Nine stages. Every dollar decided again.
Growth compounds — or leaks — at nine specific moments in a customer's life. This page maps all nine: the metric that matters, the honest benchmark, the failure mode, and exactly where Retrics acts.
Acquisition is a price. Know what it buys.
Every growth plan starts with a price: what a customer costs. Paid CAC — ad spend over ad-attributed customers — is the number agencies quote, and it flatters. Blended CAC divides every dollar of sales and marketing by every new customer, including the free ones from organic and referral that paid channels claim anyway. When paid CAC reads $42 and blended reads $68, blended is the truth. Run the business on it.
Two rules make acquisition solvent. LTV:CAC of at least 3 — below that there is not enough downstream margin to fund product, ops, and the next customer. And payback under 12 months: the ratio can look fine at 3.0 and still starve you if the cash returns over four years while the ad bill clears in thirty days. CAC is not a KPI to minimize. It is a price to justify — and the justification lives in the eight stages after this one.
LTV : CAC — and the payback clock
LTV : CAC = (AOV × orders/yr × margin × yrs) / blended CACpayback (mo) = CAC / monthly contribution per customerCAC-only dashboards. The brand optimizes the price of a customer down to the decimal while the value of a customer quietly falls — and calls the quarter a win.
The first 30 days decide the next three years.
The clock starts at checkout, but value starts at first use. Time-to-first-value in D2C is delivery time plus the days the box sits unopened — and every one of those days lowers the odds of a second order. Measure the gap between order and the first value moment: delivery confirmed, product opened, first result felt. Brands obsess over conversion-rate decimals while a three-day fulfillment delay quietly taxes every cohort behind it.
The first 30 days decide roughly 31% of a customer's eventual repeat value — usually before the brand has sent anything but a shipping notification. Onboarding is not a welcome-series template; it is timed to the use-up curve. A 30-serving product creates a reorder decision near day 25. Teach usage in week one, confirm the result in week two, and make the reorder effortless before the jar is empty — not after.
Activation rate, 30-day
activation = customers with 2nd value moment ≤ 30d / cohortTTFV = first-value date − order dateThe brand celebrates the order and goes silent until the next promo. By the time anyone writes again, the product is in a drawer and the moment is gone.
One window is worth 3.2× the rest.
Between day 30 and day 60 sits the most valuable window in the lifecycle. A customer who places order two inside it goes on to generate 3.2× the repeat value of one who orders later or not at all — measured across live stores, as-of scored. The mechanism is plain: a fast second order means the product worked, the habit sketch exists, and the brand earned a slot in the routine. Miss the window and you are re-acquiring your own customer at full CAC.
The window is not the same for every category — it is the use-up curve wearing a calendar. Consumables reorder in weeks; durables cross-sell in months. And it is not the same for every customer: a daily user exhausts a 30-day supply twice as fast as the label promises. The nudge that works is timed to the individual's projected reorder date, arrives before the decision, and asks for the same product — not a discovery email for a different one.
Second-order rate, 60-day
2nd-order rate = customers with order 2 ≤ 60d / cohort sizewindow lift = repeat value (order 2 in d30–60) ÷ (later)The window opens and closes in silence. Day 61 arrives, and the brand's first touch since the shipping confirmation is a 20%-off blast written for everyone.
Retrics scores every new customer's window as it opens and drafts the reorder nudge to land on their projected run-out date — the campaign calendar never sees it.
Frequency compounds. Basket size just adds.
Orders three through five are where a repeat buyer becomes a habit. The tell is the inter-purchase interval: the median days between a customer's orders, tracked per customer, not per segment. A tightening interval is a habit forming. A widening one is drift arriving early. Most dashboards can see neither, because they report monthly averages across everyone — and averages are where cadence goes to die.
Frequency compounds LTV harder than basket size. Moving a customer from two orders a year to three is +50% lifetime value at identical AOV; no upsell widget moves AOV 50%, and aggressive cart-stuffing often suppresses the next order entirely. Subscription is frequency made contractual — powerful, but so is manual repeat at a stable cadence, which carries full margin and no pause-button churn. Sell the sooner order, not just the bigger one.
Inter-purchase interval (IPI)
IPI = median days between order n and order n+1LTV = AOV × orders/yr × margin × active yrsMerchandising pushes a bigger cart instead of a sooner one. AOV climbs 8%, frequency slips, LTV stays flat — and the quarter is reported as a win.
Your top decile is the business.
Revenue concentration is the oldest pattern in retail and the least acted on. The top decile of customers by LTV routinely carries a third to half of revenue — the 80/20 curve bends hard at the top — and a VIP is often worth 10× the median customer. That concentration is not a risk to diversify away; it is the business, and it deserves the same named-account attention a B2B company gives its enterprise tier. Know these customers individually. There are only a few hundred of them.
The instinct is to reward VIPs with discounts. Resist it. A sitewide 20% hands the largest margin sacrifice to the people who were already paying full price, and trains price sensitivity into exactly the customers who had none. Protect, don't discount: early access to drops, restock priority, a human reply, the occasional unearned gift. Loyalty compounds on recognition. It erodes on coupons.
Top-decile revenue share
decile share = revenue of top 10% (by LTV) / total revenueThe sitewide sale. Your most loyal customers get the same 20% off as strangers — margin spent where the loyalty was already earned, sensitivity taught where there was none.
Churn never announces itself. It goes quiet.
D2C churn is non-contractual: there is no cancel button, no end date, no event to alert on. A customer simply does not order again, and the silence is indistinguishable from a normal gap — until it is not. Average churn rate is a fiction here. A 90-day gap is a dead coffee subscriber and a perfectly healthy skincare customer. The only honest unit of analysis is one customer measured against their own history.
That is what per-customer lapse probability is: given this customer's own cadence — their intervals, their trend, their seasonality — how likely is it that the current gap ends in another order? When the probability crosses threshold, the flag fires with its confidence attached: 0.94, not a vibe. Weeks before the quarterly cohort review would notice, while the customer is still reachable and the save is still cheap.
Per-customer lapse probability
P(lapse) = P(no next order | gap t, own cadence history)Churn is discovered in the quarterly cohort review — sixty days after the customer already made the decision, and long after the save was cheap.
Retrics scores lapse probability against each customer's own cadence, every day, and flags the drift at 0.94 confidence — weeks before silence reads as churn.
Some of them were coming back anyway.
Here is the uncomfortable truth about at-risk flows: some of those customers were returning regardless. Hold out a random slice of flagged customers, send them nothing, and about 2.7% come back on their own. That is the natural-return floor, and any campaign measured without it is claiming credit for gravity. Most retention dashboards do exactly that — and report the floor as performance.
Lift over the floor is the only honest number. If 9.0% of treated customers return against the 2.7% holdout, the flow earned 6.3 points — that is the revenue it created, and the only revenue worth paying for. Honest measurement is not academic hygiene; it changes decisions. Flows with beautiful open rates and zero lift get killed. Boring flows with real lift get budget. The dashboard stops flattering and starts steering.
Incremental lift over holdout
lift = return rate (treated) − return rate (holdout)honest revenue = lift × customers flagged × AOV × marginNo holdout. The flow reports every returning customer as a save — including the 2.7% who were coming back anyway — and the budget follows the flattery.
Every Retrics intervention ships with its own holdout. The ledger reports lift over the 2.7% floor — never raw returns.
Timing beats discount depth. Every time.
The reflexive win-back is a discount, and the reflexive fix for a weak win-back is a deeper one. The data disagrees. Response tracks recency far more than incentive depth: a specific nudge at roughly 1.2× the customer's own cadence — your usual is about to run out — outperforms a 30%-off apology sent at day 120. By the time the deep discount lands, you are not winning a customer back. You are buying a stranger at a loss.
Every save has a price, and some saves are worth less than zero. The expected value of a win-back is the probability of return times future margin, minus the discount, the send, and the margin surrendered on orders that would have happened anyway. Run that arithmetic and a chunk of the lapsed file is negative-margin: low-LTV customers whose most profitable next step is being let go. Spend the recovered margin on the top decile instead.
Expected value of the save
save EV = P(return) × future margin − (discount + send cost)The escalating-discount ladder: 10%, then 20%, then 30% — training lapsed customers to wait, and paying the most for the least likely returns.
Retrics drafts the win-back timed to the customer's own cadence, holds the discount as a last resort, and never sends a save the math says is negative.
If it isn't on the ledger, it didn't happen.
Retention has an attribution problem worse than acquisition's. A returning order gets claimed by the email tool, the SMS tool, the loyalty program, and the ad platform — simultaneously — until attributed revenue exceeds actual revenue and nobody can say what worked. The fix is not another model. It is a ledger: every recovered dollar entered once, attributed to the one flow that earned it, net of the natural-return floor, timestamped and auditable.
A ledger changes behavior the way honest accounting always does. Flows compete on incremental dollars instead of open rates. The unglamorous reorder nudge that quietly earns $18.9k a quarter beats the beautiful campaign with zero lift. Retention reporting starts to read like finance instead of marketing — and when the CFO asks what the program returned, the answer is a number with receipts, decided again every single month.
Recovered revenue, net
recovered = Σ (returned order margin − floor share), by flowFive tools each claim the same returning order. Total attributed revenue exceeds actual revenue, and the budget meeting runs on fiction.
Every dollar Retrics recovers lands on the ledger with the flow that earned it — counted once, net of the floor, ready for your CFO.
Retention isn't a stage. It's the loop.