The case for fewer KPIs in UA

A typical mobile game UA dashboard carries somewhere between thirty and sixty metrics. A typical weekly UA review changes three or four things: a budget, a bid, a creative rotation, sometimes a geography. The ratio between what the team tracks and what it acts on is the clearest evidence that most of the tracking isn't measurement. It's reassurance.

The argument here is that a UA operation should run on a small number of KPIs, probably fewer than eight, each explicitly paired with a decision it can trigger, and that everything else should drop to diagnostic data, consulted only when a KPI moves. Plenty of experienced growth leads will object that mobile UA is too multi-dimensional for that, and that the extra metrics are cheap insurance against missing something. The insurance isn't cheap. The team pays for it in attention, and attention is the scarcest input a UA team has.

How dashboards grow

Metrics accumulate through a simple ratchet. Something goes wrong. Someone adds the metric that would have caught it, and nobody ever removes it, because removing a metric is a decision and adding one is a reflex. A new network arrives with its own reporting conventions and its metrics join the panel. A stakeholder asks a question once and a tile gets built to answer it forever; creative testing at scale multiplies rows, because every variant needs its own line, and cheaper reporting tooling multiplies columns, because every question is now trivial to answer permanently. Nothing in either process asks whether a metric has changed a decision in the past year.

There is also a defensive motive. A large dashboard is protection. When a quarter goes badly, the team can point to the metric that was flagging the problem. The fact that nobody acted on it is a separate conversation.

Why more metrics produce worse buying

The mechanism isn't that extra metrics are wrong; it's that they compete for the same interpretive bandwidth and dilute the signal of the ones that matter.

With forty metrics on a panel, some will move materially every week by chance alone. The team learns, correctly, that most movements mean nothing, and calibrates towards ignoring them; that calibration then leaks onto the three metrics that actually should trigger action. A rising CPI on the primary network gets the same shrug as a wobble in day-one retention on a minor geography, because the dashboard has trained the team to expect noise.

A second effect: when many metrics are available, the review meeting selects the ones that support whatever the room already believes. A buyer who wants to keep a campaign defends it on day-thirty ROAS if day-seven looks bad, or on install volume if ROAS looks bad. A large metric set makes every position defensible and therefore makes no position decisive.

The trade-off nobody states

Fewer KPIs means accepting that some problems surface later than they might have. A lean panel won't flag a creative fatigue pattern on a secondary network in week one; it will flag it in week three, when the network's blended return crosses a threshold. That delay is a real cost.

The compensating gain is that when the lean panel does flag something, the team believes it and moves. In practice the reaction time saved by a credible signal exceeds the detection time lost by a sparser one, because most of the delay in UA operations is not in noticing problems but in agreeing they are problems. A dashboard that produces consensus quickly is worth more than one that produces early, disputed warnings.

A construction rule

Build the KPI set backwards from decisions, not forwards from data.

Start by listing the decisions the team actually makes at each cadence, not the ones it could in theory make. Daily: pause or continue a campaign, rotate a creative. Weekly: move budget between networks and geographies, change bid targets, approve concepts for production. Monthly: set the total envelope; add or drop a partner; revisit payback assumptions.

For each decision, name the single metric that would most change the answer, and the threshold at which it would. If two decisions share a metric, good. If a metric doesn't map to any decision it is diagnostic, and it leaves the panel.

An illustrative result for a mid-sized hybrid-casual studio might be six KPIs, each with its threshold and action in one clause:

• Blended cost per payer above the plan figure for two consecutive weeks: freeze the total envelope until it returns.
• Day-seven ROAS by network below the target band for two weeks: cut that network's budget by a fixed step, say a fifth.
• Predicted day-ninety payback by geography beyond the ceiling for a month: pause scaling in that geography and open the diagnostic layer.
• Creative concept win rate against the control below one in eight over a quarter: change the concept brief, not the production volume.
• Organic install share of total falling below its floor for three weeks: commission a baseline check before approving any further paid scale.
• Incremental installs per thousand dollars from the most recent holdout below the hurdle: reallocate from that network before the next holdout, not after.

Nothing else appears until a KPI crosses a line, at which point the diagnostic layer opens for that KPI only.

Note what is absent. No CTR, no IPM, no CPM. Those are inputs to CPI, and CPI is an input to cost per payer. They matter to the buyer running the account daily. They don't belong in the review that decides where money goes.

What changes in the room

The review meeting gets shorter, and more disagreeable in a useful way, because with six numbers everyone can hold the whole picture at once. Movements are either above threshold or not. Discussion concentrates on the one or two that crossed, and the question is which of the pre-agreed actions to take, not whether the movement is real.

The other change is political. A lean KPI set forces the team to declare in advance what it will do when a number moves. That commitment is uncomfortable, because it removes the option of arguing the metric away later. Studios that adopt this tend to find that the discomfort is the point: the dashboard stops being a place to look for reasons and becomes a place where the decisions already exist, waiting for the number to arrive.

The diagnostic metrics don't disappear. They move to a second screen that opens on demand, and someone still owns their integrity. But they stop being the thing the team stares at on Monday morning, and that is where most of the improvement comes from.

Related archive reading

• More creative testing can make you worse
• ROAS targets versus payback windows