Completions
Quiz →

Operating model · Free · No call required

The DTC Retention Operating Model

Retention is usually treated as a campaign problem and funded as one. It is decided earlier than that, in four moments most operations do not instrument, and measured with a number that is frequently wrong in a direction that flatters.

Published August 22, 2026· 9-minute read · Written so you can run it without hiring anyone

Where retention is actually decided

Four moments. None of them is the campaign, and three of them happen before anyone in marketing is looking.

The second order, not the first

Whether a customer returns is largely settled by what happened between order one and order two — delivery accuracy, the first support contact, whether the product matched the page. Most retention spend goes to month six.

A win-back campaign is repairing a decision the operation made months earlier. It is cheaper to change the thing that caused the decision.

The moment before the cancel click

Save flows fire at the cancellation screen, which is after the decision. The signals that predicted it — a skipped shipment, a support ticket, a drop in engagement, a failed payment retried twice — were available weeks earlier.

A save offer presented to someone who has already decided reads as a toll. The same offer presented at the skipped shipment reads as service.

The failed payment

Involuntary churn is routinely a double-digit share of total churn and is almost entirely a systems problem: card expiry, a retry schedule that gives up too early, a dunning email that lands in promotions.

It is the cheapest retention available and the least discussed, because it belongs to no team. Marketing calls it billing; billing calls it marketing.

The replenishment window

Every consumable has a window in which a reorder is natural and outside which it requires persuasion. That window differs per product and per cohort, and is usually approximated with one global cadence.

A single cadence is wrong for most of the catalog by construction. The window is derivable from your own order history rather than from a benchmark.

The LTV number you are probably reporting

Three failures account for most retention numbers that cannot be acted on. Each is arithmetic rather than opinion, so each can be checked this week.

LTV computed on an immature cohort
A cohort three months old cannot report lifetime value. Averaging it with a cohort three years old produces a number that moves when the mix changes and not when the business does — which is why it can improve in a quarter where retention fell.
Revenue retention hiding customer churn
Rising average order value can hold revenue flat while the customer count falls. Both numbers are true; only one of them tells you whether people are staying. Report them separately or the good one will be quoted.
Attribution that ends at the first order
Channels are usually judged on acquisition cost against first-order value. A channel that acquires customers who reorder twice is worth multiples of one that does not, and a first-order model cannot see the difference — so budget flows to the wrong channel with full confidence.

Cancellation reasons that become actions

Most operations collect a cancellation reason and route all of them to the same place, which is a chart. The reason only earns its collection cost if each value has a different owner and a different response.

  • Too expensive — often means the value was not visible, not that the price was wrong. Test the articulation before the discount.
  • Too much product — a cadence problem wearing a churn costume. The save is a schedule change, not an offer.
  • Quality or fit — the only reason on this list that should reach the product team the same week, and the one most often aggregated into a dashboard nobody reads.
  • Switched to a competitor — worth knowing which one and on what dimension. This is the only reason where a discount is sometimes the correct response.
  • Life circumstances — genuinely uncontrollable, and important to separate out, because leaving it in makes every other bucket look smaller than it is.

A free-text box collects better reasons than a dropdown and is harder to act on. Run both: the dropdown for routing, the box for the sentence that tells you what the dropdown missed.

The record every channel reads from

Every retention tactic above assumes the underlying data agrees with itself. Three records decide whether it does.

One product record, read by every surface
The product detail page, the ad feed, the marketplace listing, and the support macro should read from the same record. When they do not, the disagreement surfaces to the customer as a broken promise and to you as a return.
One customer record, spanning anonymous and known
The session before the account, the order, the support ticket, and the subscription state need to resolve to one person. Retention decisions made on a partial record systematically misjudge the customers who use more than one channel — who are usually the valuable ones.
One definition of active
Finance, marketing, and the subscription platform frequently disagree about who counts as active. Every retention number downstream inherits that disagreement, which is why two teams can both be right and produce different answers.

Five questions to ask before you buy anything

  1. 1Which cohort is our LTV computed on, and has it matured?
  2. 2What share of our churn is failed payments rather than decisions?
  3. 3What signal fires before a cancellation, and how many days of warning does it give?
  4. 4When a cancellation reason is "quality", who receives it and in what timeframe?
  5. 5If the PDP, the ad feed, and the marketplace listing disagree, which one is authoritative?

If the second question has no answer, start there. Involuntary churn is the only retention work that requires no persuasion, no creative, and no discount.

If you would rather not run it yourself

The assessment names which of the four moments is costing you most, and what it would take to instrument it. It is diagnostic and it ends in a document you own whether or not anything follows.

The AI Readiness AssessmentWhat an engagement contains

Also: the deployment blueprint · the embedded executive engagement

Ready to talk now? Book the 30-minute consultation. Scope and pricing are settled on the call.

Free — no call required

The Multi-Location AI Swarm Blueprint

Get the deployment order for a marketing swarm across multiple locations — which agent goes first, what it depends on, and what has to be true before agent number two pays for itself.

  • The dependency map: which four foundation agents every other agent reads from, and why deploying a surface agent first strands it.
  • The 90-day sequence, week by week, with the acceptance test that closes each week.
  • The four loops that have to close — capture, decide, act, emit — and the failure mode when any one of them stays open.

Opens on this page immediately. No attachment, no waiting on an email.