Join the waitlist

Sample case study

Tidewell Home tested two fixes for first-time buyers who do not come back

Tidewell HomeHome goods sold online and in 40 stores420,000 customers · $96M revenue · Insights team of 2

Sample case study. Tidewell Home is a fictional brand and the data is simulated to show how Cufo works. These are not customer results.

First-time buyers at Tidewell Home are coming back less than a year ago. Cufo linked part of the drop to late deliveries and suspects colours that look different from the photos. Of two finished tests, a cheap change to the order confirmation page worked and was adopted, while a daylight-photo test was not decisive, so the colour question stays open. Three decisions were made without a test, and the Black Friday shipping offer is a proposed test that has not been approved or run.

of first-time buyers from April to June ordered again within 90 days. A year earlier: 23.5%
17.8%
of first online orders arrived after the promised date (12,350 of 44,100)
28%
return rate for Harbor duvet covers from lot HL-07, against 5.9% for earlier lots
17.4%
of September checkout starts left at the shipping step (75,560 of 184,300)
41%

Situation

The situation

Tidewell Home sells bedding, throws, tableware, rugs and curtains, online and in 40 stores. It has 420,000 customers and an insights team of two. This sample follows the week of 19 October 2026.

First-time buyers are coming back less. Of those who bought for the first time from April to June, 17.8% ordered again within 90 days. A year earlier it was 23.5%. The head of insights has to decide where the Q1 retention budget goes, by 13 November. A Black Friday shipping offer also has to be set, because 41% of checkout starts leave at the shipping step.

The decisions here are the head of insights and growth’s.

1Data

See: What was collected

Cufo brought seven sources together. Personal details are masked before analysis, and each source shows its age and its known problems.

Sources and records

  • Online and store orders412,000
  • Carrier scans44,100
  • Support conversations18,400
  • Product reviews9,800
  • Returns portal21,300
  • Checkout funnel and site search184,300
  • August wishlist survey1,140
  • In all691,040

Gaps Cufo flagged

  • Support and reviews reach only 6% of first-time buyers. The other 94% have told Tidewell nothing, so the reason many do not return is unknown.
  • The checkout funnel data is 4 days old. The access key for its export expired on 15 October.
  • UK checkout sessions before cookie consent are not recorded, so UK checkout figures are partial.
  • The August wishlist survey is 52 days old. Only 118 of its 1,140 answers came from loyal customers, so splits by loyal customers are low confidence.
  • 17,140 store receipts are not yet linked to an online account.
  • 640 support chats still hold phone numbers or addresses and are waiting to be redacted.

2Analytics

Think: What it means

Cufo coded six themes from support, review and return text and set them beside carrier scans and orders. Four are shown here.

Themes

Each chart shows mentions a week over the 12 weeks before the case, oldest on the left.

  • Late delivery on first orders

    640 mentions

    Orders arrive after the promised date, tracking stalls, and buyers say nobody told them it would be late.

    6212 weeks agoLast week
  • Colour not as pictured

    1,590 mentions

    Fabric looks greyer, warmer, thinner or glossier than in the photos. It is 38% of 4,180 first-order returns.

    14112 weeks agoLast week
  • Shipping cost at checkout

    640 mentions

    Shipping cost appears at the last step. Some small baskets leave, and some buyers say they would pay if they knew the date.

    6812 weeks agoLast week
  • Harbor duvet cover shrinks after washing

    500 mentions

    The cover shrinks after one wash, even when the label is followed. Mentions jumped in a single week.

    7012 weeks agoLast week

In customers’ words

Illustrative lines from the simulated data, with personal details removed.

  • Lovely rug, wrong week. Delivery took 13 days and nobody told me it would be late.

    Review

  • Photos must be heavily edited. The 'oat' is basically grey. Disappointing for the price.

    Review

  • I'd pay the $14.95 if I knew the date. Right now it's just a range.

    Support

The finding

First-time buyers whose first order arrived late reorder less, but the data cannot yet show that lateness is the cause.

Evidence

  • Fact12,350 of 44,100 first online orders (28%) arrived after the promised date.
  • FactBuyers with a late first order reordered within 90 days at 12.4%, against 19.9% for on-time orders.
  • FactLate orders skew to rugs and furniture. After adjusting for category mix, late delivery explains an estimated 1 to 3 points of the 5.7-point drop.
  • FactSupport and reviews reach only 6% of first-time buyers. The other 94% have said nothing.

Confidence

Medium

Carrier scans cover every online order, but the link to reordering is a correlation.

Considered and set aside

  • Late delivery as the whole story: after adjusting for category mix it explains an estimated 1 to 3 of the 5.7 points, so most of the drop is still unexplained.
  • Photos as the main cause of colour returns: the daylight-photo test did not move returns decisively.
  • A general rise in Harbor duvet returns: returns break by production lot, 17.4% for lot HL-07 against 5.9% for earlier lots, and the new mill skipped pre-washing.

3Experiments

Try: What was tested

Two tests finished before this week, and each set its rule before it started. Two more are proposed for Black Friday. They would share one control: half of checkout sessions keep today's $99 free-shipping line and a quarter see each change.

The tests

  1. Test 1 of 4Complete

    Daylight photos on the 40 bestselling textiles

    Hypothesis Colour returns happen because studio light flatters the fabric, and daylight photos set the right expectation.

    Design
    A/B test. People were split at random into two groups, one with the change and one without.
    Measure
    First-order return rate, textiles. Lower is better.
    Dates
    8 Jun to 5 Jul 2026
    The rule
    Adopt only if the difference is clear at p below 0.05, in the right direction.
    Rule set in advance
    Yes, before the test started

    What the numbers say

    No clear difference

    7.40% for the control (663 of 8,960) against 7.00% for the variant (627 of 8,960), a difference of 0.40 points in the variant's favour; the test could not tell this apart from chance (p = 0.30).

    6.0%6.5%7.0%7.5%8.0%ControlVariant

    Control: studio photosVariant: daylight photos with a colour cardLines show 95% intervals

    Exact figures
    GroupTotalCountedRate95% interval
    Control: studio photos8,9606637.40%6.88% to 7.96%
    Variant: daylight photos with a colour card8,9606277.00%6.49% to 7.54%
    Difference, variant minus control−0.40 points, 95% interval −1.16 to 0.36 (Newcombe)
    Relative change5.4% lower
    p-value, pooled two-proportion z-test0.2981 (the rule needs below 0.05)

    Total: people or sessions in each group. Counted: how many of them did what the measure counts. All figures are computed on this page from those two counts.

    What the brand did

    Inconclusive Returns moved the right way but not decisively, and colour complaints did not fall either. The photos were not adopted. Photos are not the main cause of colour returns, so the colour question stayed open and went into the day-30 survey for first-time buyers.

  2. Test 2 of 4Complete

    Matching items on the order confirmation page

    Hypothesis Showing matching items right after purchase raises second orders within 30 days.

    Design
    A/B test. People were split at random into two groups, one with the change and one without.
    Measure
    Second order within 30 days. Higher is better.
    Dates
    4 to 31 Aug 2026
    The rule
    Adopt only if the difference is clear at p below 0.05, in the right direction.
    Rule set in advance
    Yes, before the test started

    What the numbers say

    Clear win for the variant

    11.8% for the control (2,148 of 18,200) against 13.2% for the variant (2,402 of 18,200), a difference of 1.4 points in the variant's favour; the test can tell this apart from chance (p < 0.001), so by the rule set for it the variant did better.

    11%12%13%14%ControlVariant

    Control: receipt onlyVariant: three matching itemsLines show 95% intervals

    Exact figures
    GroupTotalCountedRate95% interval
    Control: receipt only18,2002,14811.80%11.34% to 12.28%
    Variant: three matching items18,2002,40213.20%12.71% to 13.70%
    Difference, variant minus control+1.40 points, 95% interval 0.72 to 2.08 (Newcombe)
    Relative change11.8% higher
    p-value, pooled two-proportion z-test0.0001 (the rule needs below 0.05)

    Total: people or sessions in each group. Counted: how many of them did what the measure counts. All figures are computed on this page from those two counts.

    What the brand did

    Adopted Adopted as the default on 2 Sep. A small, cheap change worked, but it does not help buyers whose first order went wrong. Whether it lifts the 90-day repeat rate is checked on 2 Dec.

  3. Test 3 of 4Planned

    Free shipping from $75 instead of $99

    Hypothesis A $75 free-shipping line lifts checkout completion enough to pay for the subsidy.

    Design
    A/B test. People were split at random into two groups, one with the change and one without.
    Measure
    Checkout completion. Higher is better.
    Dates
    Proposed for about 22 Oct to 18 Nov 2026
    The rule
    Adopt only if the chance that the variant is better reaches 95%.
    Rule set in advance
    Not yet registered

    What the numbers say

    Not run yet

    What the brand did

    Planned Proposed, not approved and not run. It would stop if margin per order falls below $16.40, from $18.40. The estimate is 2% to 5% more completed checkouts, and a test this size can only see effects of about 4% or more.

  4. Test 4 of 4Planned

    Delivery-date promise at checkout

    Hypothesis A guaranteed delivery date lifts checkout completion more than a lower shipping line.

    Design
    A/B test. People were split at random into two groups, one with the change and one without.
    Measure
    Checkout completion. Higher is better.
    Dates
    Proposed for about 22 Oct to 18 Nov 2026
    The rule
    Adopt only if the chance that the variant is better reaches 95%.
    Rule set in advance
    Not yet registered

    What the numbers say

    Not run yet

    What the brand did

    Planned Proposed, not approved and not run. It shows an "Arrives by" date with a $10 credit if late, and would stop if late-delivery credits pass 8 per 100 orders. The estimate is 3% to 7% more completed checkouts.

4Strategy

Act: What was decided

Five decisions are on record. One came from a test, three were made without one, and one is still open.

Decisions on the record

Tested: decided from a test. Untested: decided without one. Pending: not decided yet.

  1. Run the spring bedding sale at 20% off, not 30%.

    Bedding revenue during the sale was 11% above last year, inside the 8% to 15% expected. Gross margin on bedding fell 3.6 points against 1 to 3 expected, because more shoppers stacked the sale with a first-order code. Codes no longer apply to sale items.

    Untested
  2. Make the matching items block on the order confirmation page the default.

    Next check on 2 Dec 2026: the 90-day repeat rate of August buyers.

    Tested
  3. Ship tableware in moulded-pulp packaging from 1 Nov.

    Damaged in transit is 14% of first-buyer tickets. The change is reversible and costs $0.40 an order, so there was no test. Expected: 25% to 45% fewer damage tickets on tableware, checked on 15 Dec.

    Untested
  4. Hold the 1,800-unit Harbor duvet reorder until the mill confirms pre-washing.

    Returns data already answered the question, so there was nothing to test. Expected return rate on the next lot: 5% to 8%, checked on 15 Feb 2027.

    Untested
  5. Set the Black Friday shipping offer. Proposed: test the $75 line and the delivery-date promise against today's $99 line.

    Not decided. The test costs an estimated $7,500 to $12,000, above the $5,000 the head of insights can approve alone, so the CEO must sign off.

    Pending

What changed next

  • The confirmation page result moved retention work to the first order itself: delivery dates and colour accuracy.
  • The day-30 survey for first-time buyers is drafted for 1,500 September buyers. It has not been sent. Its text reminder would reach 1,188 people without written consent, so it goes only to the 312 who gave it.
  • A second ask, 200 chat interviews with lapsed first-time buyers, is also drafted. 23 invitees were asked for feedback in the last 45 days and must be replaced first.
  • Harbor is still open: the fix has not shipped, buyers have not been told, and returns have not recovered. 3 of 5 tracker steps are not done.
  • Late delivery is still open: 28% of first online orders are late against a target of 12%, and the fix has not shipped.
  • Buying commitments for the 2027 range are due 15 Dec. Whether loyal customers want bedding or outdoor next is still unanswered.

Takeaways

What this case shows

  1. A test that does not decide is still useful: the photo test turned a guess about colour into a survey question.

  2. Cheap, reversible decisions can skip a test if they are written down with a number to check back against.

  3. Keep what was found, tested, decided and still open apart. The Black Friday offer is proposed, not decided.

More

More sample case studies

  • Sample case study 01

    Fernhill Botanics

    Clean skincare for sensitive skin, sold online

    Refills of a best-selling serum fell. A reminder test was inconclusive; reviews and tickets pointed to a new pump that leaks or jams.

    116pump complaints in the six weeks after the pump changed

  • Sample case study 03

    Brightloop

    Language-learning app with AI speaking practice, sold to individuals and schools

    A language app checks a price rise it never tested, finds shared logins wiping out progress, and plans a family plan test before launch.

    574support tickets in 12 weeks about progress overwritten on a shared login

  • Sample case study 04

    Larkfield Mobile

    A challenger mobile network in 14 European markets

    A fictional mobile network sent an app update to everyone at once. See how Cufo found the cause, stopped a risky test and planned a staged fix.

    2.1% to 9.8%port-in eSIM failures, before 9 Oct and at the peak on 12 Oct

All sample case studies

The demo shows the Fernhill Botanics case inside Cufo, with the same simulated data: open the demo.

Waitlist

Hear when Cufo opens

Cufo is in private development. Early access opens to a small group of brands first.

Join the waitlist