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

Customer Experience, Retention & Loyalty › Troubleshooting

Customer Churn Rate Increasing

Why more customers are leaving: a definition that moved, cohorts hiding the loss, onboarding that misses first value, a broken promise and failed payments.
Troubleshootingintermediate17 min readUpdated

First, check the definition and the base, then read churn by cohort and channel

Start with the figure. The churn and retention metrics guide gives the customer churn rate as customers lost during the period divided by customers active at the start of the period, with three rules a rising series often breaks: the base is the customers at the start of the period, so customers acquired during it do not dilute the figure; the period is reported beside the rate, because monthly and annual rates are not comparable; and voluntary churn, where the customer chose to leave, is kept apart from involuntary churn, where a payment failed. Ask whether every period in the series was built the same way. A dashboard rebuilt in the spring, a finance team that now counts paused subscriptions as cancelled, or a retailer whose inactivity window shrank will each produce a rise that is partly or wholly arithmetic. The churn rate definition is the reference for what belongs in the numerator and the base.

Then change the view. A blended rate across all customers is dominated by long-standing loyal customers and can hold flat while each new group is retained worse than the last; the loss appears only when those newer customers are a large share of the base, by which time it has run for months. The metrics guide's answer is a cohort view, grouping customers by when they were acquired and following each group at the same age, split by acquisition channel, first product or campaign so a source that brings customers who leave is visible on its own. Read the latest cohorts against older ones at the same age, then look along the diagonal for a calendar period in which every cohort worsened at once, which points to an event rather than a mix change.

The KPI tree turns the finding into a question about one branch. The metrics guide's retention tree puts the outcome at the top, splits customers retained into voluntary and involuntary churn, hangs the drivers from the diagnosis beneath voluntary churn and the leading indicators beneath each driver, with an owner for every driver: a rise in the involuntary branch has a billing owner, a rise in early activation an onboarding owner, a rise that tracks complaints a service owner. This page sits beside the customer retention strategy guide, which owns the full diagnosis of why customers leave and the four levers, and the customer retention audit scorecard, which rates the whole retention set-up; it starts earlier than either, from a figure that has risen, and asks whether the figure is real before any lever is chosen. Where the business sells trials, the diagnosis for trial users not converting to paid starts from the trial instead.

Diagnosis table

Diagnosis

SymptomCheckLikely causeFix
The rate stepped up in the period a report, a system or an owner changed, while cancellations and complaints did notHow the numerator and the base were built in each period, and whether paused, lapsed and failed-payment customers were counted the same way throughoutThe churn definition or the base changed, so the figure moved before any customer didWrite one definition with a start-of-period base, restate the series with it and hold it where every team reads it
The blended rate drifted up slowly while nothing changed for existing customersRetention by acquisition cohort at the same age, split by channel and offerRecent cohorts retained worse than older ones, hidden until they became a large share of the baseTrace the worse cohorts to their source and fix or cap it; set the retention KPI to a cohort cell
Losses cluster in the first weeks and grow in step with acquisitionWhere new customers stall between paying and getting what they paid for, in bookings, deliveries and first-visit recordsNew customers never reaching first value, so early loss rises with acquisitionDefine first value, remove the steps that stall customers and set a reach-out for each stall
Cancellation reasons and complaints say the experience was not what the marketing saidThe promise register: what each channel promises against the step where customers test itA marketing promise the experience does not keep, so customers feel misled after purchaseFix the step or change the claim, with an owner and a date, and tell every team that repeats the promise
Leavers had a failed delivery, a billing error or a complaint in the weeks before they wentComplaints and repeat contacts tagged by root cause, and whether each was acknowledged, fixed and followed upService failures that were never recovered, each one a customer who decided to leaveRun the recovery sequence with defined remedies, and take the largest root causes to their owners monthly
A large share of recorded losses are expired or declined cards, or the worst cohorts all came from one channel or deep offerInvoluntary losses apart from voluntary ones; cohort curves by channel and acquisition offerFailed payments counted as churn and never retried, or marginal customers from saturated channelsRetry payments with a reminder and report them separately; split channel curves and stop buying customers who never stay

Fix, verify and prevent each cause

The definition or the base changed before any customer did

The metrics guide is blunt about why churn is argued over: it depends completely on definitions that are easy to get wrong and easy to change quietly. What counts as a customer, when a pause becomes a cancellation, and whether a customer who buys once a year is active in a month they did not buy are all choices, and different teams make different ones. A rate whose base once included customers acquired during the period and now does not will rise; a retailer whose inactivity window was shortened will see lapsed customers appear from nowhere.

Fix: use the definition the guide sets out, customers lost over customers active at the start of the period, with the period stated and voluntary and involuntary losses apart; restate every period of the series with it, so the comparison is like with like; and hold the definition where finance, marketing and product all read it rather than in one analyst's query.

Verify: rebuild the last several periods from raw customer records with the written definition and confirm the rise survives; if the restated series is flat, the problem was the figure and the investigation stops here.

Prevent: give the definition a page beside the dashboard, record every change to it with the date and the reason, and treat a change to the base, the window or the lapse rule as a reviewable change that restates history rather than a tweak to one report.

Recent cohorts retained worse, hidden by the blended rate

The cohort analysis tutorial explains the hiding: the blended retention rate most dashboards report is a weighted average down a column of the cohort table, with the weights set by cohort sizes, which is why it rises whenever acquisition grows, regardless of whether retention changed, and a worse-retained group of newer customers stays hidden inside it. The tutorial's reading rules locate the cause. A steady decline down a column, across cohorts at the same age, suggests the acquisition mix is drifting towards worse-fit customers, often because paid channels are saturating; a step change in a single cohort points to something in that acquisition period, such as a promotion that pulled in bargain hunters; and the attributes to split by are the ones that held at acquisition, channel first.

Fix: build the table on the first paid event with channel and offer recorded at acquisition, find the cohorts and the source behind the fall, and act on the source, whether that means capping a discount, changing a channel's targeting or annotating a promotional cohort and excluding it from projections.

Verify: the next complete cohort from the changed source sits level with older cohorts at the same age, and the blended rate stops drifting once that cohort is a fair share of the base.

Prevent: define the retention KPI as a specific cell of the table, such as retention at a chosen age for the latest complete cohort, so the figure cannot improve because acquisition grew; refresh the table monthly with the incomplete period masked; and record each comparison and its explanation so the question is not re-asked next quarter.

New customers never reach first value

The tutorial on onboarding customers for services and products treats the period between paying and getting what was paid for as where most avoidable early loss happens, and the retention strategy guide calls early loss the largest and cheapest opportunity in most businesses, because a new customer has formed no attachment and any friction is decisive. When this is the cause, the cohort curve shows a steep early drop and the losses grow in step with acquisition.

Fix: follow the tutorial's steps. Define first value in the customer's own terms, then draw one customer's path from purchase to that moment; the tutorial draws on Nielsen Norman Group's introduction to journey mapping, which describes a journey map as a visualisation of the process a person goes through to accomplish a goal, and here the goal is first value. List every step the customer must take or know, remove or shorten the ones that stall them, time each message to the moment it is needed, give the first person the customer meets a short checklist beyond the job, then decide what a stall looks like in the business's own records and set a reach-out for each one.

Verify: compare the share of new customers reaching first value, and the early churn of the next cohort, with a similar earlier period, and talk to a few new customers again.

Prevent: keep the path, the step list and the stall rules in one document with an owner, check the records for stalls on a fixed day each week, and review the document whenever the service, the product or the promise changes.

A promise the experience does not keep

Customers judge the experience against what they were told, and the tutorial on checking marketing promises against the experience makes the point that a customer can feel misled even when every step worked as designed, because the promise was in a photograph, a tone of voice or a word like easy. The cancellation reasons and complaints that accompany this cause say the product, the delivery or the result was not what the marketing led them to expect, and the losses often concentrate in a cohort acquired by one campaign.

Fix: run the promise check. Collect the promises from every channel, add the ones customers infer, find the step where each is tested, gather evidence from complaints, returns, reviews and contacts tagged to that step, and judge each promise kept, partly kept or broken. Fix the step when the promise is why customers choose the business and the fix is affordable; change the claim when it cannot be kept reliably; and correct or withdraw a broken objective claim, such as a stated time, price or result, through the approvals process rather than softening it.

Verify: the complaints and cancellation reasons tagged to that step fall in the next cohort, and the register shows each decision done by its date.

Prevent: give every decision an owner and a date in the promise register, tell every team that repeats the promise, including agencies, sales scripts and saved replies, and add a short promise check to campaign planning so a new campaign's promises are tested before launch.

Service failures that were never recovered

The service recovery and complaints guide describes a complaint as a customer who chose to tell the business rather than simply leave, and the metrics guide lists repeat contacts about the same issue, unresolved complaints, long waits for a reply and failed deliveries among the service indicators that precede churn. When this is the cause, the customers who left had a failed delivery, a billing error or a poor contact in the weeks before, and the recovery either never happened or stopped at a refund.

Fix: run the guide's recovery sequence for every failure: acknowledge quickly and specifically, apologise without excuses, understand what the customer needs, fix the original problem, make proportionate amends, follow up to confirm the fix worked, and record the root cause. Give front-line staff defined remedies for the common failures, discretion within a written limit and a fast escalation route, so no customer waits for a permission that signals nobody is in charge.

Verify: read the retention of customers who complained after the change against those who complained before it, and check that repeat contacts about the same issue are falling.

Prevent: tag every complaint by root cause rather than symptom, count the causes monthly and take the largest to the people who can remove them, with marketing in that review because many complaints trace back to a promise, and keep the recovery measures on the retention dashboard rather than only on the support team's.

Failed payments never retried, or marginal customers from saturated channels

Two causes share this row because neither is fixed by an experience lever. The metrics guide separates involuntary churn, accounts closed after a payment failure, from voluntary churn, and notes that it is usually cheap to fix and often unmeasured; the retention strategy guide's diagnosis gives it an operations owner with retries and reminders as the lever. The tutorial on recovering failed payments walks that lever through from the customer's side, from the first service message to the account that is never fixed. The retention and payback guide owns the other half: newer cohorts often retain worse because the acquisition mix changed, as paid channels scale into weaker-fit customers and promotions pull in bargain hunters who lapse when the offer ends, so a falling plateau on the cohort curves is read by splitting the table by channel and offer before concluding anything about the product.

Fix: report involuntary losses on their own line, retry failed payments and send a reminder that lets the customer update their card, and count a customer as lost only after the retries have run; then read each channel's cohort curve at the same age against its acquisition cost, and stop or cap the channels and offers whose customers were never going to stay, which the payback guide calls often the most effective retention intervention once the segment curves are known.

Verify: involuntary losses fall and voluntary churn, read on its own, tells a clearer story; the next cohort from the remaining channels sits level with older cohorts at the same age.

Prevent: keep voluntary and involuntary churn as separate branches of the retention tree with separate owners, read channel and offer curves whenever acquisition spend moves, and agree a payback ceiling per channel with finance so spend is not increased into a channel whose customers leave.

Prevention: one written definition, a monthly cohort read and leading indicators with owners

Every cause above was visible earlier than it was found: the definition changed without a record, the worse cohorts were small for months, the stalls sat in the booking system, the promise was tested at a step nobody inspected, the complaints were closed without a cause, the failed payments were counted and forgotten. The routine that prevents the next rise is three habits with named owners.

The first is one definition. Write the retention measure once, with the base, the period, the lapse rule and the treatment of pauses and failed payments stated; agree it across finance, marketing and product; and hold it where every report reads it. Any change to it is recorded with the date and the reason and restates the history, so the series is never compared across two definitions. The measuring customer experience hub sets this behavioural measure beside perception and operational measures and asks that the three be read together, which is the second reason to write it down: a churn figure says how many left, never why, and the reasons come from the other two.

The second habit is the monthly cohort read. Refresh the table, add the new cohort, extend every row by a column, mask the incomplete period, and read it three ways: down the column at a fixed age for drift across cohorts, along the newest rows for the early drop, and along the diagonal for a calendar event. Split by channel and offer every time, because that is where a saturating channel or a deep discount first shows, and it is the reading that separates newer customers being worse from everyone getting worse this month. Keep the retention KPI as a cohort cell, with repeat purchase rate or active-customer rate beside it where the business has no cancellation event, and note each comparison and its explanation in the same place each month, as the cohort tutorial asks.

The third is a short list of leading indicators with owners. The metrics guide groups them as behavioural, such as a lengthening gap since the last purchase, falling usage, a downgrade, a removed payment method or a visit to the cancellation page; service, such as repeat contacts, unresolved complaints and failed deliveries; and perception, such as low transactional scores, which count only once they have been shown to predict behaviour in the business's own data. Choose the few that predict churn here, by comparing what customers who left did in the weeks before leaving with what similar customers who stayed did, and drop any that do not. Give each indicator an owner and a response, so a flagged customer gets a lifecycle message, a call or an account review rather than a line on a report. The exit question that the guide to designing the cancellation experience asks of every leaver feeds the same list with reasons in the customer's words, and a relationship score is read alongside it rather than instead of it, which is why the diagnosis for a relationship survey score that is falling starts by asking whether the score was ever shown to predict leaving.

The economic case for the routine is the one the retention strategy guide takes from Harvard Business Review's long run of articles on customer loyalty: keeping a customer is usually cheaper than replacing one, and retained customers tend to buy more over time. As the retention and payback guide puts it, an improvement in early retention is applied to every customer the business acquires from then on, which raises lifetime value for every later cohort. When several habits are missing at once, score the whole set-up with the customer retention audit scorecard before fixing anything; when the rise sits inside a wider stall, where churn is cancelling acquisition, the diagnosis for marketing growth that has plateaued reads the two together; and if the business reached for rewards before fixing the reasons customers leave, the diagnosis for a loyalty programme that is not increasing retention explains why the programme changed nothing. The customer experience troubleshooting hub holds the other symptom pages of this collection.

A meal delivery service finds the cohort behind the rise (illustrative)

Illustrative example — not a real result

Zorrabel Meals

Zorrabel Meals, an invented prepared meal delivery service with about 11,400 active subscribers, reports monthly churn rising for four months, from 5.2% to 7.9%. The retention lead starts with the figure. Until March the dashboard's base had included customers who joined during the month, which flattered the earlier periods; a report rebuild moved new months to a start-of-month base without restating history, so about a third of the apparent rise is the correction. Restated on one definition, the series still rises, from 6.1% to 7.9%.

The cohort view, split by acquisition channel, explains the rest. Subscribers acquired through a new half-price first-box partnership that began in April retain 46% after one order against 74% for the other channels, and by July they are a fifth of the base, while cohorts from before April hold level at the same age. Separating involuntary losses shows that 24% of the recorded leavers in the latest month were declined or expired cards that were never retried, and none of them had cancelled.

The service restates the series on the start-of-month base, caps the partnership at a quarter of its previous volume while the offer is reworked, retries failed payments twice with a reminder that lets customers update their card, and reports involuntary losses on their own line. It adds a weekly read of first-order repeat by cohort and channel, owned by the retention lead, with a note beside each change. Two months later voluntary churn reads 5.4% on the restated definition, 61% of the failed payments have recovered, and the partnership cohort, now acquired on a smaller discount, retains 71% after one order.

Common mistakes

  • Reading the series across two definitions. A rebuilt base or a shortened lapse window produces a rise that no customer caused; restate history before reading it.
  • Trusting the blended rate. It improves whenever acquisition grows, whatever happens to retention; the KPI belongs on a cohort cell.
  • Sending offers before the diagnosis. A discount treats the symptom, rewards the threat to leave and blurs the next cohort reading.
  • Counting failed payments as decisions. Involuntary losses have a billing owner, a retry and a reminder, not a retention campaign.
  • Blaming the product for a mix change. Split the cohorts by channel and offer first; a saturating channel looks like a worse product.
  • Closing complaints without a cause. A refund without a tagged root cause leaves the failure to recur at the same rate.
  • Treating the fix as finished when the rate falls. Without the written definition, the monthly cohort read and indicators with owners, the next rise waits for the next review.

Frequently asked questions

Churn has risen for several months in a row. Is that enough to say the experience got worse?

Not yet. A rising series can be produced by a change in how the figure is built, by a growing share of newer customers who were always retained worse, or by failed payments that nobody retried, and none of those says anything about the experience of a customer who chose to stay or leave. Check that the definition and the base are the same in every period, split voluntary from involuntary loss, then read retention by acquisition cohort at the same age. If recent cohorts are worse than older ones at the same age, something changed for new customers; if every cohort worsened in the same calendar period, something happened to everyone at once. Only then is it worth asking which lever to pull.

Should we send an offer to customers who look likely to leave while we investigate?

Resist it until the cause is known. An offer sent before the diagnosis treats the symptom, teaches customers that threatening to leave is rewarded, and muddies the cohort reading you are about to rely on, because retained customers in the next period will include people who stayed for the discount rather than for the product. If the losses turn out to be failed payments, the right action is a retry and a reminder, not a discount; if they are new customers who never reached first value, the fix is in the first weeks, where an offer arrives too late to matter. Keep a holdout when any retention campaign does run, so the business can tell whether it worked.

How do we keep a rise from going unnoticed for months next time?

Put the retention measure on a cohort cell rather than a blended rate, so it cannot be flattered by acquisition growth, and give it one written definition that every team reports from. Read the cohort table monthly, by channel and offer as well as by month, so a worse-retained source shows up while it is still a small share of the base. Then watch a short list of leading indicators that have been checked against what customers actually did, each with an owner who acts when it moves. A rise that is caught in a cohort column two months in is a problem to fix; the same rise found in a blended rate a year later is a lost base.

Terms used on this page

Frameworks

Where this page is used

Sources

  1. Harvard Business Review — Harvard Business Review
  2. Journey Mapping 101 — Nielsen Norman Group

Related