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

Marketing Analytics › Measurement Planning

New and Returning Customers in Marketing Data

Why analytics new users are not new customers, how to define and record customer status, and how to report acquisition and repeat sales by channel.
Guideintermediate9 min readUpdated

New users are not new customers

The reports and dimensions guide sets out what the user metrics count: active users in most standard reports, total users for anyone who sent an event, and new users for first visits. Which visits are grouped into one user depends on the property's reporting identity setting and on the identifiers available, and for a visitor who never signs in the identifier is a device-level one tied to a browser on a device. A new user is therefore a new identifier, and the comparison of sessions and users explains why user counts drift from the number of real people.

A new customer is a different kind of fact. It lives in the order system or the CRM, and it says that this person or organisation had not bought before. The two diverge in both directions:

  • Existing customers counted as new users. A subscriber who reorders from a work laptop, a customer who cleared cookies, a buyer who declined measurement on earlier visits and accepted this time, someone who switched browser.
  • New customers counted as returning users. A first-time buyer whose earlier sessions over a few weeks happened on the same device is a returning user and a new customer.

Neither error is a fault in analytics. The figure measures devices correctly; it was never designed to measure customers. The e-commerce conversion hub splits conversion rate by new against returning visitors, which is a legitimate visitor-level analysis; the mistake is reading that split as acquisition.

Defining a new customer

Write the definition into the measurement plan once, with an owner, because every later report depends on it. Five decisions make it precise.

  1. What a customer is. An account, an email address in the order system, a household, or, for business-to-business sales, an organisation. Guest checkouts need a matching rule in the order system, run on the business's own First-Party Data and never in analytics.
  2. What counts as the first purchase. The first order, the first paid subscription, the first signed contract or the first invoice, depending on the business.
  3. What happens to cancelled and refunded first orders. A simple rule is that the customer remains new until a first order is kept.
  4. When a lapsed customer counts as reactivated. A third status, with the gap set from the business's own purchase intervals, so that win-back campaigns are credited for winning back rather than for acquiring.
  5. Which channels and stores the history covers. A customer who has bought in a shop but never online is new to the website and not new to the business; decide which question the report answers, and say so.

Counted by month rather than by order, the same statuses plus the customers who stopped buying form the bridge in the growth accounting tutorial, where reactivated customers appear as resurrected and every movement must add up to the change in the active base.

Recording customer status at conversion

The order system knows, at the moment an order is placed, whether the buyer has ordered before. That is the moment to record status, and the order system is the place to compute it. Two ways to get it into the reporting, and the strongest set-ups use both.

Send it with the conversion. The confirmation page or the server pushes the purchase to the data layer with a customer status parameter (new, returning, reactivated) that the order system computed. Google's e-commerce developer guide documents the purchase event and its transaction identifier, value and items; status travels beside them as a custom parameter. The events and parameters guide explains that a parameter has to be registered as a custom definition before standard reports and explorations can show it, while the BigQuery export holds it either way. The advantage is speed: the split appears in analytics reports as orders arrive.

Join it afterwards. The purchase event's transaction identifier is also in the order table, so the two can be joined in a warehouse. Google's BigQuery export documentation covers the link, and the BigQuery export basics guide shows where the transaction identifier sits in the export's e-commerce record. The order table supplies the status, computed with the window functions the SQL for marketers guide uses to find each customer's first order. The advantages are completeness and flexibility: the join covers orders whose analytics event was lost, and the status can be recomputed when the definition changes.

What not to do is infer status from analytics itself: from a returning-user flag, from the first visit event, from the device identifier or from a cookie the site sets. Each of these measures the browser. Customer status is a business fact and comes from the business's records.

Reporting acquisition and repeat sales by channel

With status on every order, each channel gets two lines instead of one.

  • Acquisition: first orders and their revenue, by channel. New customers by channel are the denominator of channel CAC, and the LTV and CAC guide explains why the numerator should be the business's full acquisition cost.
  • Repeat: orders from existing customers and their revenue, by channel. This is retention revenue that a channel touched, not revenue it created, and it is judged differently.

Then follow the acquired customers. Group them into a Cohort by the month of their first order and acquisition channel, and read later revenue and Repeat Purchase Rate at the same ages. A channel whose new customers come back is worth more per customer than one whose new customers do not, and LTV by acquisition channel is the figure that eventually shows it.

Double counting is the trap in this report. One customer can see a social ad, click a search ad and open an email before buying, and every platform involved may count the purchase as its own. Count new customers once, in the order system, and attribute each to one channel by the business's own rule (first touch, last touch or another documented rule), applied consistently. The test is simple: the channel lines must add up to the total number of new customers in the order system. If platform-reported new customers add up to more, that is the overlap, and it belongs in a note rather than in the total.

When paid channels claim returning customers

Some channels touch returning customers far more than new ones by design. Retargeting reaches people who already visited; brand search reaches people who already know the name; email reaches people already on the list; voucher affiliates reach people already at the checkout. Under last-click rules they collect conversions that look cheap, because many of the customers were coming back anyway.

The status split makes this visible. Compare each channel's share of all orders with its share of first orders, as the troubleshooting page on retargeting and brand search recommends. A channel with a large share of orders and a tiny share of first orders is serving existing demand. That can still be worth paying for, but it has to be proved with a holdout or an incrementality test rather than assumed from attribution, and its cost should be judged against repeat revenue it actually added, not against acquisition targets.

Two practical consequences follow. Report channel cost per new customer beside cost per order, so that acquisition budgets are judged on acquisition. And keep audiences for prospecting campaigns separate from audiences for existing customers, using customer lists and exclusions where a platform supports them, so that each budget buys what it is meant to; how each platform handles those audiences is covered in its own section. The existing customer exclusion test is the design for checking that separation by region, counting new customers and existing customers' orders with the status recorded here.

A customer status reporting table

The table is the customer status section of the reporting specification. Each line names its source, because the most common error is taking the right number from the wrong system.

MeasureDefinitionSourceUse it for
New usersFirst visits from a device or browser not seen beforeAnalyticsReach of the site among new devices; never acquisition
New customersCustomers whose first kept order or contract falls in the periodOrder system or CRMAcquisition volume; CAC denominator
Returning customersCustomers with an earlier order who ordered in the periodOrder system or CRMRetention and repeat revenue
Reactivated customersCustomers ordering after a gap longer than the defined lapseOrder system or CRMWin-back campaigns
Customer status on each orderNew, returning or reactivated, computed at the time of the orderOrder system, sent with the purchase or joined on the transaction identifierEvery split by channel
Acquisition revenue by channelRevenue from first orders, one channel per customer by the business's ruleOrder system joined to sourceAcquisition performance
Repeat revenue by channelRevenue from returning and reactivated customers' ordersOrder system joined to sourceRetention touchpoints; judged by incrementality
Cohort revenue by acquisition channelLater revenue of each month's new customers, at fixed agesOrder systemCustomer quality by channel

A KPI tree built on these lines splits revenue into acquisition revenue and repeat revenue near the top, so that a change in either can be traced to its own drivers.

Illustrative example — not a real result

Saltwhisk Coffee Roasters splits its orders

Saltwhisk Coffee Roasters is an invented coffee roaster that sells bags on its website alongside a subscription. Its analytics report shows 1,150 of a month's 1,800 web orders (64%) as purchases by new users, and the team has been calling that acquisition. The analyst joins the purchase events to the order table on the transaction identifier and takes customer status from the order system: only 410 orders (23%) are first orders, and 1,390 come from existing customers, many of them subscribers adding a one-off bag.

Channel (last click)OrdersFirst ordersExisting customersSpendCost per orderCost per new customer
Paid social prospecting22015070£4,500£20£30
Paid search, non-brand260120140£3,600£14£30
Brand search38040340£1,200£3£30
Retargeting30025275£3,000£10£120
Email42015405———
Organic and direct22060160———

Retargeting looked like the second-cheapest paid channel per order; per new customer it is four times the cost of prospecting. The team moves the retargeting budget's acquisition target to cost per new customer, cuts its spend while running a holdout to see how much of its repeat revenue it actually adds, and reports acquisition and repeat revenue as separate lines from the next month.

Common mistakes

  • Reading new users as new customers. New users are new devices and browsers.
  • Inferring status from cookies. Customer status comes from the order system or CRM, not from analytics.
  • No written definition. Without rules for guest checkouts, refunds and lapsed customers, two reports will disagree.
  • One revenue line per channel. Acquisition and repeat revenue mean different things and need different judgements.
  • Adding platform figures. Each platform counts the same new customer; the order system counts it once.
  • Judging retargeting and brand search on cost per order. Their audiences are people who already know the business; judge acquisition on cost per new customer.
  • Forgetting other sales routes. A customer new to the website may not be new to the business; say which question the report answers.
  • Treating reactivated customers as new. Win-back and acquisition are different jobs with different economics.

The measurement planning hub places this guide beside the measurement plan, and the guide to measuring a sale or promotion applies the same status split to the customers a promotion brings in.

Frequently asked questions

Why does analytics say most of our buyers are new when we know most are repeat customers?

Because analytics' new users are new devices and browsers, not new customers. A loyal customer who buys on a new phone, in a different browser, after clearing cookies or after declining measurement on earlier visits arrives as a new user. The analytics figure is correct for what it counts. Customer status has to come from the order system or CRM, which knows whether this person has bought before.

Should customer status be sent to analytics or joined afterwards?

Both, if you can. Sending a status parameter with the purchase event, computed by the order system when the order is placed, makes the split visible in reports and explorations quickly. Joining the order table to the analytics export on the transaction identifier afterwards is the record: it covers orders whose event was lost, and it can be recomputed if the definition of a new customer changes. If only one is possible, the join is more reliable.

How should a reactivated customer be counted?

As its own category, defined in advance. A customer who last bought long ago is neither new nor routinely returning, and campaigns that win them back deserve credit for something different from acquisition. Set the gap that counts as lapsed from the business's own purchase intervals, record reactivated as a third status, and report it separately rather than folding it into either of the other two.

Terms used on this page

Frameworks

Where this page is used

Sources

  1. Google Analytics Help — Google Analytics Help
  2. [GA4] Set up BigQuery Export — Google Analytics Help
  3. Measure ecommerce (GA4 developer guide) — Google for Developers

Related