Performance review template

A performance review is one row per line the plan had. Each row puts the planned spend and result beside what actually happened, states the variance, names the cause as specifically as the evidence supports, and ends in one decision with one owner and a date.

The three columns kept apart are the whole discipline: the variance is arithmetic, the cause is an explanation, and the decision is a choice. Reviews that blur them produce a conversation about the decision before anybody has agreed what happened.

Run it at the end of every flight and on a regular cadence in between. The end-of-flight review is the one that changes the next plan; the regular one is what stops a flight running to its end while somebody quietly hopes it improves.

Run it also after anything unusual — a peak, an outage, a measurement change — while the details are still recoverable.

The most common way a review goes wrong is that all three happen at once.

Somebody says a line underperformed and should be cut, and the next twenty minutes are an argument about cutting it. Nobody has established what the variance actually was, whether it is outside the range the plan expected, or whether the cause has anything to do with the line at all.

Three columns fix this by forcing the order. The variance is arithmetic: planned against actual, on spend and on result, stated plainly. The cause is an explanation, held to the evidence. The decision is a choice, made after both.

It sounds procedural and it changes the meeting, because the first two are usually uncontroversial once written down, and the disagreement narrows to the part that is genuinely a judgement.

The planned columns come out of the media plan or the flight plan, unedited.

This matters more than it looks. Reconstructed plans are, reliably, plans that describe what happened: a target quietly softened, a budget remembered as smaller, a metric recalled as the one that moved. None of it is dishonest and all of it destroys the comparison.

So the review is written beside the plan, on the same units — the same lines, the same metrics, the same currency in minor units. Where the plan left a target blank because there was no baseline, the review leaves it blank too and says so. A blank is a fact about what was known at the time.

A review that only covers exceptions produces a distorted record.

It also misses the most useful pattern available: which lines are reliably predictable. A line that has spent within a small margin of its plan and hit its target for four flights running is the most valuable thing in the account, because it is the one you can build a plan around. That fact is invisible if it is never written down.

Those rows take thirty seconds each. Planned, actual, variance small, cause none, decision continue, next review at the end of the next flight.

The cause column is where reviews earn or lose their credibility.

Write only what the evidence supports, and name the evidence: the export, the date range, the recorded event, the test result. Where there is a candidate explanation but nothing to check it against, say that — a likely cause with no evidence is still a theory, and labelling it as one is what stops it being quoted as a finding six months later.

Where the cause is simply not known, write not known. This is the entry teams resist most and it is the most valuable one in the document. A plausible story gets remembered as a fact and shapes the next plan, and the next plan is then built on something nobody ever checked.

Three causes are worth separating explicitly, because they look identical in a platform's reporting: a delivery change, a measurement change, and a real change in performance. Treating the second as the third is how a working campaign gets turned off.

Every row ends in one of four words: continue, change, stop, investigate.

Continue and stop are self-explanatory. Change requires the change to be named — not optimise, but move the weight to these weeks, or rebuild this audience, or replace this creative set. Investigate is a legitimate decision and it is what the unknown-cause rows usually get, with the question written down and a date.

Then the owner, by name, assigned in the room while everybody is present. A decision assigned in somebody's absence is a decision that gets read in an email and not made.

The next-review column is what makes the document cumulative.

Without it, each review begins from the current numbers and everybody's recollection of what was decided last time. With it, the review begins by reading the previous decisions and asking what happened to them, which takes ten minutes and changes the quality of everything that follows.

It also produces the only honest measure of whether reviews are working: the proportion of decisions from last time that were actually carried out.

The notes column carries the events that make a period unrepresentative: a peak, an outage, a measurement change, a supply problem, a competitor's promotion, a public holiday.

Writing them down at the time is the only way they survive. Six months later a line's poor month sits in the record with no explanation, gets treated as evidence about the line, and influences a plan that should never have counted it.

Seven checks. Every planned figure came from the plan rather than from memory. Every line the plan had has a row, including the ones that went as expected. Variance, cause and decision sit in separate columns. Every cause names its evidence or says not known. Every decision is one of continue, change, stop or investigate, with any change named specifically. Every decision has an owner and a next-review date, assigned in the room. And anything that distorted the period is recorded while it is still recoverable.