Customer Retention Review

1. Executive summary

Net revenue$512.3k21 months, refunds excluded

Growth YoY+28.5%Jan–Sep, $242.1k

Customers3,4121.37 orders each

Repeat rate24.8%ordered 2+ times

Average order$109.50paid orders

Growth is real, and more of it now comes from people who have bought before: returning customers were 21.3% of revenue in 2025 and 30.5% so far in 2026. But that is mostly the customer base getting bigger. The rate at which each new cohort comes back has not moved, and it depends heavily on where customers first came from.

Channel quality varies threefold.

31.9% of customers first acquired through Email order again within 90 days, against 10.9% for Marketplace. They are also worth $191.50 each so far, against $132.15.

Retention is flat across cohorts.

Every quarterly cohort with a full window repeats at 24.1% to 26.5% within 180 days. Nothing in the last 18 months has changed how likely a new customer is to come back.

A quarter of revenue sits with lapsing regulars.

482 customers have bought two or more times but not in the last 180 days. They account for 24.7% of all revenue ($126,672).

Most customers buy once.

75.2% of customers have a single order. 1,733 of them last bought more than 180 days ago.

Recommended next steps

Build a post-purchase email series at days 14, 30 and 60 for every first-time buyer, and measure the change in 90-day repeat rate by channel.
Run a win-back offer for the 482 at-risk regulars before the November peak.
Review acquisition spend on Marketplace and Social against their lower repeat rates. Judge channels on 90-day revenue per customer, not first-order revenue.

Larkspur & Co. · Customer retention reviewWork sample, placeholder data2

2. Trend and channel quality

Monthly net revenue from first orders (navy) and repeat orders (teal). The peak month was 2026-08 at $30,938.

Customers by first channel

Channel	Customers	90-day repeat	Revenue / cust.
Email	404	31.9%	$191.50
Referral	369	25.2%	$174.03
Search	1,025	19.7%	$157.21
Social	785	12.7%	$134.00
Marketplace	330	10.9%	$132.15

Customers acquired by 1 July 2026, so each has a full 90-day window. Revenue per customer is all net revenue to date.

Reading this: Email and Referral customers arrive already warm and come back. Marketplace and Social bring volume at first purchase, but fewer of those buyers return, so their true cost per retained customer is higher than first-order figures suggest.

Larkspur & Co. · Customer retention reviewWork sample, placeholder data3

3. Cohorts and customer segments

Repeat rate by first-order quarter

Cohort	Customers	90 days	180 days
2025Q1	453	18.1%	24.1%
2025Q2	449	19.1%	26.5%
2025Q3	447	19.5%	25.9%
2025Q4	536	19.2%	24.2%
2026Q1	502	19.9%	25.7%
2026Q2	520	19.6%	n/a
2026Q3	505	n/a	n/a

n/a: the cohort has not yet had the full window.

Segments as at 30 September 2026

Segment	Customers	Revenue	Share
Champions	97	$36,571	7.1%
Loyal	268	$64,185	12.5%
At risk	482	$126,672	24.7%
New	314	$35,124	6.9%
Single order, recent	518	$57,960	11.3%
Lapsed one-off	1,733	$191,826	37.4%

Segment rules: Champions 3+ orders and last order within 120 days; Loyal 2+ orders within 180 days; At risk 2+ orders, none in 180 days; New 1 order within 60 days; Single order, recent 1 order 61–180 days ago; Lapsed one-off 1 order over 180 days ago.

Larkspur & Co. · Customer retention reviewWork sample, placeholder data4

4. Data quality and method

Cleaning log

Step	Rows
Rows in the raw export	4,996
Channel values standardised (case, spaces)	144
Test orders removed	12
Duplicate order rows removed	85
Blank regions set to Unknown	60
Refunded orders excluded from revenue	220
Paid orders analysed	4,679

Definitions

Revenue: order value of paid orders. Refunded orders are excluded from revenue and order counts.

First channel: the channel of a customer's first paid order.

Repeat within N days: another paid order 1 to N days after the first. Only customers with a full window are counted.

Growth: Jan–Sep 2026 against Jan–Sep 2025.

How the analysis is built

A single Jupyter notebook (pandas, NumPy, Matplotlib) reads the raw export, logs each cleaning step, calculates every figure and chart in this report, and writes them to a metrics file and clean CSVs. The report is generated from those outputs, so re-running the notebook on a new export refreshes it with no hand-typed numbers.

A separate check script recalculates the headline figures from the raw export without pandas and confirms they match the notebook and this PDF.

Files

File	What it is
analysis.ipynb	The executed notebook, with outputs
data/orders_raw.csv	Raw order export
output/clean_orders.csv, customers.csv	Clean paid orders and one row per customer with segment
output/metrics.json	Every figure used in this report

Larkspur & Co. · Customer retention reviewWork sample, placeholder data5