Dunwell Studio home

Sample work

Five questions. Five complete studies.

Each study answers one business question with a written report, a live dashboard and a PDF. Together they show how Dunwell works: a clear question, checked data, a tested method and a practical next step.

Four studies use public data from real businesses. One uses synthetic data to demonstrate a method. None are Dunwell clients.

  1. 01Growth Check
  2. 02Campaign Impact Check
  3. 03Stock & Demand Planner
  4. 04Conversion Check
  5. 05Returns & Margin Lab
What moved revenue, year 1 to year 2Year 1 revenue£9.16mLost customers−£1.02mReturning spent less−£0.55mNew customers+£1.40mGuest checkouts+£0.34mYear 2 revenue£9.33mBars start at £7m

01 · Growth Check · Public data

Growth Check

The question
Sales grew 1.9%. Was the business actually getting healthier?
How we answered it
Split two years of orders into lost, returning and new customers, followed every newcomer for 180 days, and tested a churn rule against a 12-signal model.
What we found
New buyers covered £1.57m lost from existing customers. Customers won in autumn came back less often, and 1,193 at-risk customers are listed by value.

UK online gift shop · UCI Online Retail II · 1.07 million order lines. Independent study using public data. Not a Dunwell client.

Read the report Open the dashboard PDF
Buyers per 1,000 customers, two weeks after the email5.7No email8.8Women's email12.5Men's email

02 · Campaign Impact Check · Public data

Campaign Impact Check

The question
Did the email campaign sell more, and should it go to fewer, better-chosen customers?
How we answered it
Used a random hold-out of 21,306 customers to measure what each email caused, then tested six targeting models against sending to everyone.
What we found
Both emails sold more; the men's email added about 7 buyers per 1,000. No targeting model beat sending it to everyone.

US online retailer · Hillstrom email experiment · 64,000 customers. Independent study using public data. Not a Dunwell client.

Read the report Open the dashboard PDF
Shortage and waste cost per 100 units, test weekBest simple rule13.9Forecast-based plan7.6−45%lower shortage and waste cost17%of demand lost to stockouts

03 · Stock & Demand Planner · Public data

Stock & Demand Planner

The question
How much demand do empty shelves hide, and how much should each store order?
How we answered it
Rebuilt the demand hidden by out-of-stock hours, forecast seven days ahead with a range, and tested the order plan against simple rules on a held-back week.
What we found
About one sale in six was lost to empty shelves. Planning from the forecast range cut shortage and waste costs by 45% against the best simple rule.

Online fresh-grocery retailer · FreshRetailNet-50K · 50,000 store-product pairs. Independent study using public data. Not a Dunwell client.

Read the report Open the dashboard PDF
Share of visits reaching each stepViewed a product20.5%Added to cart5.3%Started checkout1.9%Added payment1.3%Bought1.0%One purchase in four had no add-to-cart event

04 · Conversion Check · Public data

Conversion Check

The question
Where do visitors drop out before buying, and what should be tested first?
How we answered it
Checked the tracking before trusting it, rebuilt every visit into a funnel, compared groups like for like, and tested a model that spots likely buyers early.
What we found
One purchase in four had no add-to-cart event, so tracking comes first. The model's top 10% of January visits held 38% of the buyers.

Google Merchandise Store · Google Analytics sample · 270,154 visitors. Independent study using public data. Not a Dunwell client.

Read the report Open the dashboard PDF
Contribution before returns, last 12 months (simulated)Kept · £226kLost to returns · £322k59%of contribution lost28%of items returned

05 · Returns & Margin Lab · Synthetic data

Returns & Margin Lab

The question
How much do returns really cost, and can we predict which items will come back?
How we answered it
Built a contribution model with every cost assumption in the open, tested every product grouping for real differences, and tried to predict returns at checkout.
What we found
In this simulation, returns took 59% of contribution. Nothing predicted which items would come back, so the fix is shop-wide, not a list of products.

TheLook, a clothing shop Google made up for public use · 180,516 order items. Synthetic demonstration. Not a client result.

Read the report Open the dashboard PDF

What a project includes

What's standard, and what's extra.

Standard in every study

  • A one-page summary and a written report
  • A metrics sheet with the key numbers
  • Up to five actions, ranked by effort and likely value
  • An optional 30-minute walkthrough

Paid extras, quoted separately

  • A live dashboard like the ones shown here
  • Prediction models, such as the churn and buyer models

The samples show the full range of work. You pay only for what your question requires. See pricing.

Want this analysis for your shop?

Send a short brief with your main question. I'll confirm your data can answer it and reply with the scope, price and a start date.

Start a brief See pricing