The dashboard invites a bigger bet
An online clothing shop owner must choose next month's advertising plan. More paid orders make the broad campaign look attractive, but a return-processing backlog leaves refunds out of the sales report. Scaling that apparent success could leave less money for overhead.
Price the orders customers keep
Hypothetical example; all figures are assumptions, not reported results. Order forecasts count additional purchases caused by advertising. Customers pay $60 per order. A retained order leaves $30 after product and delivery costs. A returned order is fully refunded and loses $10 in shipping and handling; returned stock is assumed resalable.
Both plans fit a $1,200 advertising allowance. The broad plan spends $1,200 for 100 orders, including 30 expected returns. Contribution is 70 × $30 − 30 × $10 − $1,200 = $600.
The focused plan spends $600 for 60 orders from previous buyers, including six expected returns. Contribution is 54 × $30 − 6 × $10 − $600 = $960. Count eventual refunds, even if they are paid later.
Make the queue visible
Choose the focused plan under these assumptions: it leaves $360 more contribution within the same budget. Contribution must still cover overhead; receipts also include money owed back to customers.
Match order_id across sales and returns; inspect return_requested and refund_posted. A $60 payment marked return_requested=yes, refund_posted=no still appears in sales while the refund waits. Processing it reverses the sale and leaves the $10 return cost.
If requested returns lack refunds, include those refunds and handling costs before approving spend. Compare customer cohorts after the same return window. The forecasted return difference needs checking; a smaller budget alone does not cause fewer returns.
When the broad plan earns its place
Now change the example: hold other assumptions fixed, but expect only 20 broad-campaign returns. Contribution becomes 80 × $30 − 20 × $10 − $1,200 = $1,000. It beats $960. Choose broad below 21 returns; at 21 the plans tie. Above that threshold, choose focused, unless verified inputs change.