in Modern MMPs

Your attribution model runs inside your MMP. The question is whether your platform makes model selection accessible or treats it as an enterprise feature locked behind sales calls.

Configuration Accessibility: Legacy MMPs often require support tickets to change attribution models, with 48-72 hour implementation timelines. Modern platforms let you switch models in dashboard settings and see results immediately. This matters because you need to test which model matches your reality, and testing requires quick iteration.

Model Transparency: Some platforms show you attribution model logic clearly. Others treat it as a black box. Before committing to a platform, verify you can see exactly how credit is assigned. If you can't reproduce the calculation manually, you can't validate whether it's correct or debug when something breaks.

Historical Recalculation: Better platforms let you apply new attribution models to historical data, generating comparison reports showing how model changes would have affected past decisions. This lets you validate new models before switching live traffic to them. Without this feature, you're flying blind when you change models.

Multiple Model Support: Your attribution model isn't universal. You might want last-click for gaming campaigns and first-click for brand campaigns within the same app. Verify your MMP supports multiple active models with clear segmentation. Otherwise you're stuck choosing one model for everything, which guarantees some campaigns get measured incorrectly.

Cost Transparency: Some platforms charge separately for advanced attribution models. Your contract might include last-click but require upgrades for time-decay or data-driven attribution. This creates perverse incentives where you stick with the wrong model because switching costs money. Platforms like Linkrunner include all attribution models at base pricing (starting at ₹0.8 per install) specifically to remove this barrier.

For questions to ask during vendor evaluation, see 15 Questions to Ask in MMP Demos.

Key Takeaways

Attribution model choice directly affects where you allocate budget. Get it wrong and you'll systematically overfund channels that look good in reports but don't actually drive incremental conversions.

Match your model to your vertical's actual user behaviour. Gaming apps with instant decisions need last-click. Fintech apps with week-long research need first-click. Subscription apps with multi-stage journeys need linear or position-based models.

Simpler models beat complex ones when you have limited data volume, small teams, or low channel overlap. Don't adopt multi-touch attribution because it sounds sophisticated. Adopt it when your data supports it and your team can interpret it correctly.

Validate your model with holdout tests. If pausing a channel that shows great attributed performance barely affects actual conversions, your model is rewarding correlation not causation. Fix this before compounding budget allocation mistakes.

Switch models carefully. Your team needs retraining, your historical comparisons break, and ad platform algorithms need time to adjust. Make the switch during low-stakes periods, not right before major launches.

Your MMP's attribution flexibility matters as much as its tracking accuracy. A platform that locks you into one model or charges extra for model changes will constrain your analysis options when your business evolves.

Frequently Asked Questions

Which attribution model do most apps use? Last-click attribution is most common, used by approximately 60-70% of mobile apps primarily because it's the default setting in most MMPs. However, "most common" doesn't mean "most appropriate". Many apps would get better measurement from alternative models but never revisit the default choice.

Can I use different attribution models for different campaigns? Yes, and you should. Your prospecting campaigns might need first-click attribution while your retargeting campaigns need last-click. Verify your MMP supports segmented attribution rules. Not all platforms do, particularly at lower pricing tiers.

How do attribution windows interact with attribution models? Attribution windows define the time period during which a click can receive credit. Attribution models define how credit is divided among eligible clicks within that window. You need to configure both correctly. A 7-day attribution window with last-click gives all credit to the most recent click within 7 days. A 30-day window with first-click gives all credit to the first click within 30 days.

Does iOS SKAN support multi-touch attribution? No. SKAN provides only aggregated, anonymous conversion data without user-level tracking. This means multi-touch attribution models can't function for iOS 14.5+ traffic. You're limited to install-level attribution using postback data. This makes Android data even more valuable since it still supports proper multi-touch measurement.

How often should I review my attribution model choice? Quarterly is appropriate for most apps. Review whenever you enter a new vertical, launch major new campaigns, or change your channel mix significantly. If gaming represents 20% of your app's revenue and suddenly becomes 60%, your attribution model might need adjustment.

What's the best attribution model for organic installs? None. Attribution models handle marketing touchpoints. Organic installs by definition have no paid marketing touchpoint to attribute. Track them separately in your reporting and don't try to force them into attribution models. Some platforms try to attribute organic installs to past marketing touches, which creates misleading data about marketing efficiency.

Do I need data-driven attribution if I'm spending ₹50 lakh/month? Not necessarily. Data-driven attribution needs conversion volume, not just spend. If you're spending ₹50 lakh across 3 channels driving 5,000 monthly conversions, you don't have enough data. If that same spend drives 50,000 monthly conversions across 8 channels, data-driven attribution might meaningfully improve your optimisation decisions.

If your current attribution model is giving you directionally correct answers but requires manual CSV exports, pivot tables, and weekly reconciliation meetings to generate actionable insights, you're spending analytical capacity on reporting overhead rather than optimisation decisions.

Modern MMPs surface attribution data in real-time dashboards with cohort cuts, creative-level breakdowns, and channel comparison views that work out of the box. Linkrunner provides this starting at ₹0.8 per attributed install with all attribution models included, no additional fees for advanced features. If you're ready to stop fighting your reporting stack and start making faster budget allocation decisions, request a demo from Linkrunner.