Which marketing numbers actually mean anything

There is one test for whether a metric belongs on your dashboard: if it moved 30% next month, would you do anything differently? If the honest answer is no, it is not a metric, it is decoration — and every decorative number on a dashboard makes the useful ones harder to see.

One outcome, a few leading indicators

Pick a single outcome number that the business actually cares about — revenue, qualified leads, trials started, orders. That is the lagging indicator: it is what you are trying to move, and it moves last.

Then pick two or three leading indicators that move earlier and predict it. These are what you check weekly, because they tell you whether the work is heading somewhere before the outcome confirms it. The pairing is what makes a dashboard useful; either half alone is misleading.

• For SEO: number of queries where you appear at all, and average position on your target set. Both move months before the traffic does.
• For content: pages that entered the top 20, and clicks on the specific pages you published this quarter.
• For social: replies and saves rather than impressions, and profile visits rather than followers.
• For email: reply rate and click-to-open on the segments you actually sell to, plus list health (bounces, complaints).
• For paid: cost per qualified action, not cost per click — and the qualification has to be real.

The numbers that are usually noise

• Impressions. They measure how often something was shown, which you do not control and cannot spend.
• Followers. A count that only goes up is not a measurement of anything.
• Page views without a destination. A visit that leads nowhere is indistinguishable from a bounce, in business terms.
• Open rate. Privacy features inflate it enough that it is now directional at best.
• Keyword rankings for terms nobody searches. Position 1 on a phrase with three searches a month is a screenshot, not a result.
• Time on page, as a target. It rises when people cannot find what they came for.
• Anything reported without a comparison. A number with no baseline and no expectation cannot be good or bad.

None of these are worthless as context. They become harmful when they are reported as results, because they can be improved without the business improving at all — and eventually someone optimises for them.

Attribution, briefly and honestly

Attribution is worse than most dashboards admit. Cookie restrictions, cross-device journeys, dark social and the simple fact that people search a brand name after seeing it elsewhere all mean that a meaningful share of your results will be credited to "direct" or "organic search" regardless of what caused them.

Two practical responses. Ask new customers, in one optional field, how they heard about you — self-reported attribution is imprecise but it catches the channels analytics cannot see. And judge channels by whether the outcome number moves when you change what you do, rather than by the credit the platform assigns itself.

Sample size, or why most weekly reviews are theatre

A site with 200 conversions a month cannot detect a 10% improvement in a week. The noise is larger than the effect, and reviewing weekly guarantees you will find a story in the noise and act on it. That is worse than not looking.

• Under ~100 conversions a month: review monthly, and only act on changes large enough to be visible without a chart.
• 100–1,000: monthly review, quarterly judgement on anything strategic.
• Above that: weekly is defensible for operational numbers, monthly for direction.
• Always: give a change enough time to be measurable before layering another change on top of it.

Timeframes per channel

Different work reports back at different speeds, and judging a channel on the wrong clock is how good work gets killed:

1. Paid: days. Fast feedback is the whole point of paying for it.
2. Email: days to a week. Send, measure, adjust the next send.
3. Social: weeks. Individual posts are noisy; the format mix is the signal.
4. Content and SEO: two to four months for a new page to settle, longer on a young domain. Judging an article at ten days tells you only that it has been indexed.
5. Brand and reputation: quarters. Mostly not measurable in a dashboard, which is not a reason to stop doing it.

A monthly review that works

Thirty minutes, same shape every month:

1. Write the expectation first, before opening anything. "I expect trials to be roughly flat and query coverage to rise, because the three pages we published target new topics."
2. Look at the outcome number against last month and the same month last year, if seasonality applies.
3. Look at each leading indicator. Did it move in the direction the plan predicted?
4. Find one thing that surprised you and understand it before doing anything else. Surprises are where the information is.
5. Decide one change for next month. One — more than that and next month's review cannot attribute anything.
6. Write down what you expect from the change. That sentence is what makes next month's review honest.

The written expectation is the entire discipline. Without it, every review becomes an explanation of what already happened, which always sounds convincing and never predicts anything.

A dashboard that fits on one screen

• Revenue or qualified leads, this month against last.
• Query coverage and average position on the target set.
• Clicks to the pages published in the last quarter.
• Email list health: replies, unsubscribes, bounces.
• Cost per qualified action, if you spend on ads.
• One qualitative line: what customers said this month that you did not expect.

Six things. If a seventh is genuinely needed, something else comes off — a dashboard that requires scrolling is a dashboard nobody reads to the bottom of.

Frequently asked questions