Ask most ecommerce teams about churn and you will see the same thing: one number on a monthly board slide labelled "90-day non-purchasers." One threshold. One segment. One rate.

Five structural patterns account for approximately 85 to 90% of preventable value erosion across ecommerce ecosystems, validated across 50+ premium brands, benchmarked against Klaviyo's 2025 and 2026 datasets of 167,000+ and 183,000+ accounts respectively, and groa° implementation data. Each pattern has a different cause, a different timing window, and a different economic consequence. Each demands a different response. Collapsing them into one number guarantees delayed response and misdirected spend.

Two definitions before the framework.

Preventable value erosion: loss of margin or lifetime value that emerges from system architecture, not from unavoidable category churn. If the system had detected the signal and intervened in time, the value would have been recoverable.

Governed system: a retention engine that uses live behavioural signals and economic weighting to trigger interventions calibrated to lifecycle position, with every action governed by where each customer actually sits in their journey and what it will cost to leave them there.

This article names every leak, shows how to recognise it in your own data, and explains what the economic cost looks like when it compounds unaddressed.

The leak map

Before the detail, an orientation.

Leaks 1 through 5 map directly to Flywheel orbits 2 through 4: Activation, Value Core, and Loyalty. In practice, a single customer can move through multiple leaks as they drift from Activation toward departure. The map below shows where each one lives and what it costs.

The leaks are structural views, not mutually exclusive customer sets. In practice, many brands will find the same customers showing up across Leaks 2, 3, and 4 simultaneously. That co-occurrence is itself a signal. It means the Value Core has stalled at the system level, and addressing any single leak without the others will produce only partial recovery.

Why churn stays hidden until it is too late

"Churn belongs to the system, not to a team, because no single function governs customer progression end to end."

Consider what that means in a typical brand. Acquisition owns the moment a stranger subscribes. CRM owns the moment they click. Finance owns the moment they buy. There is no owner for what happens between those moments: between subscribe and first purchase, between first and second, between second purchase and the slow cooling that precedes departure. Deterioration travels through those seams without triggering a response. By the time revenue signals it, the intervention window has already closed.

Most dashboards surface churn only after inactivity has hardened into lapse. Tools prioritise traffic, events, and attribution. They explain arrival and conversion and lose continuity once transactions complete. Engagement decay and slowed return cycles sit outside their design.

In groa° OS, these five leaks become live diagnostic views updated at behavioural speed. The framework below describes what they are and what detecting them requires.

Leak 1: One-and-done buyers

Orbit 2, Activation | CAC permanently unrecovered

In a mid-market skincare brand, 71 out of every 100 first-time buyers placed no second order within six months. The acquisition dashboard showed strong conversion. The Activation orbit had stalled entirely.

This is the most common leak and the most expensive. The first purchase completes, the customer never returns, and acquisition cost goes unrecovered. Compounding cannot begin because the relationship never forms.

Acquisition dashboards declare success at the point of conversion. The reporting architecture was designed to measure arrivals. Nobody tracks whether the customer came back.

The signal: track the share of any cohort that fails to make a second purchase within a defined window for your category. When it runs above 70%, Activation has stalled.

Customers who return within the early window of their expected cycle exhibit materially higher lifetime value than those who delay. Faster return means more purchase cycles within any measurement window, earlier acquisition cost recovery, and extended lifetime value from the same acquisition investment. Every week of delay in the Activation window reduces the probability that a second purchase occurs at all.

A governed system does not wait for dormancy to surface this leak. It detects the signal within the Activation orbit's defined window and intervenes with product education and sequenced progression logic calibrated to that customer's entry signals and category context. The intervention arrives when the customer is still receptive, not when the calendar says day seven has passed.

Leak 2: High-risk cohorts

Orbit 3, Value Core | CAC at risk whilst customer appears active

Picture a customer placing orders every eight weeks. Revenue is stable. Then look closer: their unique site visits have dropped 60% in three months, their email open rate has halved, and their last click was on a sale event four months ago. They are still buying. Their orbit energy has already collapsed.

Engagement collapses before purchasing stops. These customers transact, and browsing, response, and loyalty signals fall below healthy bands simultaneously. Revenue appears stable. The Value Core is drifting economically.

The problem is that revenue masks engagement decay. Teams see transactions. Trajectory is invisible. A customer can be generating orders and approaching departure simultaneously, and standard reporting aggregates behaviour and loses the individual signal against that customer's own baseline.

The signal: engagement velocity falling sharply against a customer's trailing history. For high-value customers, stable order counts and flat AOV combined with a 40%+ drop in site visits and email click rate is the pattern. When it appears, the Value Core is losing energy even if purchase frequency has not yet reflected this.

Governed systems respond to this signal before it appears in purchase data. Orbit-aware intervention logic activates to restore progression and protect acquisition cost recovery while the customer is still transacting. The intervention is calibrated to where the customer sits in the Flywheel, what their energy trajectory shows, and what category signals indicate about their likely next move. A generic re-engagement campaign sends the same message to everyone showing this signal. Orbit-aware intervention makes them different problems with different responses.

Leak 3: Frequency droppers

Cross-orbit, Velocity Decay | LTV compression, invisible in revenue

A coffee subscriber who used to order every 28 days is now ordering every 41 days. Revenue this month looks fine. They are now buying 46% less frequently than before. If nothing changes, their 12-month lifetime value will tend to compress by a similar proportion, simply because they complete fewer purchase cycles in the window.

Return cadence slowing beyond expected thresholds is velocity decay. Lifetime value contracts as purchase intervals stretch. The customer remains in the system and is losing orbit velocity.

Static segments cannot detect this. A customer who bought twice in three months and a customer who bought twice in twelve months both sit inside "Placed Order at Least Once" in Klaviyo. One has recovered acquisition cost and is building margin. The other is drifting toward dormancy. The segment treats them identically.

The signal: the relationship between a customer's current inter-order interval and the historical average for their segment and category. When a segment that historically buys every 42 days is now averaging 61 days, you have a velocity problem, even when revenue this month is flat.

Velocity decay is the leading indicator of churn. When second-purchase rates fall from 30% to 22%, or inter-order intervals stretch materially beyond category norms, Flywheel velocity has declined significantly. By the time the revenue line reflects this, recovery costs substantially more than prevention would have.

A governed system fires a replenishment intervention anchored to expected repurchase timing, using live product context and progression logic designed to restore cadence before the interval drifts further. When shampoo buyers in the expected 45-day band reach day 43 without an order, the system queues a replenishment touchpoint and suppresses generic sale campaigns until they repurchase or genuinely drift. That is what governed intervention looks like in practice.

Leak 4: Value decliners 

Orbit 3, Value Core | Margin erosion at stable order count

A customer places five orders in a year. The first was at full price, £85. By the fourth, they are ordering only during promotions and the basket has dropped to £42. Order count is stable. Repeat purchase rate looks fine. Contribution margin on this customer has already declined substantially.

Average order value falls and discount reliance rises. Revenue continues, and contribution margin erodes as baskets shrink and profitable product mix gives way to lower-margin behaviour.

Order count staying constant is the trap. Without margin-aware reporting across the last three to five orders, the customer appears retained whilst profitability drains. Teams report stable repeat purchase rates and miss that the economic value of each repeat is deteriorating.

The signal: AOV trend and promotional dependency together over the last three to five orders. When both move in the wrong direction simultaneously, margin is eroding regardless of what the top-line repeat rate shows.

A governed system detects this pattern and activates a cross-sell journey driven by demonstrated category and basket affinity. The goal is to rebuild AOV and mix quality through higher-margin product adjacencies, protecting margin whilst maintaining engagement. The content and timing are calibrated to the customer's purchase history and stated preferences.

Leak 5: Dormant VIPs

Orbit 4, Loyalty | Profit engine silent attrition

Losing just 200 of your top 2,000 customers can, in many ecommerce P&Ls, erase the profit generated by retaining an extra 5,000 customers from the middle of the file. The economics of churn are segment-weighted. Volume in the middle does not compensate for attrition at the top.

High-value customers disengaging quietly is the most consequential leak of the five. Order rhythm, spend concentration, and advocacy behaviours flatten. The brand's most important profit engine becomes silent attrition.

VIP lists are static labels. A customer tagged "VIP" eighteen months ago may have already churned economically. The label does not re-evaluate itself against live behaviour, so the governance system treats a formerly high-value customer as active long after they have stopped behaving like one.

The signal: a top-decile customer departing from their personal repurchase cadence, calibrated to category and behavioural history. A generic dormancy threshold applied uniformly across all profiles misses this signal entirely. The customer who buys every 25 days and has now gone 35 days is a Leak 5 signal. The same 35-day gap means nothing for a customer with a 90-day cycle.

Each leak can be operationalised in Klaviyo as a live segment built around a customer's own behavioural baseline, combined with a simple set of thresholds, long before any infrastructure rebuild. The architecture that makes it continuous and autonomous is what groa° OS provides. The segmentation logic that exposes it can be started now.

A Flywheel-governed system detects when high-value customers slow, fall behind their expected return window, and begin eroding contribution margin, and triggers intervention while profit is still recoverable.

The compounding problem nobody expects

When Leak 5 goes unaddressed, Leaks 1 through 4 get harder to fix.

High dormancy degrades deliverability across the entire list. Worse inbox placement reduces open rates in every orbit. The engagement benchmarks that detect Leak 2 become less reliable. The velocity signals that identify Leak 3 are harder to read through a degraded engagement environment.

Reactivation failures compound upstream. This is the hidden multiplier most brands miss.

The case data illustrates this precisely. A global fragrance house entered the Flywheel methodology with dormancy at 52%. Deliverability suppression was cascading across every orbit: Capture, Activation, Value Core, and Loyalty. After systematic Reactivation work, the dormant rate dropped from 52% to 31%. Revenue per recipient across campaigns improved 156%. This was not a creative win. It was the structural effect of fixing Leak 5, which restored inbox placement and made orbit signals readable again across all four upstream leaks.

The diagnostic approach: where to look this week

The leaks interact. To see how, run this exercise on a cohort from the last quarter.

Take a group of customers who first purchased in a recent month. Track them forward.

How many placed a second order within the expected window for your category? If fewer than 30%, Leak 1 is active and the Activation orbit has stalled.

Of those who did return, how many have stretched their inter-order interval by 30% or more compared to their first return? That is Leak 3 operating inside the customers who appeared to activate.

Of those still transacting, how many have placed their last two orders on promotional pricing with a falling basket? That is Leak 4 eroding margin while the repeat rate line looks flat.

The revenue line for this cohort may look stable or even positive. The economics are already deteriorating across three simultaneous leaks.

In Klaviyo, the most common segments are behavioural proxies: "Placed Order at Least Once," "Opened Email in Last 90 Days," "Has Not Purchased in 90 Days." These describe states. They do not describe trajectories, economic weight, or distance to recovery.

The five leaks impose economic ordering on that data. They tell you where customers are, which direction they are moving, how fast, and what it will cost if movement continues unchecked.

"Effort cannot resolve a structural mismatch."

Architecture that detects these signals against cohort-level baselines, scores them by economic weight, and prescribes orbit-specific intervention before the leak hardens into permanent loss is what closes the gap.

The takeaway

The five leaks are structural, which means they respond to structural solutions. Detecting them requires architecture that tracks trajectory, not state: customer energy velocity, orbit-level progression, and economic weighting applied in real time against each customer's own behavioural baseline. Static segments describe what has happened. They do not govern what happens next.

The compounding consequence is the one most brands underestimate. Leak 5 does not stay in Leak 5. Dormancy degrades deliverability. Degraded deliverability blunts the engagement signals that detect Leaks 2 and 3. Upstream leaks become harder to read through a degraded sending environment. The five interact, and the interaction is always directional. Fixing one in isolation recovers margin at the margin. Fixing the architecture recovers the system.

The map is now visible. The economic cost of each leak is measurable. The diagnostic question is a single one: does your current system detect these patterns at the speed they move, or does it surface them after the intervention window has already closed?

The Retention Maturity Assessment maps your system against each of the five leaks. You will see which leaks are currently active, their estimated annualised economic cost, and which two interventions would recover the most margin fastest. Free, under 10 minutes.

maturity.groa360.ai