ARPU and churn, measured consistently
Most cross-sector ARPU comparisons are comparing denominators. Once the definitions are held constant, the level differences shrink and the growth spreads do not.
2.7x to 3.8x
Inside every sector measured, the top quintile grows ARPU between 2.7 and 3.8 times faster than the median. The operator matters more than the category.
Key takeaways
- 01The spread between median and top-quintile ARPU growth inside a sector is wider than almost any gap between sector medians. The operator matters more than the category.
- 02ARPU level differences across sectors are mostly a denominator artefact: a telecom line, a shared streaming account and a marketplace active user are not the same unit.
- 03Churn comparisons break earlier than ARPU comparisons, because gross logo, revenue and net churn diverge by more than the sector gaps do.
- 04Growth rates travel across sectors in a way levels do not, because a percentage change survives a denominator that is merely consistent.
- 05Take your sector's spread from this report, not its median.
Romà Llambés · CEO, Markin · Reviewed 2026-09-10 · Every report is reviewed twice a year, and re-run out of cycle whenever a source it rests on changes. · Method and sources
The question
What differences between sectors survive once ARPU and churn are defined the same way?
Cross-sector ARPU levels are largely an artefact of the denominator: active users, subscribers, accounts and households produce different numbers from the same revenue. Once definitions are held constant, level comparisons lose most of their meaning, while the spread between median and top-quintile growth inside a sector survives and remains the decision-relevant figure.
- Sample
- 147 B2C operators across nine sector groupings, from the Markin benchmark set. Levels are monthly ARPU in USD converted at period-average rates. Rows are labelled by how much of the level column comes from disclosed figures rather than derived ones.
- Period
- TTM to Q2 2026. Growth rates are trailing twelve months to the end of Q2 2026.
- Reviewed
- 2026-09-10Reviewed twice a year
Findings
What we found
- 01Derived in this report
The spread between median and top-quintile ARPU growth is larger than almost every gap between sector medians, which means the operator matters more than the category.
Inside every sector in the set, the top quintile grows ARPU at a multiple of the median. Those multiples sit in a band that is wide by any standard, and it is a within-sector figure, so it is immune to the denominator problem that makes the level column unreliable. If you take one number from this report, take your sector's spread rather than its median.
Do thisTake your sector's spread from the table, not its median, and set the target against it.
- 02Derived in this report
Level differences between sectors are dominated by the denominator, not by revenue per relationship.
A postpaid telecom line, a streaming account shared by a household, and a marketplace monthly active are three different units. Revenue per household is much closer across sectors than revenue per active user, and neither is the number an operator can act on. The only actionable version is your own ARPU on your own denominator, tracked over time.
Do thisState the denominator next to every ARPU figure your team publishes or consumes.
- 03Derived in this report
Churn comparisons fail earlier than ARPU comparisons, because gross logo, revenue and net churn diverge by more than most sector gaps.
A base can report negative net revenue churn while losing customers steadily, if the remaining ones expand. That is a meaningful business fact and a useless comparison point. Before any cross-sector churn figure is quoted, the definition and the denominator have to be stated, and in most published comparisons neither is.
Do thisAgree one internal definition of churn, with the denominator fixed, before any cross-team target is set on it.
- 04Derived in this report
Growth rates are more comparable across sectors than levels, because a percentage change survives a denominator that is merely consistent rather than identical.
This is the practical reason the benchmark set publishes growth alongside level and treats the spread as the decision metric. An operator can act on a growth gap. It cannot act on a level gap that is mostly a definitional artefact.
Do thisTrack your own ARPU growth over time on a consistent denominator, and stop comparing levels across sectors.
Charts and tables
The data
- Streaming2.9x
- Telecom (postpaid)3.5x
- Telecom (prepaid)3.8x
- Fintech2.7x
- Travel2.7x
- Marketplaces2.7x
- Gaming (live services)2.8x
- Media / publishing3.8x
- Retail / subscription commerce3.4x
The spread is a ratio computed inside each sector, so it is comparable across rows in a way that the ARPU levels are not.
| Sector | Operators | Region | Median level | Top-quintile level | Median growth | Top-quintile growth | Spread |
|---|---|---|---|---|---|---|---|
| Streaming | 18 | US / EU | $9.40 | $14.80 | 3.4% | 9.8% | 2.9x |
| Telecom (postpaid) | 24 | US / EU | $34.60 | $48.20 | 1.9% | 6.7% | 3.5x |
| Telecom (prepaid) | 21 | LATAM / SEA | $4.10 | $7.30 | 2.1% | 7.9% | 3.8x |
| Fintech | 19 | US / EU | $6.80 | $13.10 | 4.2% | 11.3% | 2.7x |
| Travel | 12 | Global | $11.20 | $19.60 | 2.8% | 7.5% | 2.7x |
| Marketplaces | 16 | Global | $7.50 | $15.40 | 3.7% | 10.1% | 2.7x |
| Gaming (live services) | 14 | Global | $5.20 | $12.90 | 5.1% | 14.2% | 2.8x |
| Media / publishing | 13 | US / EU | $8.10 | $13.70 | 2.2% | 8.4% | 3.8x |
| Retail / subscription commerce | 10 | US / EU | $12.60 | $22.40 | 3.0% | 10.2% | 3.4x |
| Denominator | What it counts | Effect on the ARPU figure |
|---|---|---|
| Monthly active users | Anyone who used the service in the period, paying or not | Lowest figure; moves with engagement as well as with revenue |
| Paying subscribers | Accounts with an active paid entitlement | Higher; the most common disclosure in subscription businesses |
| Accounts | Billing relationships, which may hold several lines or profiles | Higher again; standard in telecom, where a line is the unit |
| Households | Distinct households served, where inferable | Highest; closest to comparable across sectors and rarely disclosed |
Run it on your numbers
Look up your sector, and compare the spread rather than the level
Levels are anchored to the region shown and are not comparable across sectors. The spread is a ratio computed inside each sector, so it is.
- Median growth
- 3.4%
- Top-quintile growth
- 9.8%
- Operators, region
- 18, US / EU
Spread in your sector
2.9x
Top-quintile growth divided by median growth. This is the comparable figure: it is a ratio computed inside one sector, on one denominator.
Distance to the top quintile: 9.8 percentage points of ARPU growth.
Implications
What this means for the person who has to act
Growth lead
Stop benchmarking your ARPU level against another sector. Benchmark the spread between median and top-quintile growth inside your own, because that is the part execution can move.
Finance and investor relations
State the denominator next to every ARPU figure you publish or consume. Active users, subscribers, accounts and households turn one revenue line into four different numbers.
Analytics
Publish one internal definition of ARPU and one of churn, with the denominator fixed, before any cross-team target is set on either. Gross logo churn, revenue churn and net revenue churn diverge on the same base by more than most sector gaps.
Method
Definitions that change the answer
- ARPU
- Recognised revenue for the period divided by the average number of units in the denominator. The denominator is the entire argument: active users, paying subscribers, accounts and households are four different numbers from the same revenue line.
- Active user
- Used here as monthly active, where an operator discloses it. Definitions of activity differ by sector and sometimes by operator, which is exactly why the level column is not comparable across rows.
- Churn
- Gross logo churn per month unless stated. Revenue churn, net churn and cohort-based retention give materially different answers on the same base, and the difference is larger than most cross-sector gaps.
- Spread
- Top-quintile growth divided by median growth within a sector. Unitless, so it survives denominator differences and currency conversion. This is the number worth comparing across sectors.
Estimator
The model, written out
Two quantities, and the report is mostly about the difference between them: the ARPU level of a sector, which is denominator-dependent, and the within-sector growth spread, which is not.
ARPU
A = R / U
Revenue R over denominator U. Four common choices of U produce four different values of A from one revenue line.
Why levels do not travel
log A = log R - log U
Cross-sector variance in log A absorbs all the variance in how U is defined. Comparing two sectors on A compares their denominators as much as their economics.
Why growth does travel
if U' = c x U then A'_t / A'_(t-1) = A_t / A_(t-1)
Any denominator that is wrong by a constant factor cancels out of a growth rate. This is the exact reason the spread column is the comparable one.
Spread
S = g_p80 / g_median
Top-quintile TTM ARPU growth over median TTM ARPU growth, within a sector. Unitless, currency-free and denominator-free.
What has to hold
- Each operator's denominator definition is stable across the twelve-month window. A constant error cancels; a mid-window restatement does not.
- Operators inside a sector row are comparable enough to rank, which is a weaker requirement than being comparable enough to average.
- Currency conversion at period-average rates, applied to levels only. Growth is computed in reporting currency before conversion.
What breaks it
- An operator restating what counts as an active user during the window, which injects a step change straight into its growth rate.
- Sector rows thin enough that the top quintile is one operator, at which point the spread is a story about that operator.
- Survivorship: operators large enough to disclose or be estimated reliably. Small operators, where variance is highest, are absent.
- Mixing gross logo, revenue and net churn in one comparison, which moves the answer by more than the sector gaps being compared.
Method
How to reproduce this
- 01
Take the level column for any two sectors and note the denominator each uses. If they differ, stop: the comparison is not a comparison.
- 02
Recompute both on the same denominator, using a consistently defined active or paying base. Most of the visible gap moves.
- 03
Now take the spread column, which is a ratio of two growth rates from the same operators on the same denominator. It is unaffected by the previous step.
- 04
Compare spreads. A high spread means operator execution, not sector economics, is driving the difference.
- 05
Apply the same treatment to churn before comparing it: gross logo, revenue and net churn are not interchangeable, and net churn can be positive while customers leave.
Uncertainty
How wrong each figure could be
The top-quintile figure is an order statistic on a small per-sector sample, so its uncertainty is a sampling property of the rank, not of the arithmetic. The intervals below are distribution-free 90% intervals for the 0.8 quantile, derived from the binomial coverage of order statistics at the row sizes actually in the set (n = 12 to 24 per sector, 147 operators pooled).
Top quintile at n = 18
A 90% distribution-free interval covering roughly the 67th to the 100th percentile. At this row size the quintile is indicative, not precise.
12th to 18th ranked operator
Top quintile at n = 24
Roughly the 67th to the 96th percentile. Larger rows tighten the estimate but do not make a single sector row precise.
16th to 23rd ranked operator
Top quintile pooled across all 147 operators
Only the pooled estimate is tight. This is why the report argues from the pattern across sectors rather than from any one row.
75th to 86th percentile
Spread of 2.7x to 3.8x
Individual sector spreads should be treated as indicative to about half a turn. The finding is that every sector spread is large, not that one sector's is larger than another's.
read as a pattern, not as nine measurements
Robustness
What survives a different assumption
Each row changes a definitional or sampling choice and asks whether the conclusion, that within-sector spread beats between-sector differences, survives.
| Change one assumption | What changes | Result | Conclusion |
|---|---|---|---|
| All levels recomputed on a single denominator | One definition of the paying base | Level gaps between sectors shrink substantially. Spreads are unchanged by construction. | Holds |
| Churn switched from gross logo to revenue churn | Different churn definition | Sector churn ranks reorder. The ARPU spread conclusion is untouched. | Holds |
| Derived rows dropped, disclosed rows only | Sample restricted | Fewer sectors, wider intervals, same direction. No sector reverses. | Holds |
| Quintile replaced by the single best operator | Order statistic changed to the maximum | Spreads rise and become unstable. This is the wrong statistic and the report does not use it. | Flips |
| Growth measured over four quarters instead of trailing twelve months | Window redefined | Individual values move; the ranking of sectors by spread is broadly preserved. | Holds |
The comparison that survives every treatment is the within-sector spread, and the comparison that survives none of them is the cross-sector level. The report is therefore written to be used one way only: take your own sector's spread as a target for what better execution looks like, and ignore the level column when arguing with another sector.
Limits
What this cannot tell you
- Levels are anchored to the regions listed per row and are not global figures. Comparing a US streaming level with a South East Asian prepaid level tells you about geography, not about the sectors.
- Some rows are partly derived rather than disclosed, and are labelled that way. Derived rows carry more uncertainty in the level column and roughly the same uncertainty in the growth column.
- This is an observational set of operators large enough to disclose or be estimated reliably. It under-represents small operators, where variance is higher.
- The set does not establish causation. A high spread shows that operators inside a sector diverge; it does not show what caused the divergence.
Falsifiability
What would change our mind
- 01
Widespread disclosure of household-level ARPU across sectors. Level comparisons would become meaningful for the first time.
- 02
Evidence that sector level differences persist once every operator is restated onto one denominator and one currency.
- 03
A sector where the median-to-top-quintile growth spread is close to one. That would mean execution creates little room there, and the spread would stop being the decision-relevant figure.
Provenance
Sources, and what each one was used for
- Derived in this reportMarkin B2C ARPU benchmark set 2026
All level, growth and spread figures. 147 operators, TTM to Q2 2026.
- Public company disclosureNetflix, Spotify, T-Mobile, Verizon and Vodafone investor disclosures
Anchoring the disclosed rows and the denominator definitions used by each sector.
- Derived in this reportDenominator analysis
The second table, and the finding that level differences are dominated by the unit rather than by revenue per relationship.
Questions this report answers
- What is a good ARPU for my industry?
- How do ARPU benchmarks compare across sectors?
- Why is our ARPU lower than the industry average?
- How should ARPU be defined and measured?
- Is churn comparable between industries?
Frequently asked
- Can I compare my ARPU to another sector's?
- Not usefully. Sector levels are anchored to different denominators, active users, subscribers, accounts or households, and to different regions. The comparable figure is the spread between median and top-quintile growth inside your own sector, because it is a ratio computed on one denominator.
- Why does the denominator matter more than the revenue?
- Because the same revenue line divided by monthly actives, paying subscribers, accounts or households produces four materially different numbers. Most published cross-sector ARPU gaps are that arithmetic rather than a difference in how much customers are worth.
- What is the spread and why is it the number to use?
- The spread is top-quintile ARPU growth divided by median growth within a sector. It is unitless, computed on one denominator and one currency, so it survives every definitional problem that makes the level column unreliable, and it says how much room execution creates in your category.
- Is churn comparable across sectors?
- Less so than ARPU. Gross logo churn, revenue churn and net revenue churn diverge on the same base by more than most sector gaps, and net churn can look healthy while customers leave. Any cross-sector churn figure without a stated definition and denominator should be discarded.
Act on this
- PlaybookRun a retention offer before renewal without paying customers who were staying anywayPre-renewal save offers on Stripe, sized by risk instead of by list membership.
- ResearchThe retention discount lineMost save offers are evaluated on how many customers stayed. The only number that decides is how many stayed who would otherwise have left.
- ResearchRevenue that carries marginEvery growth decision on this list raises reported revenue. Four of them raise contribution margin. They are not the same four most teams prioritise.
- GuideHow to increase ARPU for B2C companies in 2026Five ARPU levers ranked by durability, the diagnostic to run first, and why sustained ARPU growth is a decision throughput problem rather than an ideas problem.
- GuideThe operating model behind sustained ARPU growthWhy ARPU growth stalls at the same ceiling in most B2C businesses, and the operating model that changes decision throughput instead of headcount.