Growth value and decision metrics
9 min read
Lesson objectives
- Choose a value and guardrail metric that distinguishes customer and business outcomes from traffic volume.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
When a marketing team says growth reached 30%, first ask what grew. Visits or opened accounts may rise while orders with a positive contribution margin fall. Growth that can guide investment is an increase in a defined form of customer value after costs and delayed effects are considered. Start with the value model before selecting a dashboard. For a subscription business, management may need the contribution of a customer cohort after service and acquisition costs, rather than registrations alone. For a marketplace connecting buyers and sellers, counting one side is insufficient: buyer numbers can rise while sellers cannot fulfill demand. State the unit of the decision explicitly: a new customer, a completed order, or a cohort that joined in a given month. Fixing that unit makes comparisons across channels and periods interpretable.
An outcome metric describes the result to improve; driver metrics show where a team may intervene. Suppose net order contribution equals order revenue less merchandise, fulfillment, support and expected return costs. It can be decomposed into the number of orders multiplied by average contribution per order. Orders can then be decomposed into qualified visitors, the share reaching a suitable product page, and the share completing a real purchase. This is an explanatory tree, not a proven causal equation. A steep discount might attract more visitors and reduce contribution per order at the same time. Click or registration rate does not become the central measure of customer value, sometimes called a north star metric, simply because it is easy to collect. Define a leading team metric that is close to both customer value and business value, and place a financial outcome and protective measures beside it.
Every metric needs a short definition card: numerator, denominator, time window, person or order unit, event source, exclusion rule and data maturity point. A reported 5% conversion rate is ambiguous when one team divides orders by sessions and another divides purchasers by users. A person may place two orders under the first definition but is counted once under the second. Mixing those denominators invalidates a campaign comparison. Distinguish an early signal from a later outcome as well. Completing account setup may predict use, but it is not paid use. If an event definition changes during an experiment, pause the time comparison or recompute the baseline with the same definition. Correct arithmetic cannot rescue an inference based on unlike definitions.
A guardrail is a measure that constrains the decision when an apparent improvement harms quality, rights or sustainability. For subscriptions, it could be refund rate, billing complaints or support response time. Set the decision rule before seeing the result: increase spend if net cohort contribution improves over a sufficient period and refunds remain below an agreed threshold. Choosing the threshold from the observed result makes almost any number defensible after the fact. Distinguish a stopping boundary, which halts a harmful experiment, from a long-term improvement target. Missing guardrail data do not imply that the guardrail is healthy; a scaling decision may need to wait until returns have matured. A named owner for definitions and escalation is useful, but an ownership chart does not explain how the guardrail protects value.
To turn the tree into a weekly decision, attach each branch to an action that can be tested. Fewer qualified visits might call for a search visibility check or a budget review. A fall in completed orders with steady visits may indicate payment friction or a faulty measurement event; reconcile actual order records before changing the page. Lower contribution alongside more orders could reflect a discount, higher service cost or more returns. Do not jump from a moving indicator to one cause. Write at least two competing explanations, the evidence that would distinguish them, and what the team will do if that evidence is unavailable. This makes a growth dashboard a diagnostic and learning tool rather than a ranking of teams by arrow color.
When a report draws from web analytics, do not automatically equate a recorded purchase event with a paid order in the order system. A browser may send the event twice after a page refresh, and it may not send it when tracking is blocked. Reconcile periodically: unique paid orders in commerce records, purchase events with a unique order identifier, and the difference by device, channel and day. If commerce records show 800 orders while analytics shows 760 unique order identifiers, the gap is 40 orders, or 5% of the commerce count. Multiplying 760 by 800/760 does not establish corrected figures for every channel; missing observations may be unevenly distributed. A directional trend can be monitored with an explicit confidence warning, but a large budget transfer requires an estimate of how this gap could change the ranking of alternatives. Product-specific measurement behavior should be checked against its documentation, while the company’s own order and value definitions come from its transaction records.
Present the result to management as a decision rather than a crowded dashboard. For this case: over comparable two-month horizons, B cohort contribution after campaign cost is 12,200 against 1,000 for A. Keep A spending bounded while testing whether later sales or incremental impact changes the ranking, and monitor returns before scaling B. This statement separates what was calculated from what remains unknown and ties uncertainty to an action. If most B buyers would have purchased without the ad, the funding decision could change even while B descriptive contribution stays higher. If B margin falls as auction prices rise with scale, do not extrapolate its old average performance to a larger budget without testing. In every presentation ask which changed fact would reverse the decision, and make that the next measurement priority.
A two-sided product also needs supply-demand balance. If buyer requests rise 20% while capable sellers remain fixed, completed-order share may fall and waiting time may rise. A created-orders metric would reward a campaign that harms the experience. Use completed positive-contribution orders over a common horizon as the outcome, with fulfillment time and seller opportunity as guardrails. A rise on one side alongside deterioration on the other requires a capacity decision, not a celebratory arrow.
Lesson summary
Separate growth value from activity and set a guardrail and decision rule.