How to increase customer lifetime value: choose a lever and measure it

To work out how to increase customer lifetime value (LTV), look at five levers: retention, purchase frequency, order value, margin and acquisition quality. Where a model multiplies order value, frequency and margin, a gain in one of them raises LTV by the same percentage. The same gain in retention adds more the higher your retention already is.

So your own order data, not an industry average, decides which lever to pull first. And an intervention's effect only shows when you compare it with a random control group that didn't get it.

The definition of LTV and how it differs from other metrics are in the guide to customer retention metrics, churn and LTV. The calculation itself lives in the site's calculators.

Each lever maps to a different number in your order data. The numbers are only comparable when they share one customer definition and one period. For example, use the last 12 months and customers whose first order is at least a year old. Treat returns and cancellations the same way in every number.

• Retention: The share of one year's customers who order again the next year. The margin-adjusted model in the site's LTV calculator uses its complement, annual churn.
• Purchase frequency: Orders per active customer per year.
• Order value: Average value per order after returns and cancellations, also called average order value (AOV).
• Margin: The gross profit share left from an order after product cost, discounts, shipping and returns.
• Acquisition quality: The same four numbers split by acquisition source, plus each source's customer acquisition cost (CAC).

The first four levers change the value of the customers you have today. The fifth changes the value of the cohorts you acquire from now on. You only see its result once new customers complete the measurement window.

The levers also overlap: one intervention often pulls one lever while it pushes another. A discount, for example, can lift frequency and cut margin.

Take a hypothetical store with an order value of $60, 4 orders per customer per year, a 50% gross margin and 60% annual retention.

The margin-adjusted model of the site's LTV calculator values this customer at $300. Improve each lever by 10% on its own, with the other inputs fixed, and retention adds more value than the others:

Order value, frequency and margin grow each year's gross profit by the same share, so all three give the same result. A retention gain means more customers stay each year. Each customer who stays keeps ordering in later years, so the effect builds from year to year.

Hypothetical example: starting retention changes the result

The value of the same relative retention gain depends on how high retention is today. Run the same store at three starting retention levels and check the result with a five-year horizon, and the ranking shifts:

In a business with 30% retention, a 10% retention gain lifts LTV by less than half as much as the same gain in order value.

In this example, the retention lever beats order value above a starting retention of about 48% with no horizon. With a five-year horizon, the threshold is about 54%.

The no-horizon model grows without limit as retention approaches 100%. That is why the two columns split so far apart in the 80% row. If your retention is high, test the decision with a horizon too.

Hypothetical example: a cheap source can leave less value

Say two acquisition sources bring customers with the same order value, frequency and margin, but different retention. A customer from source A costs $100 to acquire and one from source B costs $150:

Source B costs 50% more per customer. Yet after acquisition cost it leaves $250 of gross profit per customer, against $140 for source A.

LTV is 2.40 times acquisition cost for A and 2.67 times for B. Shifting budget on cost per customer alone would pick the source that leaves less value in this example.

Hypothetical example: a discount can lift frequency and lower value

An intervention that pulls one lever can push another. If the same store gives up 5 margin points for a discount that lifts frequency by 10%, LTV falls by 1%.

In this example, the margin loss has to stay under 4.55 points for the 10% frequency gain to hold LTV steady. Bundles raise a similar question: if baskets grow while the order count falls, LTV can stand still. You only see which way it moves when you measure both inputs together.

Sunil Gupta, Donald Lehmann and Jennifer Ames Stuart used public data from five companies in their customer valuation study. It was published in the Journal of Marketing Research in 2004. A 1% gain in retention raised customer value by 2.45% to 6.75%, depending on the firm. A 1% gain in margin raised it by about 1%.

A 1% improvement in acquisition cost moved customer value by only 0.02% to 0.32%. The authors note that the higher a company's current retention, the higher the impact of improving it. That's the same direction as the second table in the hypothetical example.

The finding rests on these five firms' quarterly data up to March 2002. The authors say they left out the cost of improving retention or margin and ignored interactions among acquisition, retention and margins. Your own data still decides which lever leads in your business.

An LTV intervention becomes measurable when you split customers at random as they qualify for it. One group gets the intervention, and the control group doesn't. Compare the groups by assignment, not by who opened the message. Give both groups the same follow-up window, and write down what the control group still receives in that time, such as order and shipping notifications.

LTV builds over years, and a test can't wait that long. So pick gross profit per assigned customer as the main metric and measure it over a window you set in advance. Choose the window from your customers' usual time between orders, and leave out customers who haven't completed it yet. Group size and long-running holdouts are covered in the guide to control groups and holdouts.

Write the decision rule before the test. Say what difference in gross profit per assigned customer counts as a win, and which guardrail stops the test at what limit. If the result is unclear, report it as unclear, not as no effect.

A hypothetical decision rule could read like this. After 180 days, gross profit per assigned customer must be at least $5 above the control group. The gap in unsubscribe rate must stay within 0.5 points. If both hold, roll the intervention out to every eligible customer.

You set the thresholds from the cost of the intervention. What matters is writing them down before you see the results.

The flow pages in the journey library define each flow's own control-group metric, such as the incremental second-purchase rate. At the lever level, add two measures: gross profit per assigned customer and the other levers the flow is assumed not to touch.

Retention: customers drifting away and customers already lost

You can pull the retention lever at two points. The tiered churn intervention flow is for customers whose churn risk crosses a threshold. The lapsed customer win-back flow is for former customers whose paid relationship has ended. The flow pages measure incremental usage recovery and incremental win-back against a control group.

For LTV, the real question is whether returning customers stay and whether they cover the cost of the incentive. Compare gross profit per assigned customer, net of incentive cost, in both groups, and track the 12-month result next to the short window. Use unsubscribes and complaints as guardrails.

Purchase frequency: the second order

For purchase frequency, the first step to measure is the move from a first order to a second. The first-to-second purchase flow aims to bring first-time buyers back around the natural time they are likely to need another purchase. It defines its own success measure as the incremental second-purchase rate against a control group.

For this lever, the main metric is orders and gross profit per assigned customer over the same window. If you use a discount, track margin separately: as in the hypothetical example, orders can rise while value falls.

Order value: complementary products and shipping thresholds

Interventions that grow order value either add an item to the basket or set a threshold. The flow that offers a verified complementary product is the cross-sell and complementary offer page. Whether a threshold pays for itself is worked out in the guide to free shipping thresholds.

Order value alone misleads, because it can rise while fewer customers order. Randomize by customer, not by session, so the same person doesn't see two offers. The main metric is gross profit per assigned customer, and the guardrail is the share of customers who order within the window.

Margin: discount depth and cost per order

Margin interventions work on discount depth or on cost per order: a shallower discount, fewer returns, cheaper shipping. In the test, the control group keeps today's discount and the test group gets the shallower one.

If margin rises while retention falls, the gain is smaller than it looks. So the guardrail is the share of customers who order again within the window. Write the decision rule in gross profit, not in margin rate: a higher margin rate that comes with fewer orders can leave less per customer.

Acquisition quality: the customers each source brings

One way to compare acquisition quality by source is Google Analytics 4 (GA4). Its user lifetime exploration can show the source, medium and campaign that drove users with the highest lifetime revenue. It also gives the average LTV of users acquired by each first user medium. That figure is revenue, not gross profit.

GA4's help page says LTV percentiles are often zero because most users don't buy. Filtering to purchasers only gives a more useful view of the distribution.

A gap between sources is still an observation, since sources also differ in audience, offer and timing. Gupta and his co-authors consider it quite likely that some acquisition programs, such as price promotions, attract customers with low retention. They state this as an expectation, not as a measured result.

To see whether a targeting or offer change brings more valuable customers, use your ad platform's experiment feature. Google Ads custom experiments share the original campaign's traffic and budget, test the change and compare its performance with the original campaign. For Search campaigns you can split by cookie or by search. As of November 2026 the feature covers Search, Display, Video and Hotel Ads campaigns.

Tag first orders from the experiment arm and from the original campaign as separate cohorts. Then compare their gross profit per customer over the same window. To see what acquired customers order months later, track them in your own order data. To read LTV per source against acquisition cost, use the LTV:CAC ratio calculator.

Single tests show each intervention's effect within its own window. They don't show whether customers are worth more across all your messages combined.

Some customer relationship management (CRM) platforms offer a permanent group kept out of every campaign for this. Braze's Global Control Group documentation describes holding a percentage of all users out of campaigns and Canvases. The group is there to analyze the overall impact of messaging over time.

As of November 2026, the same page says users in this group still receive campaigns triggered through the application programming interface (API). You can also add exclusion settings for transactional messages that should reach everyone. Comparing this group's 12-month gross profit per customer with that of messaged customers shows the program's total contribution to LTV.

The hypothetical example assumes the same retention every year. ChartMogul's LTV page is about subscription businesses. It says the basic formula assumes churn is linear, while real churn is highest in the first few months, which overstates LTV.

In ecommerce too, the step from a first to a second order may differ from later years. If it does, measure that first step as its own lever. These two sentences are this post's inference, not the source's.

The model doesn't discount future gross profit. The LTV calculator's page also notes that none of its models discounts future revenue for the time value of money. So a long lifespan inflates the figure. That's why the five-year check matters for a business with high retention.

The average customer isn't a real customer either. The same relative gain plays out differently for a loyal frequent buyer and for a one-order customer. Scoring customers by recency, frequency and monetary value (RFM) is the subject of the RFM analysis guide.

Finally, improving every lever by the same share is a convenience of the calculation. Only your own tests tell you how far each lever can move.

Repeat the example's calculation with your own numbers to choose the first lever:

1. Pull retention, frequency, order value and margin from the last 12 months with one customer definition, and split them by acquisition source.
2. Change each input on its own by a realistic amount. Take that amount from your own controlled tests, not from other companies' averages.
3. Add the intervention's cost and its side effects on the other levers: what a discount takes from margin, what a bundle takes from frequency.
4. Start with the lever that raises LTV most and whose effect you can measure against a control group within your chosen window. Write the decision rule before the test.

Run every scenario through the margin-adjusted model of the LTV calculator with the same inputs, and the results stay comparable.