Time Between Orders: Timing Replenishment and Win-Back Emails From Your Own Data

Most replenishment and win-back emails fire on a round number of days. The usual way to pick a better number, averaging the gap between orders, leaves out every customer who has not reordered yet. Here is how to build a survival curve from your order data and read send times off it.

Replenishment and win-back emails depend on one number: how long customers usually wait before ordering again. Get the time between orders wrong and the reminder lands too early, when the customer still has plenty, or too late, after they bought elsewhere. Hello Retail’s guide to replenishment puts the problem simply: a single interval cannot fit a population with different consumption rates.

Most stores pick that interval from a benchmark or from the average gap between orders in their own data. Both are a start. The second has a flaw that is easy to miss, and fixing it changes the answer. This guide shows how to measure time between orders properly with a survival curve, using order data you already have in WooCommerce or Magento.

Benchmarks give you a starting range

Eightx published a 2026 benchmark of the median days between orders by vertical. It uses the median because order gaps are skewed, and it states that no single public report publishes this table, so the figures are assembled from several sources and portfolio observations.

VerticalMedian days between ordersTypical range
Supplements and vitamins3028 to 35
Pet food and supplies3025 to 45
Food and beverage4530 to 60
Beauty and skincare7560 to 120
Apparel and fashion12090 to 180
Home goods and furniture270180 to 365
Consumer electronics365270 to 730
Source: Eightx, Average Days Between Orders by Vertical: 2026 Benchmark.

The ranges are wide, and Eightx itself recommends calibrating against your own cohort data. Two stores selling skincare can differ by months depending on pack sizes and product mix.

Why the average gap between orders is too short

The obvious calculation takes every customer with two or more orders and averages the days between them. It ignores everyone who has bought once and not yet bought again. Some of those customers will reorder next week; others never will. Leaving them out keeps only the people who came back, and among recent customers the only ones who have come back are the quick ones. The result is an interval that is too short.

Observed and censored gaps between orders for five customers Five customers on a timeline ending today. Two placed a second order, so their gap is observed. Three have not reordered yet: one bought recently, one bought months ago, and one long ago. Their gaps are at least as long as the time since their first order, but the true value is unknown. A naive average uses only the two observed gaps. today Customer 1 Customer 2 Customer 3 Customer 4 Customer 5 reordered: gap observed reordered quickly not yet, recent buyer not yet, after 5 months probably never
Dashed lines are censored: we know the gap is at least this long, not how long it will be. A naive average drops all three.

Statisticians call those dashed lines censored observations, and the problem is old. Drye, Wetherill and Pinnock wrote in 2001 that, for this kind of time-to-event data, simple means and summary statistics can be seriously biased and misleading. In one of their examples, a simple mean put average membership at just over four years, while the estimate that accounted for censoring was nearly seven. Our post on lost sales from stockouts deals with the same kind of hidden data on the demand side.

Measuring time between orders with a survival curve

A survival curve, usually estimated with the Kaplan-Meier method, answers a better question: of the customers who bought on day 0, what share has not yet reordered by day t? Customers who have not reordered still count, up to the day you observe them, and then drop out of the calculation without being treated as either reorders or losses.

Build the durations

Each row is one order, the number of days to that customer’s next order, and whether the next order has happened. In WooCommerce, wc_order_stats carries a customer_id that points to wc_customer_lookup, the table WooCommerce Analytics uses to identify customers, including guests by email. This query needs MySQL 8 or MariaDB 10.2 or later.

WITH o AS (
  SELECT s.customer_id, DATE(s.date_created) AS order_date
  FROM wp_wc_order_stats s
  WHERE s.status IN ('wc-completed', 'wc-processing')
    AND s.parent_id = 0
),
seq AS (
  SELECT customer_id, order_date,
         ROW_NUMBER() OVER (PARTITION BY customer_id ORDER BY order_date) AS n,
         LEAD(order_date) OVER (PARTITION BY customer_id ORDER BY order_date) AS next_date
  FROM o
)
SELECT customer_id, n,
       DATEDIFF(COALESCE(next_date, CURDATE()), order_date) AS days,
       next_date IS NOT NULL AS reordered
FROM seq;

For product-level replenishment, add a filter so the next order must contain the same product or category. On Magento, the same query runs on sales_order, grouped by customer_email so guest orders are included.

Fit the curve

The Python library lifelines has a ready-made KaplanMeierFitter that takes exactly these two columns.

import pandas as pd
from lifelines import KaplanMeierFitter

df = pd.read_csv("order_gaps.csv")   # customer_id, n, days, reordered
first = df[df.n == 1]                # time from first to second order

kmf = KaplanMeierFitter()
kmf.fit(first["days"], event_observed=first["reordered"])

print(kmf.survival_function_at_times([14, 30, 45, 60, 90, 180]))

Fit separate curves for first-to-second and later orders. A first-time buyer and a fifth-time buyer behave differently, and the second purchase is usually the one a flow is trying to win.

Reading send times off the curve

Illustrative survival curve of customers who have not yet placed a second order The share of customers who have not reordered starts at 100 percent, falls steeply between day 25 and day 60, and flattens at about 62 percent by day 180. The replenishment reminder is placed just before the steep part, around day 28. The win-back email is placed where the curve flattens, around day 95. The flat level shows the 62 percent who are unlikely to reorder. 100% 60% 0% 0 30 60 90 120 180 Days since first order Replenishment just before the drop Win-back where it flattens about 62% never reorder
Illustrative curve for one product category. The flat tail is as useful as the drop: it tells you how many customers a flow can realistically reach.

When fewer than half of customers ever reorder, the median time to reorder does not exist, because the curve never falls below 50 percent. A more useful approach is to work with the share who eventually reorder. In the example below, 38 percent of first-time buyers reorder within six months. The timing points are read as fractions of those eventual reorders.

MeasureNaive averageSurvival curve
Customers used1,100 who reorderedAll 4,000 first-time buyers
Share who reorder within 180 daysNot available38%
Day by which a quarter of reorders have happenedNot availableDay 30
Typical gap34 days (mean of repeaters)Day 41 (half of eventual reorders)
Day by which 90% of reorders have happenedNot availableDay 95
Illustrative figures for one category over 12 months of first orders.

Those numbers translate directly into a flow. Send the replenishment reminder a few days before reorders start to pick up, here around day 26 to 28. Send the win-back after most reorders that are going to happen have happened, here around day 95; a customer who has not ordered by then is behaving like the 62 percent who never do. Hello Retail’s guide also recommends suppressing the reminder if the customer bought in the last seven days, and adjusting for pack size, since a larger pack lasts longer.

Segments worth a separate curve

SegmentWhy it differsWhat changes in the flow
Product or categoryConsumption cycles differSeparate reminder day per product
Pack sizeA larger pack lasts longerScale the day by quantity bought
Order number (first, second, later)Loyal customers reorder faster and more reliablyShorter interval and less incentive for repeat buyers
First order on a discountDiscount-led buyers often reorder lessExpect a lower plateau; judge the flow on its own baseline
Acquisition channelGift buyers and marketplace shoppers rarely replenishExclude or send a different message

To compare segments over time, line them up by the month of first order. QuantumLayers covers how to read that kind of table in Cohort Analysis; a survival curve is the same idea measured in days instead of calendar months.

Checking that the new timing works

  1. Keep a random 10 percent of eligible customers out of the flow as a holdout.
  2. After two full cycles, compare the survival curves of the emailed group and the holdout.
  3. Look at the gap between the curves, not the email’s click or conversion rate; some of those orders would have happened anyway.
  4. If the curves only separate after the reminder date, the timing is right. If they separate before it, customers were already reordering and the reminder came late.
  5. Refit the curves each quarter; pack sizes, prices and the product mix change the answer.

Sources