# Subsets

Aiolum story: https://aiolum.com/story/subsets/ · added 2026-11-06 · updated 2026-11-06

**In short.** Subsets (AI churn prediction and retention experiments for subscription media, Denmark): More than 500 lifecycle experiments run on the platform, July 2026. Evidence grade E: no revenue figure was checked. Team: 4 co-founders at the Y Combinator launch in July 2023; YC lists a team of 7 in November 2026. Launched July 2023; still running.

AI churn prediction and retention experiments for subscription media · Denmark · AI-native · team of 2 to 10 · still running · group: Documents, data and slides

- Outcome: More than 500 lifecycle experiments run on the platform, July 2026 (evidence grade of the headline number: E; no revenue figure)
- Launch: July 2023; months from launch to the outcome: 36
- Team: 4 co-founders at the Y Combinator launch in July 2023; YC lists a team of 7 in November 2026
- Outside funding: Y Combinator (Summer 2023); US$6.1M raised in total according to its YC page, including a US$1.65M pre-seed round in February 2024
- Main factor: Niche
- Meters, 1 to 4: evidence strength 1, copyable pattern 1, capital needed 3
- Tags: AI-native, Small team, Funded, Running, B2B, Subscription media software

Links:

- Visit subsets.com: https://www.subsets.com/

**The product.** A web platform for the retention teams of subscription businesses, mostly news publishers and streaming services. It trains a machine learning model on each customer's own subscriber data, finds groups of subscribers who are likely to cancel, explains which behaviour puts them at risk, runs A/B tests of retention actions through the customer's existing email and CRM tools, and turns the tests that work into automatic flows. It is sold through demos; no prices are published.

**Why it worked.** Subscription publishers know how many subscribers they lose, but their CRM tools report opens and clicks, not whether a campaign kept anyone. Subsets sells the missing step: a test with a control group, measured on retention and lifetime value, that a commercial team can run without data scientists or engineers.

Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; E = no revenue figure found, so the headline is a non-money fact and the other facts carry their own grades; calc = computed on this page from graded facts.

## Summary

Subsets is a retention platform for consumer subscription businesses. Retention is the share of subscribers who stay from one period to the next; its opposite is churn, the share who cancel. Subsets builds a predictive model, a statistical model that estimates from past behaviour which subscribers are likely to cancel, and then runs A/B tests: one group of subscribers (the treatment group) gets an action such as an email series or an offer, a similar group (the control group) gets nothing, and the difference in retention shows what the action did.

Four co-founders, Oliver, Martin, Matias and Nikolai, launched it on Y Combinator's site on 17 July 2023, in the Summer 2023 batch. Y Combinator lists Martin Johnsen (CEO), Oliver Brandt (CTO) and Nikolai Skelbo as founders, the company as founded in 2023, and a team of 7 in Copenhagen. Customers named by the company include The Atlantic, NBCUniversal, Daily Mail, Matas, Hearst, The Athletic and Børsen. The company said in June 2025 that it supported more than 20 million active subscriptions across its customers, and in July 2026 that more than 500 experiments had been run on the platform.

No revenue figure for Subsets was found, so this story has evidence grade E. The company said in February 2024 that it had raised US$1.65 million in pre-seed funding, and its Y Combinator page says it has raised US$6.1 million in total. Those are investments, not revenue.

## Pricing

Subsets prices in November 2026: not published; the site offers only "Request a demo" (grade C)

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Subsets platform |  |  | price not published; sold after a demo |

The home page, the product pages and the careers page show no prices and no plans. The site sends every visitor to a demo request form. The table records that no price is published.

## How much revenue does Subsets make? The numbers, graded

A Copenhagen company started by four co-founders in Y Combinator's Summer 2023 batch, now 7 people with outside funding; traction is counts of experiments and subscriptions, and no revenue figure was found

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Experiments run on the platform, July 2026 | more than 500 | C | Company blog post of 7 July 2026, "Subsets crosses 500+ lifecycle experiments". Covers onboarding, activation, engagement, renewal, upsell, winback and churn prevention. | [6] |
| Active subscriptions supported, June 2025 | more than 20 million, across its customers | C | Company blog post of 16 June 2025. These are the customers' subscriptions, not Subsets' own customers. The same post said it had launched over 100 retention experiments and created more than 200 behavioural audiences that year. | [5] |
| Revenue of Subsets | not published | C | No payment-processor profile, founder interview with figures or filing was found. See "What is not verified". | [1, 3, 4] |
| Customers named | The Atlantic, NBCUniversal, Daily Mail, Matas Group and Hearst; The Athletic and Børsen | C | The first five from the Y Combinator company page, the last two from the February 2024 funding post. Case studies on the site cover DailyMail+, Matas, Hearst, The Philadelphia Inquirer and The Independent. | [1, 4, 3] |
| Price | not published | C | Home page, read 6 November 2026: only "Request a demo". | [3] |
| How the model works | a model trained on each customer's first-party data; gradient boosting and temporal sequencing | C | Home page: "Subsets trains a unique machine learning model on top of your proprietary 1st party data". Funding post, February 2024: "a proprietary AI framework based on gradient boosting and temporal sequencing". | [3, 4] |
| Team at launch, July 2023 | 4 co-founders | C | YC launch post: "We are Oliver, Martin, Matias, and Nikolai - the co-founders of Subsets". The February 2024 funding post is signed by Martin, Oliver and Nikolai. | [2, 4] |
| Team, November 2026 | 7 | C | Y Combinator company page, read 6 November 2026. The careers page calls it "a fast-growing, lean team" in Copenhagen. | [1, 7] |
| Funding | US$6.1 million in total; US$1.65 million pre-seed in February 2024 | C | YC page: "Subsets has raised $6.1m". Funding post, 27 February 2024: pre-seed "led by Upfin and Y Combinator", with Cuesta Labs, Sandhill Markets and angel investors. The careers page also names FinTech Collective as an investor. | [1, 4, 7] |
| Example customer result | Matas: 29 percentage points higher 30-day retention in one test | C | Case study published by Subsets, 17 October 2025, with quotes from Matas' Head of Loyalty. It is the customer's result, not Subsets' revenue. | [8] |
| AI models used | its own models; no outside model provider named | C | The privacy policy describes "an AI-based subscription retention platform" and names no model vendor. | [9, 3] |
| Time from launch to the headline | 36 months | calc | Calculated: July 2023 (YC launch post) to July 2026 (the 500-experiment post). | [2, 6] |

How to read this table: a grade C value is a statement by the company on its own site or its Y Combinator pages, or a statement by a named customer published by the company. A "calc" value was computed on this page. No value is grade A or B, and no row is Subsets' own revenue.

## When did Subsets launch, and how did it grow? Timeline

From a Y Combinator launch with four co-founders to more than 500 retention experiments for news publishers, in three years. Each entry carries its grade.

- **17 July 2023** (grade C [2]): Launch post on Y Combinator's site, Summer 2023 batch: "The AI retention co-pilot for subscription businesses". Four co-founders ask for introductions to subscription businesses and for pilot customers.
- **27 February 2024** (grade C [4]): The company announces a US$1.65 million pre-seed round led by Upfin and Y Combinator, and says it launched six months earlier out of Copenhagen and works with publishers such as The Athletic and Børsen.
- **16 June 2025** (grade C [5]): The company says it supports more than 20 million active subscriptions across its customers, after the INMA World Congress and a first Retention Roundtable dinner in London with 10 large publishers.
- **17 October 2025** (grade C [8]): Case study with Matas, the Scandinavian health and beauty retailer: a re-engagement test on dormant loyalty-club subscribers.
- **7 July 2026** (grade C [6]): The company says more than 500 experiments have been run on the platform.
- **November 2026** (grade C [1, 3]): Y Combinator lists a team of 7 in Copenhagen and US$6.1 million raised. The site still sells by demo only.

## Why did Subsets work?

### One industry

The product speaks the language of subscription media: articles read, app downloads, family members invited, renewals. Its blog posts, case studies and events are all for news publishers and streaming services, so each sale makes the next one easier to explain.

### Prediction joined to a test

A churn score alone does not keep a subscriber. Subsets pairs each predicted audience with a controlled test and reports the retention lift against the control group, which is the number a retention manager has to show their boss.

### No engineers on the customer side

The site says it is built for commercial teams and runs through the tools a publisher already has (Braze, HubSpot, BigQuery, Amplitude and others on the integrations list), so a customer can start without a data science project.

## How it was built

Only what the company's own pages say. They name the model type, not the libraries or the hosting.

- A machine learning model trained for each customer on its own first-party subscriber data, according to the home page.
- Gradient boosting (many small decision trees, each one correcting the errors of the ones before) combined with temporal sequencing (features built from the order of a subscriber's actions over time), according to the February 2024 funding post.
- "Explainable AI": for each audience the platform shows which behaviours drive the churn risk, for example moderate activity followed by 30 days of inactivity in the Matas case study.
- Experiment tooling: sample-size calculation, time left until results are ready, and detection of statistical significance, according to the home page.
- Connections to subscription billing, product analytics and CRM tools; the sitemap lists integration pages for Braze, HubSpot, Chargebee, BigQuery, Databricks, Amplitude and GA4 among others.

Sources: [3, 4, 8, 12]

## Who tried this before Subsets, and why did they fail?

Three companies in our database of failed startups promised to predict churn or automate retention marketing, and failed. The values below are from that database and are not independently verified.

The failed startups come from our database of dead startups, retold here in our own words; Its entries are AI-assisted summaries of public sources, so they carry no evidence grade.

### Preact (2012 to 2018; funding US$4.7M; cause: Competition)

- Promise: Predict which SaaS customers would churn from their usage data, so that teams could act before they left.
- Why it died: Its predictions were not unique enough against larger customer-success platforms such as Gainsight that covered the whole customer lifecycle.
- Record: 

### Userfox (2011 to 2014; funding US$700K; cause: Competition)

- Promise: Automated lifecycle emails triggered by user behaviour, to raise retention and conversion.
- Why it died: Mailchimp, Constant Contact and integrated CRM suites already held the market, and it lacked the money to keep up.
- Record: 

### Marketfox (2017 to 2020; funding US$2M; cause: Competition)

- Promise: Marketing automation for online shops: segmentation, email and push messages to raise customer lifetime value.
- Why it died: It could not stand out against Mailchimp, HubSpot and Klaviyo, and could not raise enough to compete with them.
- Record: 

## What changed

All three sold retention tools to a broad market and lost to large suites that already owned the customer's email and CRM. Subsets does not try to replace those suites: it sits on top of them and sends its audiences and tests through them, and it sells to one industry, subscription media, instead of to every online business.

What did not change: the large CRM and messaging vendors keep adding their own AI features, and the sitemap shows that Subsets publishes a page titled as a Braze AI alternative. Prediction alone was not enough for Preact, which is why Subsets puts the controlled test at the centre.

## Can one person copy Subsets now?

**Verdict: The idea, yes. The way it was built, not alone.**

One industry, one painful metric, and a product that proves its effect with a control group is a pattern any founder can use. Subsets itself is a funded team of 7 that sells by demo to large media companies, with data integrations and enterprise sales that one person would find hard to carry.

1. Pick one industry where losing customers is the main cost, and learn its metrics and its tools.
2. Connect to the tools the customer already uses instead of replacing them.
3. Report every action against a control group, so that the customer can see what the product changed.
4. Publish the customer results as case studies with named people.

**What is still open.** The same predict-and-test loop for other subscription businesses with weak tools: gyms, meal kits, software for small firms, paid newsletters.

## What could go wrong for Subsets?

- **What is not verified.** No revenue figure for Subsets was found. We looked for a payment-processor profile, a founder interview with figures and a filing. The headline counts (more than 500 experiments in July 2026, more than 20 million subscriptions in June 2025) are the company's own statements on its blog. The February 2024 funding post said the company had "run thousands of experiments" six months after launch, which does not match the 500 experiments announced in July 2026; the two posts may count different things. A reader could look for an annual report in the Danish company register or an interview with the founders that gives revenue.
- **Team and funding.** The company was started by four co-founders (YC launch post, July 2023); the February 2024 post is signed by three of them. Y Combinator lists a team of 7 in November 2026 and US$6.1 million raised, and the careers page names Y Combinator and FinTech Collective as investors. It is a venture-funded team, not a solo business.
- **Competition and model providers.** No other retention or churn-prediction story is on this site. The company's own sitemap lists a page positioned as an alternative to Braze's AI features, a large messaging vendor that Subsets also integrates with. The company says it trains its own gradient-boosting models on each customer's data, and none of its pages name an outside AI model provider, so it does not appear to depend on a large language model vendor.
- **Customer logos and case studies.** The customer names on the Y Combinator page (The Atlantic, NBCUniversal, Hearst and others) are the company's own statement; we did not find statements by those companies. The DailyMail+ and Matas case studies quote named managers but are published by Subsets. One DailyMail+ case study tests a 40% price rise and reports the forecast lifetime value; that is a customer experiment, not Subsets' revenue.

## Sources

1. Y Combinator company page: Subsets — https://www.ycombinator.com/companies/subsets (grade C). Read the company data in the page, 6 November 2026: description, founders, team size, founding year, batch, location and funding total.
2. Y Combinator Launches, 17 July 2023: Subsets – The AI retention co-pilot for subscription businesses — https://www.ycombinator.com/launches/J1P-subsets-the-ai-retention-co-pilot-for-subscription-businesses (grade C). Read in full. Company's own launch post. Source for the four co-founders and the original product.
3. Subsets home page, read 6 November 2026 — https://www.subsets.com/ (grade C). Read in full. Source for the product, the model description, the demo-only sales and the Daily Mail quote.
4. Subsets blog, 27 February 2024: Subsets raises $1.65M to automate retention for subscription media using explainable AI — https://www.subsets.com/blog/subsets-raises-1-65m-to-automate-retention-for-subscription-media-using-explainable-ai (grade C). Read in full. Source for the pre-seed round, the investors, the launch timing and the model type.
5. Subsets blog, 16 June 2025: Subsets is now managing over 20 million subscriptions — https://www.subsets.com/blog/subsets-is-now-managing-over-20-million-subscriptions (grade C). Read in full. Source for the subscription count and the 2025 experiment count.
6. Subsets blog, 7 July 2026: Subsets crosses 500+ lifecycle experiments — https://www.subsets.com/blog/subsets-crosses-500-lifecycle-experiments (grade C). Read in full. Source for the headline count.
7. Subsets careers page — https://www.subsets.com/careers (grade C). Read in full. Source for the team description and the investors named.
8. Subsets customer story, 17 October 2025: How Matas increased retention by +29pp with Subsets — https://www.subsets.com/customer-stories/how-matas-increased-retention-by-29-with-subsets (grade C). Read in full. A customer result published by the company, with quotes from Matas' Head of Loyalty.
9. Subsets privacy policy — https://www.subsets.com/privacy-policy (grade C). Read the sections on the services and the information processed. No AI model vendor is named.
10. Subsets customer story: How DailyMail+ increased CLV by +32% with Subsets — https://www.subsets.com/customer-stories/how-dailymail-increased-clv-by-32-with-subsets (grade C). Read in full. A customer experiment published by the company.
11. Database of failed startups, local export — /method/#dead (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.
12. Subsets sitemap, read 6 November 2026 — https://www.subsets.com/sitemap.xml (grade C). Read the list of page addresses only. Source for the integration pages and the Braze alternative page.

Contact: aiolum@livyden.com. Run Subsets? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: Subsets).