The population moves before churn appears
Netflix churn risk, read from public behavior: a behavioral stress test of 377,487 public comments shows a large reversible-friction cohort building while projected hard churn remains close to 2%. Hard churn is the visible outcome. The larger population under active friction is still moving, still returning, and still recoverable.
read this first
- Public data only. This analysis uses publicly available discussion. We had no access to Netflix subscriber or internal behavioral data.
- Model projections. The population percentages below are projections under stated assumptions. They are not measured Netflix churn.
- Retrospective analysis. The model was run after the period examined. The purpose is to test whether the behavioral movement was already visible in public data.
- Independent benchmark. Netflix does not publish its churn rate. The external comparison comes from Antenna's third-party panel estimate.
The visible number is about 2%
A conventional subscriber view focuses on the people who leave and stay gone. Our model projects that hard-churn population at roughly 2% in April 2026. That is the terminal end of the process: boycott language, permanent walk-away behavior, piracy as a replacement, or other signals associated with durable exit.
The more consequential movement is happening earlier. Alongside that hard-churn tier, the model projects an approximately 12% active-friction cohort. These customers behave differently. They cancel and return. They reassess the service around new releases. Their relationship with the product has become conditional. They still generate revenue. They also carry substantially more defection pressure than a stable subscriber.
active friction · reversible
~12%
Cancel and resubscribe, toggling around new releases. Keeps coming back.
hard churn · permanent
~2%
Boycott, piracy, walk-away. Gone for good.
Around six times more people are in motion than have reached hard churn. A dashboard built around subscriber loss will capture the hard-churn tier. The larger movement can remain hidden because these customers continue to come back. That 12% is the warning window. By the time churn appears on the dashboard, the behavioral transition has already started.
The movement is visible in the raw signal
Before applying the behavioral model, we looked at a much simpler question: how often did specific frictions appear in public Netflix discussion over time? The answer changes sharply in Q1 2026. In dfmchn's 377,487-comment Netflix corpus, price friction rose 55% and advertising friction 124% from the earlier baseline into April 2026, and cancel-intent language rose with them.
These figures come directly from keyword prevalence in the underlying corpus. There is no behavioral model involved in this part of the analysis. The public conversation itself was changing. The model then asks what that changing pressure does to the population.
The population begins to cross
Difference Machine represents a population as behavioral states connected by constrained transitions. Under low pressure, most of the Netflix population remains in passive inertia. As price and advertising friction intensify, more of the population moves into active evaluation and fluid churn behavior. By April 2026, the model projects:
The important part is the direction of travel. The population does not jump directly from satisfied subscriber to permanent churn. Pressure accumulates. Behavioral states change. Some transitions remain reversible for a period of time. Others become increasingly difficult to reverse. That creates a window in which intervention can still change the destination.
This is why churn arrives late
Historical churn is valuable for understanding loss after customers cross the boundary. Behavioral transition data adds an earlier layer. A subscriber can remain visible as an active customer while becoming more price sensitive, cancelling between releases, or repeatedly reconsidering the relationship. Financially, that customer still exists. Behaviorally, the relationship has already changed. The distinction becomes especially important for subscription products where cancellation and resubscription are easy. A customer who leaves permanently creates a clear churn event. A customer who repeatedly cancels and returns can erode lifetime value for months without ever looking like durable customer loss.
The second group is where the Netflix analysis becomes interesting. dfmchn identified a reversible churn cohort of approximately 12% of Netflix subscribers by April 2026, six times larger than permanent churn and invisible to subscription dashboards. That reversible tier is a population still in transition, and why it is the cheapest place to spend a retention dollar is the subject of our field note the cheapest customer to save.
The two curves separate
Across the period, projected active friction rises meaningfully into Q1 2026. Projected hard churn remains in a much narrower range. Every model parameter held fixed, only the observed public signal moving month to month.
data table · two-tier projection by month
| month | active friction | hard churn | total at-risk |
|---|---|---|---|
| 2025-06 | 10.7% | 1.85% | 12.5% |
| 2025-07 | 10.1% | 1.86% | 12.0% |
| 2025-08 | 10.1% | 1.81% | 11.9% |
| 2025-09 | 10.6% | 1.84% | 12.4% |
| 2025-10 | 10.2% | 1.86% | 12.1% |
| 2026-01 | 9.8% | 1.87% | 11.6% |
| 2026-02 | 10.0% | 2.32% | 12.3% |
| 2026-03 | 11.5% | 2.07% | 13.6% |
| 2026-04 | 12.0% | 2.04% | 14.0% |
November and December are absent because those months are genuinely missing from the corpus. We did not interpolate them. The widening distance between the two tiers is the useful signal. More of the population is becoming unstable while permanent loss remains comparatively low. That is exactly the period during which a business still has room to act.
A useful external check
Netflix does not publicly report churn. Antenna independently estimated Netflix monthly churn in roughly the 2.0% to 2.5% range during 2025, including an estimate of approximately 2.3% in February 2025. dfmchn's model projected Netflix hard churn at approximately 2% in April 2026, consistent with Antenna's independent 2.3% estimate, with the projected tier ranging from approximately 1.8% to 2.3% across the analysis window. The hard-churn tier was not calibrated to the Antenna estimate.
We treat this as a range-level corroboration. It does not turn the model output into measured Netflix churn. It does provide a useful sanity check: the model's independently emerging terminal tier lands in the same low-single-digit territory as the external estimate. That gives us more confidence in the structure around it, especially the much larger reversible-friction population.
benchmark: antenna third-party panel · ~2.0 to 2.5% monthly through 2025 · netflix does not self-report
One important calibration disclosure
The total April 2026 at-risk population of 14.0% should not be interpreted as a blind prediction. One cohort parameter was calibrated once so that total April at-risk equals 14.0%. The hard-churn tier was not calibrated. It emerges from the model itself.
For that reason, the most useful result in this analysis is the structure: a large reversible cohort develops alongside a much smaller terminal cohort, while the independently emerging terminal tier remains in a plausible external range. The report should be read as evidence that public behavioral signal can reveal movement inside a population before that movement resolves into permanent loss.
Under the hood
Each transition in the behavioral model has a friction-sensitivity profile derived from the population. Observed stressors apply pressure to those transitions. As pressure rises, transition weights change according to their thresholds and cross-talk relationships. A cohort-flow layer then converts those stressed transitions into projected population shares. For the monthly Netflix analysis, the stressor structure and model parameters remain fixed. The observed public signal changes month by month. That means the movement shown above is driven by changes in the underlying corpus rather than monthly retuning of the model. The full pipeline is described in the Difference Machine Method.
Source signal
The raw keyword-prevalence series behind the stress test. Every number in the analysis regenerates deterministically from the public corpus.
data table · keyword prevalence by month
| month | comments | price | ads | cancel-intent |
|---|---|---|---|---|
| 2025-06 | 29,289 | 1.45% | 0.94% | 1.26% |
| 2025-07 | 38,943 | 1.40% | 0.57% | 1.06% |
| 2025-08 | 45,251 | 1.38% | 0.61% | 1.25% |
| 2025-09 | 51,718 | 1.52% | 0.63% | 1.02% |
| 2025-10 | 56,214 | 1.46% | 0.42% | 0.84% |
| 2026-01 | 48,427 | 1.30% | 0.50% | 1.01% |
| 2026-02 | 26,727 | 1.25% | 0.79% | 1.14% |
| 2026-03 | 34,929 | 1.74% | 1.00% | 1.19% |
| 2026-04 | 45,989 | 2.02% | 1.12% | 1.37% |
Why this matters beyond Netflix
Netflix gives us a useful retrospective test because the behavioral pressure changed substantially while an external churn estimate was available for comparison. The commercial use case begins before a decision is made. A company can define the population it cares about and introduce a proposed pressure into the model. The question then becomes: what happens to the population when we apply the decision?
How much pressure can it absorb? Where do meaningful transitions begin? At what point does recoverable friction turn into durable loss? That is what Difference Machine is built to stress test.
this is one run on one population. the same pipeline reads any corpus of human text. read the other stress tests and field notes.
Bring us a population