12.2 Metrics: What to Measure and What to Ignore

Every platform shows forty numbers. Five of them change decisions, and the other thirty-five consume the forty-five minutes a week you allocated for review.

This chapter is the short list, the weekly sheet, and the two measurement errors that make an otherwise healthy account look like a failing one.

The five operational metrics

Why these five. Each has a decision attached. Cost per post decides whether a tier change is worth it. Cost per follower decides whether paid promotion is a test or a bet. Saves decide format. Funnel conversion tells you which step to fix. ARPU tells you which revenue stream to push.

A metric with no decision attached is a vanity metric, whatever it measures. That is the only definition that matters, and it is why chapter 8.3 tells you to stop looking at likes: not because likes are meaningless, but because no decision follows from them.

The weekly sheet: eight numbers

One spreadsheet, one row per week, eight columns. Ten minutes to fill, and it is the whole reporting system.

Four derived numbers, computed on the same row: follow rate (5 ÷ 4), click rate (6 ÷ 4), capture rate (7 ÷ 6), and revenue per follower (8 ÷ 5). These four are the ones that actually move, because they normalize for reach, which is the number you control least.

What the sheet is for. Not reporting — diagnosis. When reach doubles and revenue does not, columns 6 and 7 show where the funnel broke. When revenue doubles and reach does not, column 8 against column 5 shows that the audience got better rather than bigger, and the correct response is to make more of what converted rather than more posts.

What to ignore

Six numbers that consume attention and change nothing.

Follower count, as a weekly metric. A lagging indicator (8.2), driven by reach and by virality you did not plan. Track it monthly at most.

Likes. Weakly correlated with distribution and the easiest number to get.

Impressions versus reach. Two names for near-identical things; the distinction rarely changes a decision.

Engagement rate as a single figure. Averaging across formats hides which format worked, which is the only thing the number could have told you. Compute it per format or not at all.

Best-post rank. Interesting once, and useless as a trend.

Any number from a competitor's public profile. It is a different account with a different pipeline, a different budget, and often a team. Chapter 8.1's warning about choosing a platform because a large account is there applies to metrics too.

Analytics per persona, not per account

The measurement error that is rarely done correctly, and it matters the moment there is more than one surface.

The problem. A persona posts natively on the primary platform, recycles to a secondary, has an ownership channel, and collects revenue on a monetization platform. Each surface reports its own numbers, and none of them is the persona's numbers.

The rule: build the sheet per persona and aggregate the surfaces into it, rather than maintaining one sheet per platform. A persona whose revenue is flat while its primary platform's reach is falling may be growing through the secondary or the channel — and if the sheets are separate, that is invisible. The opposite case is worse: a persona whose revenue looks healthy because one surface is compensating for a collapse on another.

What "per persona" means in practice. One sheet per persona, with columns for each surface's contribution to reach and to captures. It is the same eight columns with a source label, not a new system.

And the same rule applies to cost. Splitting a shared pipeline's cost across two personas is an allocation judgment, and chapter 12.3 gives the split that is least misleading: assign all shared infrastructure to the persona that existed first, and charge the second one only its marginal cost. That choice makes the second persona look cheaper than it is, which is the direction you want the error to point when the decision is whether to keep it.

A worked row

One week, filled in, with the derived rates, to show what the sheet looks like in use. The figures are illustrative. [UNVERIFIED]

What the row tells you, and what it does not. Capture rate at 22% with a click rate at 14% says the funnel is working and the constraint is reach. The response is more reach — a format test or channel work — not a landing-page rewrite. Had capture rate been 8%, the answer would be the opposite, and that is the entire value of computing the rates rather than reading the counts.

Why one row beats a dashboard. A dashboard shows forty numbers at once and invites staring. One row per week, filled deliberately, produces a trend within a quarter and a decision within two, and it costs ten minutes. The charts can come later, if ever.

The intervention thresholds

The decision rule that prevents both panic and drift. Four thresholds, each with a different response.

Why three weeks for reach and one month for formats. Reach has ordinary variance; a two-day dip is noise. Format performance is a slower signal because it needs a sample. And revenue lags both, which is why its threshold is measured in months rather than weeks — by the time revenue moves, the cause is usually two months old.

The rule that follows from the table: never change two things in response to one signal. If reach is falling, the diagnosis in 8.3 produces one cause; fixing the format and the niche and the posting time simultaneously guarantees you learn nothing, which is the same one-variable discipline as 7.2's debugging rule.

Where this goes wrong

Forty numbers reviewed weekly. The review slot is consumed and no decision follows.

Follower count as the primary metric. Lagging, and driven by things you do not control.

Engagement rate averaged across formats. Hides the only actionable information.

One sheet per platform. The persona's real performance is invisible.

Shared cost assigned evenly across personas. Makes the second one look expensive and the first one look cheap, which is exactly backwards for the keep-or-cut decision.

Fixing three things at once. No learning, and a confusing result either way.

No derived rates. Raw counts rise with reach and tell you nothing about quality.

Changing the niche at the first dip. 8.3's five-step diagnosis exists to prevent this.

The gate

1. Can you name the five operational metrics and the decision each one drives?
2. Do you fill one row per week with eight numbers, ten minutes or less?
3. Do you compute follow rate, click rate, capture rate, and revenue per follower?
4. Is your analytics organized per persona rather than per account?
5. Do you know your four intervention thresholds, and did you change only one thing last time?
6. Can you name a metric you stopped tracking because no decision followed from it?

What you leave with

Five operational metrics each tied to a decision, an eight-column weekly sheet with four derived rates and the diagnostic use of each, six numbers to stop tracking with the reason for each, the per-persona aggregation rule that makes multi-surface performance visible plus the cost-allocation choice that errs in the useful direction, and four intervention thresholds with the one-variable discipline that keeps a response from destroying its own evidence.