Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics › Measurement Planning

Measurement for Subscription Businesses

How a subscription business measures marketing: trial, activation and first payment events, billing as the source of truth, and cohorts and payback by channel.
Guideintermediate11 min readUpdated

What changes when revenue recurs

In a shop, the order is the outcome and it happens on the website, so a purchase event carries the order value and the campaign that brought the buyer. A subscription breaks that in three ways.

First, the value arrives in instalments. The first payment is a fraction of what a customer will be worth, and how large a fraction depends on how long they stay, which nobody knows on the day they sign up. Second, most of the instalments happen without a visit. Renewals, upgrades, downgrades, failed payments and cancellations are processed by the billing system, by email or inside the product, out of sight of a tag on the marketing site. Third, the step analytics sees most clearly, the sign-up or trial start, is the step furthest from revenue, and the cheapest channel for sign-ups is often not the cheapest for paying customers.

The consequence is a division of labour. Analytics is good at the beginning: where visitors came from, which pages they read, whether they signed up and started a trial. Billing is good at the money: who paid, how much, for how long, and when they stopped. Neither answers the question marketing is asked, which is how much recurring revenue each channel produced for what it cost, until the two are joined. The SaaS industry guide lists the KPIs from sign-up to renewal; this guide covers the data those KPIs need. An online store that adds a repeat-delivery option has the same split between the first order and the deliveries after it, and the guide to subscribe-and-save programmes for online stores covers the offer, cadence and cancellation design that this data then measures.

The lifecycle events and who records them

Write down the lifecycle as a list of events, then assign each to the system that sees it reliably. A browser tag should record what happens in the browser; everything else belongs to the systems that process it.

Lifecycle eventWhat it meansRecorded byAlso sent to
Sign-upAn account was createdThe product's databaseAnalytics, as the documented sign-up event, with the method as a parameter
Trial startA trial began, with or without payment detailsThe subscription or billing systemAnalytics, as an event with the plan
ActivationThe customer did the thing that shows the product delivered value, as the business defines itProduct dataAnalytics where the product is instrumented; the CRM as a status
First paymentThe trial converted, or a paid plan startedBillingAnalytics, as an event with its value, and ad platforms as a conversion
RenewalA further period was paid forBillingThe reporting warehouse only
Plan changeAn upgrade, downgrade or switch between monthly and annualBillingThe warehouse; the CRM
Payment failure and recoveryA charge failed, and whether it was recoveredBillingThe warehouse
CancellationThe customer asked to stop, and the date the service actually endsBilling or the subscription systemThe warehouse; the CRM

Google's documentation describes events as the unit analytics uses to measure interactions, and the events and parameters guide explains the split between recommended events, such as the sign-up event, and custom ones, along with user properties for attributes like plan or subscription status that describe the person rather than the action. Two definitions need particular care. Activation is the business's own choice, so write it down with the evidence that it predicts staying. And cancellation has two dates, the request and the end of the paid period; reports that mix them disagree about when churn happened.

Billing as the source of truth

For revenue, status and plan, the billing system is the source of truth, and every other figure is reconciled to it. It knows the amounts actually charged, after discounts, refunds, credits and failed payments, in the currency charged; it knows who is active today; it knows the plan each customer is on. A first-payment event in analytics is a useful signal for campaign reports and for bidding, but it is sent from a browser that may have blocked the tag, refused consent or closed before the confirmation page loaded, and it never sees what happens next.

Agree the revenue measures once and label every report with them:

  • First payment per customer, which is the first revenue a channel can be judged on.
  • Recurring revenue by month, from billing, with annual plans spread across their months if finance does so, and stated either way.
  • Net revenue after refunds and credits, with the refund window recorded so that recent months are known to be provisional.
  • Contribution after the costs of serving the customer, which is what payback should use.

Then reconcile monthly: first payments in analytics against first payments in billing, by week. A gap is expected; a gap that changes suddenly is a tracking fault, a consent change or a checkout change, and the data quality routine is where it should be caught.

Joining analytics, billing and the CRM

The join needs one identifier that every system holds. The practical choice is the account identifier the product creates at sign-up: an internal number that only the business's systems can resolve, never an email address, as the guide to keeping personal data out of analytics explains. Write it to the billing record, to the CRM record and to analytics as the site's own user identifier, under the same consent rules as the rest of measurement.

Capture acquisition at sign-up as well, on the account record itself: the campaign values from the landing page, the first landing page and the referrer, stored the way hidden form fields store them for leads. This matters because the account record is permanent and the analytics session is not, and because visitors who declined measurement still have an account even though analytics has no session for them. Where the sign-up form also asks how people heard about the product, store the answer beside the tracked source.

The join itself is easiest in a warehouse. The BigQuery export holds every event the property collected, and the BigQuery export basics guide shows that each row carries the site's own user identifier where the site sets one, beside the pseudonymous device identifier. Load the billing and CRM tables into the same warehouse and the account identifier links them: for each account, the acquisition source, the sign-up and trial dates, the first payment and every month of revenue after it. That table is what the cohort reports in the next section read from.

Judging channels on revenue that arrives later

Channels should be judged in stages, each later stage confirming or overturning the earlier read.

  1. Trial starts and cost per trial arrive within days. They show whether the channel reaches people, and nothing about whether they will pay.
  2. First payments arrive at the end of the trial. The Trial-to-Paid Conversion Rate by channel and the cost per paying customer are the first figures that deserve a budget decision.
  3. Revenue per trial at fixed ages arrives over the following months: the revenue a month's trials have produced by the end of the first billing period, then at later ages that suit the billing cycle. Comparing channels at the same age is the point; a young cohort always looks worse than an old one, and comparing them is the commonest way this analysis goes wrong.
  4. Payback by channel comes last: the Payback Period, or months until a channel's contribution per paying customer repays the cost of acquiring that customer, the CAC. The LTV and CAC guide explains why it must be calculated on margin, and the break-even calculator for subscription businesses does the arithmetic.

Four details change the reading. Annual plans make the first payment large and the later months empty, so compare monthly and annual customers separately or spread the annual payment. The same care applies to an experiment that changes the mix, such as the billing period default test, which reads its arms on contracted revenue from billing, net of refunds, rather than on plan starts. Introductory discounts depress early revenue for the channels that carry them. Trials that require payment details convert differently from trials that do not, so a channel sending people to a different trial type is not comparable. And LTV projected from a few months of data is a projection; label it as one until the cohorts have aged.

Reporting churn and retention to marketing

Marketing needs churn by acquisition channel, because a channel whose customers leave early costs more than its cost per paying customer suggests. It also needs protection from churn it did not cause. Three habits keep the report fair.

  • Compare channels within the same cohort. If customers from one channel who started in a given month leave faster than customers from other channels who started that month, the channel is likely attracting people the product does not suit, or its message promised something the product does not do. If every channel's cohorts get worse after the same date, look at prices, product releases and service before looking at acquisition.
  • Separate early from later churn. Cancellations in the first periods are where acquisition quality shows. Later cancellations are more often about the product, the price or the service, and belong to whoever owns those.
  • Separate voluntary from involuntary churn. Failed payments are a billing problem with a billing fix, and should not count against a channel. The churn and retention metrics guide defines both, along with Churn Rate and revenue churn. The tutorial on recovering failed payments takes that billing fix through from the customer's side: the messages, the recovery window and how the recovered share is read.

A useful addition is churn by campaign message. If the customers who arrived through a promise about one feature cancel early, the evidence points at the message, which marketing can change.

A subscription measurement plan

The table is the subscription section of the measurement plan, ordered from the earliest signal to the final outcome. A KPI tree connects the rows: revenue retained from a cohort at the top, fed by first payments and churn, fed in turn by trials and activation.

QuestionMeasureSource systemGrainRead when
Is the channel reaching people?Trial starts and cost per trialAnalytics and ad platformsChannel, campaign, weekWeekly
Do trials use the product?Activation rateProduct dataChannel, trial cohortAt the end of each trial cohort
Do trials pay?Trial-to-paid rate; cost per paying customerBilling, joined by account identifierChannel, trial cohortWhen the cohort's trials have ended
Do payers stay and spend?Revenue per trial at fixed agesBilling, joinedChannel, trial cohort, ageAt each agreed age
When does the channel pay back?Months to recover acquisition cost on contributionBilling, finance, spend dataChannel, quarterQuarterly
Who leaves early, and why?Early cancellations by channel and message; voluntary versus involuntaryBilling, CRMChannel, cohortMonthly
Do the systems agree?Analytics first payments against billing first paymentsAnalytics, billingWeekMonthly reconciliation

The measurement plan tutorial turns rows like these into definitions, owners and a tracking specification.

Illustrative example — not a real result

Plumquiver Lessons compares channels at three months

Plumquiver Lessons is an invented language-learning app with a fourteen-day free trial and a £9 monthly plan. Its weekly report favours paid social, which brings far more trials at a lower cost per trial than search. The team joins March's trial cohort to billing on the account identifier and reads it at three months.

Measure (March trial cohort)Paid socialSearch
Spend£6,000£4,400
Trials started2,4001,100
Cost per trial£2.50£4.00
First payments216 (9%)209 (19%)
Cost per paying customer£28£21
Still paying in month three130171
Revenue by end of month three£4,626£5,094
Revenue per trial at three months£1.93£4.63

Search produced fewer trials but more revenue from them, at a lower cost per paying customer. Paid social's early cancellations cluster in one campaign whose ads promised fluency in a month. The team keeps paid social running with the claim removed, moves part of its budget to search, and agrees to read the same comparison at six months before changing the split again.

Common mistakes

  • Judging channels on sign-ups or trial starts. The cheapest trials are rarely the cheapest paying customers.
  • Reporting revenue from analytics. It misses renewals, refunds and plan changes; billing is the record.
  • Comparing cohorts of different ages. A young cohort always looks worse; compare at the same age.
  • No shared identifier. Without an account identifier in every system, the join falls back to guesses on dates and email addresses.
  • Using an email address as the identifier. It leaks personal data into analytics; use an internal number.
  • Blaming acquisition for all churn. Separate early from later, and voluntary from involuntary, before assigning it to a channel.
  • Payback on revenue instead of margin. A channel that repays its cost in revenue may never repay it in contribution.
  • Treating projected lifetime value as measured. Label projections until the cohorts have aged.

The measurement planning hub places this guide beside the measurement plan and KPI tree it builds on.

Frequently asked questions

Should a subscription business send revenue to analytics at all?

Send the first payment as an event with its value, because campaign reports and bidding need a revenue signal close to the click. Do not treat analytics as the revenue record. Renewals, refunds, failed payments and plan changes happen in the billing system, usually without a visit, so analytics revenue will always be incomplete. Report revenue from billing, joined to the acquisition source by the shared customer identifier, and use the analytics figure as the fast signal it is.

How long should we wait before judging a channel on revenue?

Long enough for the cohort to reach the age you have agreed to compare at, and no longer than the decision can wait. Judge early on first payments, which arrive at the end of the trial, and confirm on revenue per trial at a later fixed age that suits the billing cycle. Set the ages in the measurement plan before the campaign starts, and compare channels only at the same age, because a young cohort always looks worse than an old one.

Is high churn among customers from one channel a marketing problem?

Sometimes. If customers from one channel leave early while customers from other channels in the same months stay, the channel is probably attracting people the product does not suit, or promising something it does not deliver. If customers from every channel start leaving after the same date, the cause is more likely a price change, a product change or a service failure. Compare channels within the same cohort before assigning blame.

Terms used on this page

Tools to try

Frameworks

Where this page is used

Sources

  1. [GA4] About events — Google Analytics Help
  2. [GA4] Set up BigQuery Export — Google Analytics Help

Related