Build a weekly marketing scorecard with decisions attached

Build your weekly marketing scorecard around the decisions you need to make. Give each metric a definition, a denominator where needed, an owner, and a response to improvement, decline, or uncertainty.

For a small AI or SaaS team, start with three outcome rows and a spending or effort limit. A spreadsheet is enough. Keep the underlying dashboards available when a result needs investigation.

Choose outcomes that can change next week’s work

Write the decision at the top of the sheet: “Where should we put the next week of marketing effort?” Then choose the few measures that could change your answer.

For a product with a self-serve trial, a useful starting set is:

• New qualified trial accounts, using a written definition of customer fit.
• Activation rate among trial accounts that have had enough time to activate.
• Paid conversion among an older trial cohort whose conversion window has closed.
• Acquisition spending and founder hours against agreed limits.

For a sales-led product, substitute qualified conversations held, accepted opportunities, and later sales outcomes. Define booked and held meetings separately, then choose the measure that informs your current bottleneck.

Include activity measures when they explain a constraint. If the founder planned four customer interviews and completed none, that matters. A count of social posts deserves space when it helps diagnose whether a specific plan was executed.

Keep retention, cancellations, and revenue in a broader review, and bring one into the weekly scorecard when it could change the week’s priorities. For an AI product, you might also need a limit on inference cost per activated account or the share of outputs users must redo.

Define the denominator and the clock

“Activation: 30%” leaves too much open. Thirty percent of visitors, users, accounts, or trials? Activated within a day or a month?

Write the definition beside the metric. For example:

Accounts created in the selected signup week that completed their first successful export within seven days of signup, divided by all eligible accounts created that week. Exclude employee and test accounts. Count each account once.

The numerator must come from the denominator’s population. Dividing this week’s activations by this week’s signups can mix older accounts with new accounts and produce a misleading rate. If you restrict a row to qualified accounts, apply the same qualification rule to both counts.

Funnel tools also depend on these choices. Amplitude defines total funnel conversion as users completing every step divided by users entering the first step, subject to the selected conversion window. Its conversion-over-time view groups results by when users entered the funnel. Check your report’s counting unit and window before copying its percentage. Amplitude’s funnel documentation

Choose a fixed reporting week and time zone. A cohort is the group of accounts that signed up during that period. Record when every account in the group will have completed its observation window.

For example, a Monday-through-Sunday signup cohort with a seven-day activation window is still incomplete the following Monday: Sunday’s signups have had only about a day. Wait until the last account’s seven-day window closes before comparing the completed cohort with an earlier one. A 30-day paid-conversion measure needs an older cohort, even though you review it weekly.

For rates, show the raw counts: 12 / 40 = 30%. If the denominator is zero, report N/A. Mark missing data as missing; it does not mean there was no activity. Counts such as new qualified accounts need no denominator.

For a longer comparison, combine numerators and denominators from comparable, nonoverlapping cohorts, then divide. Averaging weekly percentages can give a tiny week the same weight as a large one.

Record the source and reporting delay

Assign one source of record to each measure. Your account database might supply trial starts, product events might supply activation, and billing records might supply paid accounts. Save the report and filters so another person can reproduce the number.

For channel reporting, name the attribution scope. GA4’s First user dimensions describe initial user acquisition; Session dimensions describe session acquisition; event-scoped dimensions assign credit for key events. Those answer different questions. Google’s traffic-source scope documentation

Use one consistent scope for a channel comparison and retain an unknown-source category. A channel label describes the credit assigned under that method. It does not establish how many customers would disappear if you stopped the channel.

Track observation windows and reporting delays separately. An account can finish its activation window while the analytics report is still processing. Google says GA4 processing can take 24 to 48 hours, and attribution credit for key events can change for up to 12 days after the event is recorded. Mark affected figures provisional and record revisions. Google’s data freshness documentation

If definitions and sources are still disputed, use the marketing analytics setup brief to resolve the measurement requirements before automating the sheet.

Write decision rules before reviewing results

Each metric owner checks the data and proposes an action. That responsibility does not imply control over every cause of the result.

Set a target from your operating needs and baseline. Then define a review trigger. A target expresses what you want; a trigger tells you when to investigate. Neither is automatically a statistical threshold.

For an activation row, the rules might be:

Define what “flat” means for your row before the review. Apply the direction correctly: a falling acquisition cost may be favorable, while a falling qualified-account count calls for a different response.

For a low-volume startup, one account may move a rate substantially. Report counts, examine individual journeys where appropriate, and compare several cohorts with consistent definitions. A longer window gives you more observations at the cost of a slower decision. It still does not prove that a marketing change caused the result.

A broken signup form warrants immediate repair even when you lack enough observations to judge a campaign.

Worked example: more trials, the same number of activations

Consider a fictional AI reporting product. Its team defines activation as a successful report export within seven days of signup. It counts accounts, excludes internal tests, and compares two signup cohorts after both observation windows have closed.

All figures are hypothetical. Assume campaign spending maps to these cohorts under the team’s documented allocation rule. Spend per activated account here includes campaign spending only; it excludes founder time and other acquisition costs.

Trial volume rose 50%, while activated accounts stayed at 12. The activation rate fell by 10 percentage points. Spend per trial stayed flat, but spend per activated account rose from $50 to $75.

Before the review, the team had agreed that a decline of at least five percentage points between completed cohorts would trigger an investigation. This illustrative operating rule is neither an industry benchmark nor evidence of statistical significance.

The team holds spending at its existing cap and assigns two checks:

The numbers alone cannot distinguish weaker traffic, onboarding friction, or faulty measurement. If missing events explain the decline, repair tracking and restate the affected rows. If accounts are getting stuck before export, inspect the obstacle. When users need guidance on an unfinished step, the guide to onboarding emails triggered by a missing activation step offers a focused next task.

The team has enough information to postpone a spending increase and assign an investigation. It still needs those checks to explain the decline.

Copy this scorecard worksheet

Use one copy of this block per metric. Keep definitions stable between reviews; date any change that breaks comparability.

Decision this metric informs:
Metric name:
Owner:
Source report and saved filters:
Attribution scope and model, if applicable:

Counting unit: account / user / session / other
Numerator or count definition:
Denominator, if applicable:
Qualification criteria, exclusions, and deduplication rule:
Cohort dates and time zone:
Observation window:
Date the last account's observation window closes:
Observation status: complete / incomplete / not applicable
Data pulled at:
Reporting status: ready for review / provisional / missing / unreliable
Known reporting delay or revision risk:

Current result, including raw counts:
Previous comparable result:
Longer baseline and included cohorts:
Target and reason:
Review trigger and reason:
Range treated as flat:

If improving:
If declining:
If flat:
If inconclusive or unreliable:

Decision made:
Action owner and due date:
Next review date:
Definition changes or data revisions:

Run the first weekly review

Before the meeting, check that the populations match, observation windows have closed, and source definitions are unchanged. Note launches, outages, pricing changes, or shifts in customer mix that might explain a movement.

Start with last week’s assignments. Then discuss rows that crossed a trigger or need a decision. An unchanged row can keep its current plan; it does not need a new experiment every week.

Choose three outcomes today. Fill in their definitions and decision rules before entering the numbers. End the first review with one concrete task, an accountable owner, and a date to examine the result.