Anatomy of a budget cut done well

A budget cut is the only moment in the year when a UA team has to say which of its spend it believes least. Most teams avoid saying it. They take the reduction proportionally across channels, defend the decision as fair, and end up with a smaller version of the same allocation, including the same slice of spend that was never earning its keep. The thesis of this piece is that a proportional cut is the worst available method, and that a cut done well should make the remaining budget more efficient than it was before the cut, not merely smaller.

That sounds like a consultant's promise, so the mechanism deserves spelling out. Every channel a team buys on has a marginal return curve: the first dollar buys the cheapest, most eager installs, and each further dollar buys slightly worse ones until the auction is pushing you into users who would not have installed at any sensible price. A budget is a stack of these curves. A proportional cut trims every curve by the same fraction, which removes some genuinely marginal spend and some genuinely productive spend in equal measure. A marginal cut trims only from the flat tails. The second method removes more waste per dollar cut, by construction.

Why proportional cuts happen anyway

The incentive structure explains the persistence of the bad method better than ignorance does. A proportional cut requires no one to admit that a channel they defended is the least productive. It produces no loser in the room. It also fits the way most teams see their data: platform dashboards report average return per channel rather than marginal return, and an average-return view makes every channel that clears the target look equally worth keeping.

The finance side has its own reason to prefer proportional cuts. They're legible. A twenty per cent reduction across the board fits in one line, whereas a cut that removes forty per cent from one channel and nothing from another invites a conversation about why anyone funded that channel at that level in the first place. That conversation is exactly the one a well-run team should want, and exactly the one a proportional cut lets everyone skip.

The second-order effect of cutting the tail

Here is the part most budget coverage misses. When a team cuts spend from the flat tail of a channel's curve, it does not only save money. It changes the signal the channel's algorithm is optimising on.

Automated buying on the large networks learns from the conversions it delivers. Spend at the tail delivers low-value conversions, and those conversions train the model toward more of the same. Pulling budget back to the productive part of the curve tightens the conversion signal, and over a few weeks the model's own targeting often improves. Teams that have made this kind of cut sometimes report that return on the remaining spend rose by more than the mechanical curve arithmetic would predict. That isn't magic; it's better data reaching the model.

The trade-off is that the tail was also where a lot of the volume came from, and volume has organic and ranking effects that do not show in channel return. A cut that goes too far into the productive part of a curve can push a game down a category chart and cost organic installs the dashboard will never attribute to the cut. UA Ledger's earlier piece on the paid-to-organic ratio, written off the back of Adjust's March report on gaming sessions and retention, is the right background reading on how large that multiplier can be. Size the cut with that multiplier in mind, not just the direct return.

A worked illustration

Consider an illustrative studio spending a monthly UA budget across four channels, asked to reduce by a quarter. The proportional method takes a quarter from each. The marginal method starts differently.

For each channel, the team pulls the last eight weeks of daily spend and attributed day-seven revenue and plots them. Two channels show a clear knee: return holds until a certain daily level, then falls away. One channel is roughly linear, still buying productively at its current level. One channel has been flat for months, with return barely above the target regardless of spend, which usually means the channel is buying the same users the others already reach.

The marginal cut removes everything above the knee on the two curved channels, takes nothing from the linear channel, and halves the flat one as a test rather than an assumption. Total reduction lands around the required quarter. In this illustration the blended return on the remaining budget rises, because the removed spend was returning below the average by definition.

The flat channel is the important call. It may be genuinely useless, or it may be providing reach the other channels depend on. The only way to know is to cut it hard for a fixed window and watch whether the other channels' cost rises. If they hold, the cut becomes permanent. If they rise, the channel was doing invisible work and gets partly restored.

A decision rule

For a team that has to cut and doesn't have clean incrementality tests on every channel, a practical rule in three steps.

Cut the tail first. Any spend above a visible knee on a channel's return curve goes before anything else.

Test the flat. Any channel whose return doesn't respond to spend gets a halving for a defined window, and the other channels' costs become the outcome measure.

Protect the slope. Any channel still buying productively at the margin gets cut last and least, even if its average return is lower than a flatter channel's.

The last point is the counterintuitive one. Average return is the wrong sort key. A channel with a mediocre average but a live slope is buying growth. A channel with a good average and a dead slope is buying maintenance. Under a cut, growth is what you keep.

What to bring to the finance conversation

The document that justifies a marginal cut is not a spreadsheet of channel averages. It is four curves, one per channel, with the proposed cut marked on each. That single page does more to move a finance conversation than any amount of argument about efficiency, because it makes visible the thing proportional cuts exist to hide: that not every dollar in the budget was doing the same work, and the team knew which ones weren't.

A team that can produce those curves on demand has also quietly solved the following quarter's problem. When the budget comes back it goes onto the slopes rather than the tails, and the allocation that emerges is the one the team should have been running all along.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• Why your best network degrades as you scale it
• Post-mortems for failed campaigns, done honestly