How to Measure Customer Reactivation: From Attributed Revenue to Incremental Lift
Your email platform says the win-back flow generated $80,000.
Did it?
Maybe.
It certainly attributed $80,000.
That is not quite the same thing.
Some of those customers may have bought anyway.
Some may have already been due to reorder.
Some may have seen paid ads.
Some may have received another campaign.
The dashboard answers:
Which touchpoint gets credit?
The commercial question is:
What actually happened because we intervened?
That is incrementality.
Attribution is a credit system
Most marketing systems work with attribution rules.
For example:
- clicked an email,
- purchased within the attribution window,
- order credited to email.
That is operationally useful.
But it is not causal proof.
Google makes this distinction explicitly in its Conversion Lift documentation.
Standard attributed conversions are counted according to campaign attribution rules.
Incremental conversions are measured by comparing a treatment group that saw the campaign with a control group that did not. (support.google.com)
That distinction belongs in customer reactivation too.
Why reactivation is especially vulnerable to over-attribution
You are marketing to previous customers.
They already know you.
Some were likely to purchase again naturally.
Imagine someone normally orders every 45 days.
You send them 20% off on day 44.
They buy on day 45.
The platform says:
WIN-BACK SUCCESS.
But did we recover them?
Or did we simply give a discount to an active customer?
Without a counterfactual, we do not know.
Start with reactivation rate
A basic metric:
Reactivation rate
=
Reactivated customers / eligible customers
Example:
10,000 eligible customers
600 purchase
Reactivation rate:
6%
Useful.
But incomplete.
Revenue per recipient
This helps compare campaigns of different sizes.
Revenue per recipient
=
Revenue / recipients
If:
$50,000 revenue
10,000 recipients
RPR:
$5.00
Now compare another segment:
$20,000 revenue
2,000 recipients
RPR:
$10.00
The smaller cohort is generating twice the revenue per recipient.
That can be commercially important.
AOV after recovery
Suppose recovered customers have:
$65 AOV
while their historical AOV was:
$92.
Why?
- deep discount?
- lower-ticket products?
- weaker customer quality?
Revenue is not enough.
Context matters.
The holdout group
This is one of the cleanest tools available.
Imagine:
Eligible cohort:
10,000
Randomly assign:
9,000 treatment
1,000 holdout
Treatment receives the recovery intervention.
Holdout does not.
After 30 days:
Treatment purchase rate:
5.4%
Holdout:
2.2%
Estimated incremental lift:
5.4% − 2.2% = 3.2 percentage points
Estimated incremental orders:
9,000 × 3.2%
= 288
If AOV is $85:
288 × $85
= $24,480 estimated incremental revenue
That is much more informative than counting all:
486 treatment purchases
as campaign-created.
Incrementality is about the counterfactual
The central question is:
What would have happened if we did nothing?
That is impossible to observe for the same individual at the same time.
So we approximate it using a comparable control group.
This is why randomization matters.
Google describes Conversion Lift in exactly this way: comparing treatment users with users deliberately held back from exposure to measure causal lift. (support.google.com)
Holdouts are not exotic
Optimizely uses holdouts extensively in experimentation and personalization. Its standard personalization setup can hold back a proportion of the audience so that the personalized treatment can be compared with a generic experience. (support.optimizely.com)
The principle is simple.
Do not ask only:
Did treatment customers buy?
Ask:
Did they buy more than comparable untreated customers?
Sample size matters
A holdout of:
10 people
is almost useless for most ecommerce purchase outcomes.
If:
1 person buys
the rate becomes:
10%.
If nobody buys:
0%.
Huge apparent difference driven by one person.
Small samples create noisy conclusions.
So the test has to balance:
- learning,
- risk,
- available cohort size.
Retention outcomes take time
Optimizely points out that retention tests are harder than simple same-session conversion tests because the outcomes can take weeks or months to appear and can be affected by many intermediate touchpoints. (optimizely.com)
This is equally true in ecommerce reactivation.
A customer may recover today.
But the deeper question is:
Did they resume a valuable relationship?
That may take longer to see.
Measure the second order after recovery
A reactivated customer who buys once under a 30% discount then disappears again is different from someone who returns to normal buying behaviour.
Track:
- recovery order,
- next order,
- 60-day repeat,
- 90-day repeat,
- subscription restart.
This tells you whether the campaign created a transaction or restarted a relationship.
Margin changes the answer
Suppose:
Incremental revenue:
$30,000
Gross margin:
70%
Gross profit:
$21,000
Campaign cost:
$8,000
Incremental contribution:
$13,000
Now compare another strategy:
Revenue:
$24,000
Margin:
80%
Campaign cost:
$3,000
Contribution:
$16,200
Lower revenue.
Better economics.
Cost per incremental recovered customer
Formula:
Recovery cost / incremental recovered customers
If:
Campaign cost:
$7,500
Incremental customers:
250
Cost per incremental recovered customer:
$30
Now compare with:
New CAC:
$75
That does not automatically prove recovery is better.
But now the business has a useful benchmark.
Do not use a holdout if you cannot interpret it properly
Holdouts introduce their own challenges.
You need to consider:
- seasonality,
- cross-channel exposure,
- sample size,
- test duration,
- contamination.
If the holdout gets hit by five other campaigns while treatment does not, the comparison becomes messy.
Experimentation does not remove judgement.
It improves it.
Separate primary and secondary metrics
For a Revenue Recapture test, a sensible hierarchy might be:
Primary
Incremental purchase rate
Secondary
- incremental revenue
- revenue per recipient
- AOV
- contribution margin
Guardrails
- unsubscribes
- complaints
- bounce rate
Long-term
- repeat purchase after recovery
Do not let a great click rate distract from poor economics.
Build the measurement plan before the send
This sounds obvious.
It is often not done.
Before launch, define:
- who is eligible,
- treatment size,
- holdout size,
- measurement window,
- attribution window,
- success metric,
- stop conditions.
Once the campaign is live, changing definitions because the results look inconvenient destroys credibility.
A worked example
Priority cohort:
12,000 customers.
Treatment:
10,800
Holdout:
1,200
Treatment purchases:
702
Treatment rate:
6.5%
Holdout purchases:
42
Holdout rate:
3.5%
Incremental lift:
3.0 percentage points
Estimated incremental orders:
324
AOV:
$78
Estimated incremental revenue:
$25,272
Contribution margin:
65%
Estimated contribution before campaign cost:
$16,427
Campaign cost:
$6,000
Estimated incremental contribution:
$10,427
Now you can make a decision.
Not:
“The flow looked good.”
But:
“This cohort appears to generate positive incremental contribution at this intervention cost.”
That is much more useful.
Why I prefer “estimated incremental” language
Even a good experiment has uncertainty.
Unless the test is designed and powered rigorously, avoid pretending the estimate is exact.
Say:
estimated incremental revenue
rather than:
recovered $25,272 exactly.
That level of precision communicates maturity.
Most experiments will not be spectacular
Optimizely’s large-scale experimentation research is a useful reminder that many ideas fail to produce statistically significant improvement; experimentation works because teams learn across many tests, not because every test wins. (optimizely.com)
That is the attitude Revenue Recapture needs.
A failed cohort test is not automatically wasted money.
It can tell you:
- wrong segment,
- wrong timing,
- wrong Return Reason,
- weak product-market logic.
That learning can save a much larger bad rollout.
Frequently Asked Questions
Is attributed revenue useless?
No. It is useful operationally. It simply should not be confused with causal incremental revenue.
Do I always need a holdout?
No. But if you want stronger causal confidence, a holdout helps.
How large should the holdout be?
It depends on audience size and expected conversion. There is no universal percentage that works for every test.
What is a good reactivation rate?
There is no single benchmark. Category, lapse depth, customer history and incentive all affect it.
The question that matters
The campaign dashboard asks:
Who got credit?
Revenue Recapture asks:
Did the customer behave differently because we did something?
That is a much harder question.
It is also the one worth answering.
Related reading
How to Prioritize Lapsed Customers
Customer Reactivation Strategy
Want the opportunity measured before you scale it?
The Revenue Recapture Analysis is designed to identify the first test and the evidence we would want before expanding it.