---
title: "Mobile App Churn: 6 Behavioural Signals That Predict It 2 Weeks Before"
description: "By the time a user uninstalls, the decision was made two weeks ago. Here are the 6 behavioural signals that predict it and how to score them."
publishedAt: "2026-07-15T14:17:00.000Z"
updatedAt: "2026-07-15T14:17:00.000Z"
author: "Amar Rawat"
categories: []
canonical: "https://www.digia.tech/post/mobile-app-churn-6-behavioral-signals-predict-2-weeks-before"
---

# Mobile App Churn: 6 Behavioural Signals That Predict It 2 Weeks Before

**TL;DR**

- By the time a user uninstalls, the decision has been made for roughly two weeks. Session frequency typically starts declining 10 to 28 days before a subscriber cancels.
- The intervention window exists, but only for teams that are instrumented to see it.
- The article covers why reactive churn campaigns underperform and the evidence behind the two-week window.
- It covers six specific behavioural signals that precede churn: session frequency decline, session depth reduction, feature reversion, in-app dismissal rate increase, abandoned intent, and absence of milestone behaviour in reactivated users.
- It covers how to build a scoring model from these signals without machine learning.
- It covers what campaign to run at each risk level.

Most churn campaigns fire at the moment of maximum futility. A user has not opened the app in 21 days. The system flags them as churned. A win-back push goes out. [Recovery rates for full uninstall win-back sit at 1 to 5%](https://mwm.ai/glossary/winback-campaign), a fraction of what the same message would achieve if it had been sent two weeks earlier, while the user was still installed, still occasionally opening the app, and still recoverable through the product itself rather than through an external campaign trying to undo a decision that has already been made.

The problem is not the campaign. The problem is the timing. [Churn is a staged process, not a sudden event. Users move from Active to Drifting to At-Risk before cancelling or uninstalling, and each stage is visible in behavioural data if you track the right metrics](https://www.airbridge.io/en/blog/how-to-spot-users-who-are-about-to-churn). The uninstall is the final, visible step in a process that started weeks earlier and left a specific, measurable trail. Teams that only look for the uninstall event are looking at the wrong end of the timeline.

This article covers what that trail looks like, six specific signals within it, and how to build a detection system that catches the decision while it is still being made, not after it has already been finalised.

## Why Reactive Churn Campaigns Underperform

A reactive churn campaign is triggered by inactivity: no session in N days, so send a re-engagement message. The problem with this trigger is that it fires after the user has already stopped finding value in the product, not while that process is happening. By the time 14 or 21 days of inactivity has accumulated, the user has already replaced the app's function with something else, formed a negative or neutral association with it, or simply moved on.


![ A user journey funnel visualization showing customer progression through key app events. The flow begins with **App Installed (510 users)**, where **67.3%** proceed to **App Login (343 users)**. From there, the journey branches: **33.2%** continue to **Feature Viewed (114 users)**, while **66.5%** complete a **Form Submitted (228 users)** event. A callout labeled **"Watch replays of users dropping"** highlights session replay functionality for analyzing user drop-offs. The diagram uses green flow paths, white event cards, and percentage labels to illustrate conversion through the customer journey.](https://cdn.sanity.io/images/53loe8pn/production/b3f1b8a6b1566271aea4e6fe63dd1dc692ea0b20-1672x941.png?w=1200&fit=max&auto=format)


[Behavioral signals catch the friction that caused churn, often weeks before any statistical risk score moves. Churn rarely starts with a decision. It starts with friction that builds over weeks, often before a customer contacts support or cancels](https://www.fullstory.com/blog/predicting-customer-churn/). A churn model that only reads inactivity is reading the outcome of the friction, not the friction itself. The intervention at that point is not solving the underlying problem. It is attempting to reverse a conclusion the user has already reached.

The conversion rate data reflects this timing failure directly. [Win-back campaigns targeting users after full uninstall recover 1 to 5% of the targeted audience, while subscription win-back with a discount, targeting users who are still installed but have cancelled billing, recovers 5 to 15%](https://mwm.ai/glossary/winback-campaign). The difference between these two numbers is not campaign quality. It is proximity to the moment the decision was still reversible. A user who has uninstalled has closed a door. A user who is drifting, still installed, still occasionally opening the app, has a door that is only partially closed.

The correct target for a churn intervention is not the user who has already left. It is the user who is in the process of leaving, while the behavioural evidence of that process is still visible and while the app itself, not an external campaign, still has the opportunity to change the trajectory.

## The Two-Week Window: What the Evidence Shows

The two-week window is not an arbitrary framing device. It reflects a consistent pattern found across subscription and engagement research.

[Session counts typically start declining 10 to 28 days before subscribers cancel, a window large enough to intervene](https://www.airbridge.io/en/blog/how-to-spot-users-who-are-about-to-churn). Setting a session frequency baseline for each user cohort during their first 30 days, and flagging users whose weekly session count drops below 50% of that baseline, catches the decline while it is still in progress rather than after it has completed.


![A line chart titled "The Gradual Decline: 6 Weeks to Churn" illustrating how customer engagement decreases before cancellation over a six-week period. The chart compares three key metrics: Daily Logins (red), Average Session Time (blue), and Core Feature Usage (orange). All three metrics show a steady downward trend from Week 1 to Week 6, with average session time declining the most significantly. The visualization highlights progressive disengagement as users approach churn, emphasizing the importance of identifying early warning signals for customer retention.](https://cdn.sanity.io/images/53loe8pn/production/71b55869b5634a611830042f97ceb2c3b7aa3fb6-1672x941.png?w=1200&fit=max&auto=format)


[The first renewal or billing window is the highest-risk moment for subscription apps. Nearly 30% of annual subscribers cancel within the first month of their subscription](https://www.airbridge.io/en/blog/how-to-spot-users-who-are-about-to-churn), which means the behavioural decline that precedes this cancellation is compressed into an even tighter window for new subscribers than for established ones. The signal detection system needs to account for this: a 14-day window for an established user's declining engagement may need to be a 7-day window for a user in their first billing cycle.

[Gartner's 2025 Customer Success research found that organisations with automated churn early-warning systems reduce annual churn rates by about 3.1 percentage points on average compared to manual monitoring](https://prometheusagency.co/insights/predictive-churn-modelling), with the operational difference being that scores are wired into tasks and playbooks, not just calculated and reported. A churn score that exists in a dashboard without a corresponding trigger to a specific campaign at a specific threshold is not a churn prevention system. It is a churn observation system. The two-week window is only useful if the detection triggers action within it, not after it has closed.

## Signal 1: Session Frequency Decline

Session frequency decline is the earliest and most reliable of the six signals. It measures how often a user opens the app relative to their own historical baseline, not relative to a fixed threshold applied uniformly across the user base.

The correct baseline is individual, not aggregate. A user who opens a banking app twice a week has a different normal cadence than a user who opens a food delivery app daily. Comparing both users against a single frequency threshold produces false positives for the low-frequency-by-design user and false negatives for the high-frequency user whose engagement has genuinely started declining but has not yet dropped below an aggregate threshold.

[Track opens per week, not session duration. A user can open the app briefly and still be drifting. Frequency is a more forward-looking predictor than time spent in the app](https://www.airbridge.io/en/blog/how-to-spot-users-who-are-about-to-churn). This distinction matters because session duration can remain stable even as frequency declines: a user who visits less often but, when they do visit, completes their usual actions, is showing an early-stage decline that duration-based metrics would miss entirely.

**Distinguishing genuine decline from normal variance.** Weekly session counts are noisy at the individual level. A single low-frequency week is not a signal. The correct approach compares a rolling average (typically a 7 or 14-day trailing window) against the user's own baseline average from their first 30 days of use, and flags a decline only when the rolling average has been below 50% of baseline for at least two consecutive measurement periods. This two-period requirement filters out normal week-to-week variance (a busy week, a holiday, a one-off disruption to routine) and only flags decline that has persisted long enough to represent a genuine behavioural shift rather than noise.

## Signal 2: Session Depth Reduction

Session depth measures how many meaningful actions a user completes per session, not how long the session lasts or how often it occurs. A session where a user opens the app, glances at one screen, and exits is behaviourally different from a session where the same user completes three or four core actions, even if both sessions last the same 90 seconds.

The threshold that matters is not an absolute number of actions. It is the ratio of the user's current session depth to their own historical baseline. A user whose typical session includes checking their balance, reviewing a transaction, and browsing a feature has a baseline of three core actions per session. A decline to one action per session, sustained across multiple sessions, is the depth-reduction signal, independent of whether the frequency of sessions has changed.

Session depth reduction frequently precedes session frequency decline, which is what makes it a genuinely leading indicator rather than a coincident one. A user who is losing interest in a product typically reduces what they do within each session before they reduce how often they open it. Catching depth reduction while frequency is still stable identifies the user earlier in the churn process than waiting for frequency to decline as well.

## Signal 3: Feature Reversion

Feature reversion is the pattern where a user who previously engaged with multiple features of the product contracts back to using only one core feature, abandoning the adjacent features they had adopted.

This signal is distinct from a user who never adopted multiple features in the first place. The reversion pattern specifically requires a prior period of broader engagement followed by a narrowing. A fintech app user who used to check their portfolio, review spending insights, and make transactions, but over the past two weeks has only opened the app to check their balance, is showing feature reversion. The narrowing itself is the signal, independent of the user's absolute activity level.

Feature reversion is diagnostically valuable because it often reveals which part of the product the user has lost confidence in or interest in, rather than the product as a whole. [Feature adoption breadth is one of the strongest predictors of retention, and users who adopt three or more features show meaningfully higher retention than users who use only core features](https://www.appcues.com/blog/a-guide-to-feature-adoption). The inverse holds for churn prediction: a contraction in feature breadth, especially a rapid one over a two-week window, is a leading indicator that the user's relationship with the broader product is narrowing, even while their relationship with the single remaining core feature may still appear healthy on a surface-level activity metric.

## Signal 4: In-App Dismissal Rate Increase

In-app dismissal rate measures how often a user dismisses, ignores, or swipes away nudges, surveys, tooltips, and prompts, relative to their own historical baseline dismissal rate.


![A side-by-side illustration of a cart abandonment and re-engagement workflow displayed on two smartphone screens. The left screen shows an online shopping cart with pastel-colored products, payment details, and a prominent "Checkout" button, along with a "Cart abandoned" label indicating the user exited before completing the purchase. The right screen displays a personalized messaging app notification reminding the customer about the abandoned cart, featuring the product image, a 10% discount offer, and a "Back to cart" call-to-action. The composition visually demonstrates how personalized messaging can recover abandoned carts and encourage customers to complete their purchase.](https://cdn.sanity.io/images/53loe8pn/production/4213ad1b973f0f0a67886de042785f7ca42d1edb-1672x941.png?w=1200&fit=max&auto=format)


This signal captures something the other five do not: emotional disengagement that precedes behavioural disengagement. [Click-through rate on in-app nudges is healthy in the 15 to 40% range, well above the 2 to 5% range typical for email and push](https://www.nvecta.com/blog/in-app-nudges/), which means a user's baseline engagement with in-app content is meaningfully higher than their engagement with external channels. A sustained increase in dismissal rate for this specific channel, one the user has historically engaged with at a healthy rate, is a stronger signal than declining engagement with push notifications, because it indicates the user is disengaging even from content delivered inside a session they are actively in.

The interpretation of this signal requires distinguishing genuine emotional checkout from simple notification fatigue caused by excessive volume. [52% of users who disable push notifications eventually churn from the app entirely](https://www.courier.com/blog/how-to-reduce-notification-fatigue-7-proven-product-strategies-for-saas), but a portion of that churn is caused by the app's own over-messaging rather than the user's disinterest in the product. Before treating a rising dismissal rate as a churn signal, check whether the volume of nudges sent to that user has also increased over the same window. If volume increased and dismissal rate increased proportionally, the signal may reflect fatigue from over-messaging rather than genuine disengagement. If dismissal rate increased while volume stayed constant or declined, the signal is more reliably interpreted as the user checking out emotionally before they check out behaviourally.

## Signal 5: Abandoned Intent

Abandoned intent covers cart abandonment, incomplete transactions, and started-but-not-finished flows, specifically instances where the user demonstrated an intention to complete an action and then did not, without any recovery attempt in a subsequent session.

A single abandoned cart or incomplete flow is not a churn signal. Users abandon carts for reasons unrelated to their relationship with the product: a distraction, a change of mind, a payment method issue. The signal is the pattern of repeated abandonment without recovery: a user who has started the same flow, or a similar flow, on three or more occasions in a two-week window without ever completing it once.

This pattern is diagnostically distinct from feature reversion because it captures active friction rather than passive disengagement. A user showing feature reversion has stopped trying to use a feature. A user showing abandoned intent is still trying, repeatedly, and repeatedly failing to complete the action. This distinction matters for the intervention: feature reversion may call for a re-engagement message reminding the user of the feature's value, while abandoned intent calls for a friction-diagnosis intervention, understanding what specifically is preventing completion (a confusing step, a technical error, a missing piece of information) and addressing that specific point.

[A subtle tooltip nudge or a coupon code at the point of cart abandonment can increase conversion, but the same intervention repeated without addressing the underlying friction becomes ineffective by the third occurrence](https://www.plotline.so/blog/in-app-nudges-ultimate-guide). Repeated abandoned intent that survives a first-line nudge intervention is a stronger churn signal than a single abandonment, because it indicates the friction is structural rather than momentary.

## Signal 6: Absence of Session-One Milestone Behaviours in Reactivated Users

The sixth signal is specific to users who have already lapsed once and returned, either organically or through a re-engagement campaign. A significant subset of these reactivated users churn again, and the signal that predicts this second churn is visible in their very first session back.

[The moment a churned customer reactivates, they must enter a dedicated, improved onboarding path. Treat them as a new, high-risk segment](https://www.stackmatix.com/blog/win-back-campaign-strategies). A returning user who does not replay the activation pattern that retained them originally, the specific sequence of actions that constituted their first-time activation, is showing a signal that the return is unlikely to stick. If a user's original activation involved completing a first transaction within their first session, and their return session shows browsing behaviour without a completed transaction, that absence is diagnostically meaningful.

[Re-churn rate, the percentage of reactivated users who churn again within a defined period, is the most critical quality metric for any win-back programme. A high re-churn rate indicates poor segmentation or offer targeting](https://www.stackmatix.com/blog/win-back-campaign-strategies) rather than a failure of the reactivation offer itself. Tracking whether reactivated users replay their original milestone behaviour in their first return session is the earliest possible signal for whether a specific reactivation will hold or will produce a second, faster churn cycle. [Non-discounted resubscribers stay 2.6 times longer than discount-acquired ones](https://www.airbridge.io/en/blog/subscription-app-win-back-sequence), which reflects the broader pattern that reactivation quality, not just reactivation volume, determines whether the second chapter of the user relationship survives.

## Building a Churn Prediction Model From These Signals Without Machine Learning

A machine learning churn model is not necessary to act on these six signals. [Starting simple before scaling is the right sequence: launch with rule-based thresholds to test intervention impact, measure recovery lift from basic campaigns before investing in statistical or machine learning models](https://clevertap.com/blog/churn-prediction/). An event-based scoring approach, built on the six signals above, is sufficient for most teams and has the advantage of being interpretable: every score component maps to a specific, explainable behaviour, which matters when a growth team needs to explain why a specific user was flagged and design the corresponding intervention.

**The scoring structure.** Assign each of the six signals a point value when triggered for a specific user, based on how strongly that signal has historically correlated with subsequent churn in your own data (established by comparing the signal's presence in your churned cohort against its presence in your retained cohort over the preceding 90 days).

A reasonable starting weight structure, to be recalibrated against your own churned versus retained cohort data:

Session frequency decline (sustained below 50% of baseline for two consecutive periods): 25 points Session depth reduction (below 50% of baseline session actions, sustained): 20 points Feature reversion (contraction from 2 or more features to 1, sustained over 2 weeks): 20 points In-app dismissal rate increase (above baseline while nudge volume is stable or declining): 15 points Abandoned intent (3 or more incomplete attempts at the same flow within 2 weeks): 15 points Absence of milestone behaviour in reactivated users (applicable only to the reactivated segment): 25 points, replacing rather than adding to the other five for this specific segment

**Score thresholds and corresponding actions.** A cumulative score of 0 to 24 represents low risk: no intervention beyond standard lifecycle messaging. A score of 25 to 49 represents moderate risk: the user qualifies for a targeted in-app re-engagement campaign, not yet an external win-back campaign. A score of 50 to 74 represents high risk: the user qualifies for a multi-channel intervention combining in-app and push. A score of 75 or above represents critical risk: the user is treated as functionally lapsed and qualifies for the full win-back sequence, even if they have not yet crossed a formal inactivity threshold.

**Calibrating the weights to your own data.** The starting weights above are a reasonable default, not a fixed formula. [Companies establish risk thresholds, typically scoring above 70 as high risk, 40 to 70 as moderate risk, and below 40 as healthy, calibrated against their own historical churn and retention data](https://www.saber.app/glossary/churn-prediction-model). Pull your churned cohort from the past 90 days and your retained cohort from the same period. For each signal, calculate what percentage of the churned cohort exhibited that signal in the two weeks before churn, and what percentage of the retained cohort exhibited the same signal without churning. A signal that appears in 80% of your churned cohort but only 10% of your retained cohort is a strong predictor and warrants a higher weight than a signal that appears in 40% of churned users and 35% of retained users, which is a weak predictor for your specific app despite being a reasonable signal in general.

## Campaign Triggers by Signal Level

Each risk tier calls for a different campaign format, not simply an escalating volume of the same message. [Different disengagement stages need different messaging. Drifting users, still active but declining, need different messaging than at-risk users who haven't opened the app in 30-plus days](https://www.airbridge.io/en/blog/how-to-spot-users-who-are-about-to-churn).

**Low risk (0 to 24 points): standard lifecycle messaging.** No dedicated churn intervention. The user continues to receive normal product communications, feature announcements, and lifecycle nudges as any healthy user would.


![A visual representation of a multi-step re-engagement campaign workflow. On the left, a push notification reminds inactive users that they have not visited the app recently, followed by an email featuring personalized content and offers to encourage them to return. On the right, the campaign flow is shown as a sequence: Push Notification (#1 Push – Re-engagement campaign) → Wait (1 Week) → Email (#2 Email Re-engagement campaign). The diagram demonstrates how businesses use automated, multi-channel messaging to re-engage inactive users and improve customer retention.](https://cdn.sanity.io/images/53loe8pn/production/c95b657c7cb7784c7fa593d76c4ca41f61bf1bbd-1672x941.png?w=1200&fit=max&auto=format)


**Moderate risk (25 to 49 points): in-app re-engagement, not push.** Because the user is still opening the app, even if less frequently, the highest-leverage intervention is in-app: a contextual nudge that fires on their next session, surfacing the specific feature or value they have drifted away from. [In-app click-through rates of 15 to 40% substantially outperform email and push](https://www.nvecta.com/blog/in-app-nudges/), and the moderate-risk user is precisely the segment where this channel advantage matters most, because they are still in-session often enough for an in-app trigger to reach them.

**High risk (50 to 74 points): multi-channel, value-first, no discount yet.** The user's session frequency has likely declined enough that in-app alone may not reach them reliably. Add a push notification, but keep the messaging value-first: reminding the user of a specific benefit or unfinished action, not leading with a discount. [Non-discounted resubscribers stay 2.6 times longer than discount-acquired ones](https://www.airbridge.io/en/blog/subscription-app-win-back-sequence), and leading with a discount at this stage risks training users to disengage in order to receive an offer, a pattern that damages the long-term economics of the intervention even when it produces a short-term reactivation.

**Critical risk (75 or above): the full win-back sequence, discount held for later touchpoints.** [A 4-email escalation sequence achieves approximately 14.7% cumulative reactivation, and the most effective first touchpoint goes out within 24 hours, while the user is still in the mental frame of the product](https://www.airbridge.io/en/blog/subscription-app-win-back-sequence). At this tier, the discount or incentive should appear in the third touchpoint of the sequence, not the first, following the same value-first logic: if the user reactivates from touchpoints one or two without needing the discount, the campaign has not spent the incentive it did not need to spend.

## Topics Not in the Brief That Teams Should Know

**The specific risk of over-triggering on false positives.** A scoring system that flags too many users at moderate or high risk trains the growth team to ignore the alerts, the same fatigue dynamic that affects users receiving too many notifications. Calibrate the score threshold against actual intervention capacity: if the moderate-risk tier is producing more flagged users per week than the team can meaningfully act on, raise the threshold rather than lowering intervention quality per user.

**Churn-reason segmentation before campaign design.** [Not all churned users are worth pursuing. Cost-related churners and billing-failure churners respond well to win-back. "Found a better app" churners typically do not](https://www.airbridge.io/en/blog/subscription-app-win-back-sequence). Before building campaign content for each risk tier, segment by the likely churn reason inferred from the signal pattern: a user showing abandoned intent around a payment flow is likely a friction-driven risk, addressable with a UX fix and a targeted nudge. A user showing broad feature reversion with no specific friction point may be a fit-driven risk, harder to address through messaging alone.

**The compounding cost of ignoring the two-week window at scale.** [Organisations with automated churn early-warning systems reduce annual churn by 3.1 percentage points on average compared to manual monitoring](https://prometheusagency.co/insights/predictive-churn-modelling). At meaningful MAU scale, a few percentage points of annual churn reduction compounds into a materially different retained user base after 12 to 18 months, which is the argument for building the scoring system even before it is perfectly calibrated, since directional accuracy at the two-week mark outperforms perfect accuracy at the point of uninstall.

**Retraining the model as product and user behaviour evolve.** A scoring model calibrated against last year's churned cohort degrades as the product changes, new features launch, and user expectations shift. [Models require regular updates as customer behaviour patterns evolve and product capabilities change](https://www.saber.app/glossary/churn-prediction-model). Recalibrate signal weights against a fresh churned versus retained cohort comparison at least quarterly, and immediately after any major feature launch or onboarding redesign that would plausibly change what a normal usage baseline looks like.

## Key Takeaways

By the time a user uninstalls, the decision has typically been made 10 to 28 days earlier. Reactive churn campaigns that fire on inactivity are targeting a decision that has already been finalised, which is why full uninstall win-back recovers only 1 to 5% compared to 5 to 15% for subscription win-back reaching users while they are still installed.

The six behavioural signals that precede churn within the two-week window are session frequency decline (relative to individual baseline, not an aggregate threshold), session depth reduction (fewer meaningful actions per session), feature reversion (contraction from multiple features back to one), in-app dismissal rate increase (checked against nudge volume to rule out fatigue), abandoned intent (repeated incomplete flows without recovery), and absence of milestone behaviour in reactivated users (the strongest predictor of a second, faster churn cycle).

A rule-based, event-driven scoring model is sufficient to act on these signals without requiring machine learning infrastructure. Weight each signal based on its correlation with churn in your own historical data, not a generic industry weighting, and recalibrate quarterly.

Each risk tier requires a different campaign format, not an escalating volume of the same message. Low risk gets standard lifecycle messaging. Moderate risk gets in-app re-engagement, which outperforms push and email at this stage because the user is still in-session often enough to be reached there. High risk gets multi-channel, value-first messaging with the discount held back. Critical risk gets the full win-back sequence with the incentive placed at the third touchpoint, not the first.

The compounding value of catching churn two weeks earlier rather than at the point of uninstall is not marginal. A few percentage points of annual churn reduction, achieved consistently through earlier detection, produces a materially different retained user base over 12 to 18 months than a reactive system ever will.

## Further Reading

**From Digia Engage:**

- [Mobile App Retention Rate: What It Is and What's Pulling It Down](https://www.digia.tech/post/mobile-app-retention-rate-what-it-is-and-whats-pulling-it-down/) — the D1, D7, and D30 retention framework that churn signals feed into
- [The Engagement Gap: Why Mobile Apps Lose Users Between Sessions](https://www.digia.tech/post/engagement-gap-why-mobile-apps-lose-users-between-sessions/) — the session interval mechanics that underlie session frequency decline as a signal
- [How to Increase Feature Adoption in Mobile Apps: A 6-Step Framework](https://www.digia.tech/post/how-to-increase-feature-adoption-mobile-apps-6-step-framework/) — the feature adoption funnel that feature reversion measures in reverse
- [When NOT to Show a Nudge: Building a Suppression Logic](https://www.digia.tech/post/when-not-to-show-a-nudge-suppression-logic/) — the suppression framework that prevents over-triggering from becoming its own churn driver
- [How to Know If Your Personalization Is Actually Working](https://www.digia.tech/post/how-to-know-if-your-personalization-is-actually-working/) — the holdout group measurement framework for validating whether churn interventions produce causal retention lift
- [Digia Engage Nudges](https://www.digia.tech/products/nudges) — event-based trigger architecture for deploying in-app re-engagement campaigns at the moderate-risk tier

**External Sources:**

- [How to Spot Users Who Are About to Churn Before They Leave](https://www.airbridge.io/en/blog/how-to-spot-users-who-are-about-to-churn) — Airbridge (10 to 28 day session decline window; staged churn process; frequency baseline methodology)
- [Predicting Customer Churn with Behavioral Signals](https://www.fullstory.com/blog/predicting-customer-churn/) — FullStory (behavioural vs. statistical signal layers; friction preceding decision)
- [Predictive Churn: Stop Mobile App Users From Leaving](https://www.dogtownmedia.com/predictive-churn-how-you-can-tell-a-user-is-leaving-your-mobile-app-before-they-do/) — Dogtown Media (71% churn within 90 days; McKinsey AI churn prevention reducing churn by up to 15%)
- [What Is Churn Prediction? A Complete Guide](https://clevertap.com/blog/churn-prediction/) — CleverTap (start simple with rule-based thresholds before scaling to ML)
- [Churn Prediction Model: Definition, Examples and Use Cases](https://www.saber.app/glossary/churn-prediction-model) — Saber (score threshold tiers; feature engineering methodology; model retraining cadence)
- [Predictive Churn Modelling: How to Build, Score, and Act](https://prometheusagency.co/insights/predictive-churn-modelling) — Prometheus Agency (Gartner 2025: 3.1 percentage point churn reduction from automated early-warning systems)
- [How to Reduce Notification Fatigue: 7 Proven Product Strategies](https://www.courier.com/blog/how-to-reduce-notification-fatigue-7-proven-product-strategies-for-saas) — Courier (52% of users who disable push eventually churn; volume overload as fatigue driver)
- [In-App Nudges: 12 Examples and 6 Design Patterns](https://www.nvecta.com/blog/in-app-nudges/) — Nvecta (15 to 40% in-app CTR vs. 2 to 5% for email and push; dismiss rate as a churn prevention metric)
- [The Ultimate Guide to In-App Nudges in 2026](https://www.plotline.so/blog/in-app-nudges-ultimate-guide) — Plotline (cart abandonment nudge mechanics; contextual relevance and timing principles)
- [Winback Campaign: Recovering Churned Mobile App Users](https://mwm.ai/glossary/winback-campaign) — MWM (1 to 5% uninstall win-back vs. 5 to 15% subscription win-back recovery rates)
- [Win-Back Email Sequence for Apps](https://www.airbridge.io/en/blog/subscription-app-win-back-sequence) — Airbridge (13.7% natural monthly reactivation; 14.7% cumulative 4-email sequence reactivation; 2.6x longer retention for non-discounted resubscribers)
- [Win-Back Campaigns: How to Re-Engage Churned Customers](https://www.stackmatix.com/blog/win-back-campaign-strategies) — Stackmatix (re-churn rate as the critical quality metric; churn-reason segmentation framework)
- [Feature Adoption Guide: Metrics, Funnel and How to Improve](https://www.appcues.com/blog/a-guide-to-feature-adoption) — Appcues (feature adoption breadth as a retention predictor)

_The event-based scoring model and tiered campaign triggers described in this article are deployable in Digia Engage using the same nudge and audience filter infrastructure used for other in-app campaigns. Session frequency, session depth, feature usage, and nudge dismissal events all flow into the audience filter layer without engineering tickets after initial SDK integration, and campaigns fire within 100ms of a qualifying event. [Book a demo](https://www.digia.tech/book-a-demo) to see how the trigger and suppression layer supports a churn scoring model, or [read the suppression logic guide](https://www.digia.tech/post/when-not-to-show-a-nudge-suppression-logic/) for the framework that prevents churn interventions from becoming their own churn driver._