What problem does it solve?
A classic paywall applies the same rule to every visitor. FT Strategies gives the example of a standard paywall that appears after three articles within a 30 day period. The description of a Wall Street Journal session makes the same point: whether a prospective subscriber has visited a publisher multiple times or is a first time visitor, the experience is the same.
A reader who leaves at the wall also stops producing ad impressions, so a publisher that depends on both revenue lines has to decide, visitor by visitor, whether a subscription or an ad impression is worth more. In 2018 Digiday reported that NZZ got 60 percent of its revenue from subscriptions and less than 40 percent from advertising, and in 2026 Reuters said that it is not sacrificing ad revenue to grow subscriptions. A dynamic paywall lets a publisher make that decision per visitor instead of per site.
How does it work?
- Collect behavior signals. The publisher records, for registered and anonymous readers, what they read, how often they come back, on which device and at what time of day, and which newsletters they receive. In 2018 Digiday reported that NZZ had scored registered readers on about 30 attributes, such as time spent on articles, frequency and newsletters received.
- Score the propensity to subscribe. A model estimates how likely each reader is to subscribe. Reuters says its system estimates how much ad revenue a visitor could generate, then decides whether to show a paywall based on how likely that visitor is to convert.
- Choose the experience. The score and the content decide whether the paywall appears now, how many free articles the reader still gets, which package is promoted and what the message says. In 2018 NZZ served the payment prompt to readers who reached the top 20 percent of its propensity score.
- Adjust to the news cycle. Reuters says the paywall tightens when the news cycle is busy and content converts well, and loosens on a slow news day.
- Learn from outcomes. Every decision and its result, subscribed or not, goes back into the model, and the publisher runs A/B tests against a control group to check that revenue moves in the right direction.
- Audience
- Customer facing
- Autonomy
- Autonomous
- Adoption
- Early adopters
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Conversion uplift | Too few to pool | 92% | 1 | 1 organization |
| Conversion uplift | Too few to pool | 5x | 1 | 1 organization |
| Interactions handled | Not pooled | Not pooled: up to 15 million | 0plus 1 up to | 1 organization |
Value drivers: Revenue growth.
Indicative value
A news publisher with 2 million registered readers
USD 38,400 to USD 1.6 million
Added first year subscription revenue per year
How this is calculated
Formula: registeredReaders * baseConversion * relativeUplift * revenuePerSubscriber. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Registered readers who see a paywall decision each year registeredReaders, readers | 2,000,000 | 2,000,000 | The reference publisher. |
| Share of those readers who subscribe each year with a fixed paywall baseConversion, fraction of readers | 0.004 | 0.01 | Editorial assumption. Replace with the conversion rate of your own registered readers. |
| Relative increase in subscriptions from the dynamic paywall relativeUplift, fraction of base subscriptions | 0.1 | 0.4 | Conservative against the benchmarks on this page, where NZZ reported in 2018 that its conversion rate rose fivefold over three years and FT Strategies, the consulting arm of the Financial Times, reports a 92% increase in conversion rate at the Financial Times. Both are the organizations' own figures without a stated baseline, and much of the NZZ rise predates its propensity model. |
| First year revenue per added subscriber revenuePerSubscriber, USD per subscriber | 48 | 200 | The low value is Reuters' flat price of USD 4 a month, reported by Digiday. The high value is an editorial assumption for a premium news subscription. |
What it leaves out: Gross first year subscription revenue only. It leaves out churn in later years, any loss of advertising revenue from readers who now see a wall, discounts, the cost of the data and model work and the effect on the readers who would have subscribed anyway.
Market estimates (analyst estimates, not deployments)
- FT Strategies reports that at the INMA Subscriptions Summit it was highlighted that 22% of 264 news brands declared using a hybrid, dynamic or smart paywall, and says that proportion has increased 4 times since 2020. How dynamic paywalls are transforming reader revenue (2025)
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Financial Times
United Kingdom · Media and entertainment · 2025
The Financial Times launched a dynamic paywall about a year before November 2025, with Zuora as its technology partner. FT Strategies, the specialist consultancy from the Financial Times, which sells paywall consulting to news publishers, says the AI paywall has become a key driver of subscription growth, with a 92% increase in conversion rate, a 118% increase in progression through the subscription funnel and a 78% uplift in subscriber lifetime value. It also says the paywall reduced manual overhead and freed engineering capacity.
- Conversion uplift: 92%
"The AI paywall has become a key driver of subscription growth, with a 92% increase in conversion rate, a 118% increase in progression through the subscription funnel and a 78% uplift in subscriber lifetime value."
Claimed by: organization
The Wall Street Journal
United States · Media and entertainment · 2018
The Wall Street Journal describes a dynamic paywall in which machine learning and a proprietary algorithm predict each visitor's likelihood to subscribe, moving away from giving everyone the same experience. That likelihood sets the paywall experience the visitor gets, in real time, up to 15 million times a week. The listed objectives include the capacity to deal with 15 million visitors a week and the ability to increase subscription sales without impacting advertising capacity. The session description gives no results; the only outcome like statement is in a speaker biography, which credits machine learning models with the Journal's highest levels of membership ever.
- Interactions handled: up to 15 million, per week
"This takes place in real time, up to 15M times per week."
Claimed by: organization
Reuters
United Kingdom · Media and entertainment · 2026
Reuters launched a dynamic paywall in 2026 on top of its subscription offering. Phil Andraos, general manager for digital, told the Digiday Publishing Summit that a machine learning system uses user behavior and content performance to estimate how much ad revenue a visitor could generate, then decides whether to show that visitor a paywall based on how likely they are to convert. The paywall tightens when the news cycle is busy and loosens when it is slow. Reuters states that its advertising business is doing better than before the paywall and gives no subscriber or conversion figures.
No outcome disclosed.
Neue Zürcher Zeitung
Switzerland · Media and entertainment · 2018
NZZ, the Swiss German language newspaper, requires readers to register and eventually pay. It scores registered users on about 30 attributes, such as time spent on articles, frequency and newsletters received, to rate their propensity to subscribe, and shows the payment prompt to readers in the top 20 percent of that score. Between 100 and 150 rules based on reading history, device and time of day change the message, text, placement and color of the prompt. Managing director Steven Neubauer says conversion rose fivefold in three years. The report dates from 2018, so it shows an early production system.
- Conversion uplift: 5x, over three years, up to June 2018
"The 200-year old German-language publisher said it has increased conversion rate by fivefold in the last three years by using this system, with 2.5 percent of people who view a payment message prompted to subscribe."
Claimed by: organization
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Reader level behavior data: pages read, frequency, recency, device, referrer and newsletters
- Subscription and cancellation history linked to the same reader identity
- Consent records for the tracking the model relies on
- Ad revenue per reader or per page, when the publisher wants to weigh ads against subscriptions
Systems to integrate
- Content management system and paywall layer that can vary the wall per request
- Subscription and billing platform for packages, prices and offers
- Customer data platform or event pipeline for reader signals
- Experimentation tool for control groups
- Consent management platform
Complexity: Medium
The build needs a reader identity and event stream that can be queried within a page load, a paywall layer in the content management system that can change the wall per request, a subscription system for packages and prices, and an experimentation setup with a control group. A rules based version, which FT Strategies describes as a practical starting point when data is limited, can run before any model is trained.
- 1
Start with rules and a measurement frame
Define the segments and wall rules you can already act on, and set up a holdout group that always gets the current paywall so that every later change has a baseline.
- 2
Train the propensity model on your own readers
Use reader behavior before the first subscription as the features and subscription within a fixed window as the label. Check that the score ranks readers better than recency alone.
- 3
Decide what is allowed to vary
Choose whether the model may change the timing of the wall, the message, the package or all three. FT Strategies lists the access model, the messaging and the packages and prices shown to readers as things a dynamic paywall can adjust, and leaves dynamic pricing out of its article because it needs a completely different set of steps. Get a legal review before any price varies per reader.
- 4
Weigh subscriptions against ads
Where advertising matters, estimate the ad revenue per reader and feed it into the wall decision next to the likelihood to subscribe. Reuters says its system estimates ad revenue per visitor and then decides on the paywall based on how likely that visitor is to convert.
- 5
Review the model on a schedule
Retrain on recent conversions, compare against the holdout every month and retire rules that no longer win in tests.
Guardrails
- A holdout group that keeps the existing paywall, so uplift is measured and not assumed
- Tracking used for scoring limited to what the reader consented to
- No individual pricing without legal review and a notice to the reader
- A ceiling on how fast the wall can tighten, so a busy news day cannot lock out loyal readers
- Free access for content the publisher treats as public interest, such as emergency information
KPIs to instrument
- Conversion rate from reader to subscriber, treatment against holdout
- Subscriber lifetime value and churn by acquisition path
- Ad revenue per reader for readers who see a wall and those who do not
- Share of paywall decisions served within the page load time
Human in the loop
The commercial and editorial leads set which content is always free, which packages exist and how much advertising revenue the publisher is willing to trade for subscriptions. Analysts review the model against the holdout each month. The model decides per visit, within those limits.
Common failure modes
- Optimizing for the quick conversion
- A model rewards offers that convert and cancel early. Measure lifetime value and cancellation, not only the first subscription.
- Ad revenue falls unnoticed
- More walls shrink the pool of anonymous readers. Track ad revenue per reader in the same dashboard as conversion.
- Training on a biased sample
- A model trained only on readers who were already shown a wall learns that wall. Keep a small share of readers on random walls so the model sees alternatives.
- Consent drift
- New signals are added to the model without checking what the reader agreed to. Review features against the consent record at every retraining.
What are the risks and rules?
EU AI Act
Minimal risk
Scoring readers for a subscription offer is not an Annex III use and does not interact with people in a way that triggers the transparency duties of Article 50. It would move toward a more regulated setting if the same score decided access to an essential service. The relevant law is data protection: profiling for marketing needs a lawful basis and respects the right to object.
Guidance
- Regulation (EU) 2016/679, Articles 21 and 22 on profiling (European Union, Europe). Gives the data subject a right to object to profiling for direct marketing and limits decisions based solely on automated processing that significantly affect them.
- Guidance on the use of storage and access technologies (PECR) (Information Commissioner's Office, Europe). Explains when cookies and similar technologies need consent under the UK rules that implement the ePrivacy Directive, which applies to the tracking signals a reader level model relies on.
Controls to put in place
- Consent and lawful basis recorded for each signal used in the score
- Documented model features, owner and monthly review against the holdout
- A way for the reader to object to personalization and see a standard paywall
- Legal review before any individual price or discount is set by the model
Frequently asked questions
- What does a dynamic paywall change compared with a meter?
A meter applies a fixed rule such as three free articles in 30 days. A dynamic paywall changes when the wall appears, what content it gates, which message it shows and which package it promotes, depending on the reader. In 2018 NZZ served its payment prompt to readers who reached the top 20 percent of a propensity to subscribe score.
- How much do conversion rates rise with an AI paywall?
Published figures come from the organizations themselves. FT Strategies, the consulting arm of the Financial Times, which also sells paywall consulting, reports a 92% increase in conversion rate and a 78% uplift in subscriber lifetime value at the Financial Times, without a baseline. NZZ said in 2018 that conversion rose fivefold over three years, but Digiday dates its propensity scoring from 2018, so much of that rise came before the model. Reuters reports no figure and says its advertising business is doing better than before the paywall.
- Does a dynamic paywall reduce advertising revenue?
It can, which is why Reuters builds the expected ad revenue of each visitor into the decision. Reuters says subscribers still see ads, visit often and are authenticated, and that its advertising business is doing better than before the paywall launched.
- Is this the same as dynamic pricing?
FT Strategies treats them as separate subjects. It counts the access model, the messaging and the packages and prices shown to readers as things a dynamic paywall can adjust, and leaves dynamic pricing out of its article because it requires a completely different set of steps.
How to cite this page
Blits.ai AI Use Case Library, "AI dynamic paywall and subscription propensity scoring", last verified 13 October 2026, https://www.blits.ai/ai-use-cases/dynamic-paywall-and-subscription-propensity. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 13 October 2026: First published