← Back to DunwellSample project built from public data. Not a Dunwell client.
Insights
Dunwell Insights · Growth Check · Sample

Sales grew 1.9%. Repeat business weakened.

New buyers kept revenue moving, while returning customers spent less. This Growth Check shows where to look, which customers to contact first, and which actions to test.

Independent sample using historical public data. Not a client engagement. A standard Growth Check includes the report, a metrics sheet and up to five ranked actions. The live dashboard and the churn model shown here are extras.

Business: UK online gift and homeware shop, mostly selling to other businesses Data: 1.07 million order lines, December 2009 to December 2011 Prepared by Dunwell Insights, October 2026 This sample also shows: the live dashboard (a paid extra), the full at-risk list, cleaned data and scripts
Net revenue each month, December 2009 to November 2011. Year 2 is shaded.
Summary for the owner

New buyers covered for weaker repeat business

What we found

    What we'd test first

    Open the live dashboardFilter every figure by year, region and customer type, work through the full at-risk list, and download the action list.
    1 · What happened

    Existing customers took £1.57m away, and new buyers just covered it

    Net revenue (sales minus cancellations) rose from £9.16m to £9.33m. That small rise is the sum of four large movements. Customers who stopped buying took £1.02m with them. Customers who stayed spent £548,000 less. New customers and guest checkouts brought in £1.74m. Take out the new customers' spending and year 2 comes in 13% below year 1.

    What moved revenue from year 1 to year 2
    Change in net revenue, £. Each customer counted once.
    Lost: bought in year 1 only. Returning: bought in both years. New: year 2 only. Guests have no account, so they get their own line, as do 23 accounts that only ever cancelled (+£1,404). With that line the movements add up exactly.
    Every customer type spent less in year 2
    Net revenue by customer type, £
    Type comes from each customer's typical order: Small under £150, Mid £150 to £400, Trade over £400. None grew.

    Both years had the same shape: quiet to Easter, flat through summer, then a climb to a November peak. September to November carried 37.5% of year 2 revenue. That peak matters for the next finding.

    Peak season carries more than a third of the year
    Returning customersNew customers (first month)Guest checkout
    Net revenue by month. Peak season is shaded. December 2009 is the first month of data, so everyone counts as new that month. That month, new customers' cancellations outweighed their sales by £3,578. The small negative amount isn't drawn, but it is in the month's total.
    2 · Finding

    Customers won in the autumn rush come back far less often

    Return rate by the month a customer first bought
    Below averageAbove averageAverage
    Share placing a second order within 180 days. First orders from March 2010 to early June 2011.
    The gap holds inside every customer type
    First bought Dec to AugFirst bought Sep to Nov
    So it isn't only that autumn brings smaller buyers. Trade buyers won in autumn return 20 points less often.

    Why it matters: the season that brings the most new customers brings the least loyal ones. Our reading is that many peak-season buyers come for a one-off gift order, and nothing in the shop's process treats them differently. That's a lead to test, not a proven cause.

    3 · Finding

    Customers who place a second order are far more likely to stay

    Bigger first orders go with higher return rates
    Share returning within 180 days, by value of the first order
    Some first products go with higher return rates
    Return rate for customers whose first order included the product, with its likely range (95%)
    Products bought first by at least 60 customers. Woodland and children's lines lead. Gift boxes and one-off decorative pieces trail.
    4 · Finding

    A simple rule ranks customers as well as machine learning
    Gradient boosting, 12 signalsSimple rule: days since last order, order count
    Customers sorted into ten equal groups by predicted risk, tested on a later period (customers as they stood on 1 June 2011). The line shows how many really stopped buying.
    The lost regular
    Customer 13093 ordered 55 times, usually every 9 days. Then nothing for 275 days.
    The simple rule alone scores this customer at 3%, because 55 past orders pull the score down. The overdue check catches it. A monthly sales total would never show it.

    Three kinds of at-risk customer, three different follow-ups
    New buyers have ordered once. Lapsed regulars used to order at least every 90 days. Occasional buyers order less often than that, so a long gap is normal for them and the follow-up should be light.
    The ten biggest at-risk customers
    5 · Finding

    Saturday sales are very low in the records

    Across two years, the records show £9,803 of Saturday sales. Sundays brought in £1.79m. Orders peak at midday and almost stop after 6pm. That fits a shop selling mostly to other businesses. The order records can't say whether customers try to order on Saturdays and give up. Website visits and Saturday enquiries would answer that.

    Sales by day of the week, both years, before cancellations
    6 · Actions

    What to test, what it could be worth, and how we'll know

    Each value is yearly revenue, estimated as one stated assumption times the shop's own numbers. Spend uses medians, so a few huge trade accounts don't inflate the result. The actions reach some of the same customers, so these are separate scenarios. They are not added together.

    How we'll measure it: for each action, hold back one in five of the customers it targets and leave them alone. The action list marks them. At 90 days, compare the two groups for an early read. The main result comes at 180 days for return rate and at 12 months for spend, the same periods the estimates use. The measured difference replaces our estimate.

    7 · How we did this

    Every step and every decision, so you can check our work

    The analysis runs from the raw file to this report with one command. The scripts, the cleaned data and the full at-risk list come with the report, so every number here can be traced and rerun.

    Cleaning the data
    Checks we ran

      Decisions log

      What the data can't tell us

      • There are sales but no costs, so this report is about revenue, not profit.
      • There is no marketing, website or stock data. Where we suggest a cause, it's a lead to check, not a proven fact.
      • About 15% of year 2 revenue came from guest checkouts, which can't be followed over time.
      • Risk scores rank customers well. Treat the exact percentage as a guide, not a promise.
      • Last year's spend shows what's at stake. It isn't money a campaign is sure to win back.
      • Findings about first orders, products and seasons are associations. They point to what to test, not proof of cause.
      • This is one shop's data. The methods carry over to other businesses. The specific numbers don't.

      This analysis is based on the data supplied and the methods described here. It is there to support decisions. It is not a promise of any result, and it is not financial, legal or tax advice.

      Every job. Done well.