# Recovery cost and measurement worksheets

Publisher: Zobi
Updated: 4 November 2026

Use actual inputs for one reporting period. The worksheets do not supply vendor rates, store results or an assumed incremental effect.

## Costs

Each charge has a state: amount entered, does not apply, or unknown. Unknown is not zero. A complete program total requires an explicit state for every cost row. Count a charge once, even when it appears in several documents.

## Contribution sensitivity

Basket contribution equals attributed-order net revenue after discounts and refunds, minus the variable costs of those orders.

Modeled additional contribution equals assumed additional share times basket contribution, minus program costs, minus otherwise-lost contribution.

When basket contribution is positive, the conditional break-even share equals program costs plus otherwise-lost contribution, divided by basket contribution. Displayed shares within 0% to 100% round up to two decimal places, so the displayed share does not fall below the modeled threshold. The JSON record retains the unrounded ratio. A share above 100% means that modeled basket cannot cover those costs. Zero or negative basket contribution has no positive-contribution threshold.

The model assumes that additional orders have the attributed basket's contribution mix. Otherwise-lost contribution is a separate estimate of margin given up on purchases that would have happened without the program, such as a discount on an otherwise full-price order. It is not the basket's total discounts. Net revenue already uses actual discounts and refunds; the loss allowance accounts for the counterfactual change on the otherwise-occurring orders. Do not enter that same loss again as a program charge. The displayed threshold and sensitivity hold the entered basket, program costs and loss allowance fixed. If the loss or contribution mix changes with the assumed share, change both inputs and run those scenarios separately.

This is a planning model. Attribution does not prove that an order was caused by a program.

## Group reconciliation

Use comparable eligible units and a common observation window. Deduplicate orders, separate subscription renewals where appropriate, and use one net-revenue and cost definition.

Compute treated contribution per eligible shopper and control contribution per eligible shopper. Subtract the control value from the treated value, multiply by the treated count, and subtract additional program costs not already included in order costs. The worksheet normalizes before subtraction and rounds only the final cent.

Totals do not contain person-level variance. They cannot establish confidence intervals, statistical power or a reliable test duration. Arithmetic does not verify randomization, eligibility, contamination or experiment validity. A positive point estimate alone is insufficient evidence of causal lift.

## Conversation cases

Write an actual shopper task, verified merchant facts and their source/version. State needed clarification, permitted actions, escalation conditions and expected behavior. Record what was actually observed. A pass or fail requires facts, a source, an expectation and an observation. An unrun or evidence-blocked case stays in that state.

The empty case sheet is a reusable template, not evidence of an assistant's performance. Source policy changes and changed permissions can invalidate an earlier result.

## Sources

- [Zettler Digital, measuring email and SMS incrementality](https://zettlerdigital.com/blogs/news/how-to-measure-email-sms-incrementality-with-klaviyo), group and order-reconciliation method. Suggested test splits or durations are not universal guarantees.
- [Common Thread Collective, measurements for profitability](https://commonthreadco.com/blogs/coachs-corner/building-useful-measurements-for-profitability), revenue, costs and appropriate causal measurement.
- [Red Stag, interpreting cart abandonment](https://redstagfulfillment.com/percentage-of-online-shoppers-abandon-their-cart/), source definitions, recoverability and additional-profit limits.

The source passages support definitions and method principles. Zobi's formulas are an arithmetic implementation, not a reported merchant result or a claim of a novel statistical method.