A budget cut was imposed on us at short notice and we lost roughly half our media spend for two quarters. It was not an experiment and it functioned as one, and what we learned reshaped the plan afterwards.

Details are generalised, since this was a real business.

What we cut and how we chose

We had no time for analysis, so we used a crude rule: cut whatever we could not defend with evidence of incremental effect.

That removed most display prospecting, a large share of retargeting, several always-on social campaigns, and two sponsorships.

We protected branded search, one performance channel with proven incrementality, and email, which cost almost nothing anyway.

In retrospect the rule was better than anything we would have designed with more time, because it forced the question of what evidence actually existed for each line.

What happened to revenue

It fell, and by a fraction of the budget reduction. Roughly speaking, a fifty percent spend cut produced a single-digit percentage revenue decline over the two quarters.

Which sounds like a triumph and requires two caveats.

The first is that some of the effect was delayed. Reductions in upper-funnel activity do not show up immediately, and we saw a further decline in the following quarter that was plausibly attributable.

The second is that the period included seasonal factors and market conditions we could not separate out.

So the honest reading is that a large share of the cut spend was doing much less than it was credited with, and a smaller share was doing something real with a lag.

The retargeting finding

The most striking single result.

We cut retargeting almost entirely and the effect on conversions was close to undetectable.

This has been reported by others running proper tests, and it was still surprising to see in our own numbers, because retargeting had reported the best return of any channel for years.

The explanation is straightforward once stated. Retargeting reaches people who have already visited and are already considering. A high proportion of the conversions it is credited with were going to occur.

It is the purest example of attributed performance diverging from incremental performance, and it is where most organisations are overspending.

What we lost that mattered

Not everything was costless and it is important to say so.

New customer acquisition fell, and specifically acquisition of people with no prior awareness of us. That did not show in revenue immediately because those people would not have purchased in the period anyway.

It showed up later as a smaller pool of people in consideration, which is exactly the mechanism that makes upper-funnel cuts look free in the short term and expensive over eighteen months.

Branded search volume declined slowly over the period, which is the cleanest available indicator of awareness and which we now watch closely.

What we did afterwards

The budget was partially restored, and we did not restore the old allocation.

Retargeting came back at a small fraction of its previous level, capped and frequency-limited.

The money went to upper-funnel activity with a proper measurement framework attached, which we had never had.

We built a holdout testing capability, which we should have had years earlier and which the cut made politically possible for the first time.

And we adopted a rule that any always-on activity must be tested for incrementality annually or be discontinued, which has since killed two more channels.

The organisational lesson

The thing I would emphasise to anybody in a similar position.

The cut was possible to survive because it forced a question nobody had wanted to ask, which is what evidence exists that this specific spend does anything.

In normal conditions that question is difficult to raise, because everybody's channel reports a positive return and challenging it is challenging a colleague.

A forced cut removes that dynamic entirely. Everything is being cut, so nothing is a personal criticism.

If I were advising somebody with an intact budget, I would suggest running a voluntary version — a deliberate, temporary, measured reduction in one channel — rather than waiting for an involuntary one. The information is worth considerably more than the spend.

What we would test first with hindsight

If somebody were starting from an intact budget and wanted the same information without the disruption, I would suggest a specific order.

Retargeting first, because it is where the gap between attributed and incremental performance is largest and because a test is easy to construct.

Branded search second, for the same reason, though the test is politically harder because the downside if you are wrong is immediate.

Always-on social third.

Each of those can be tested in a single region or a held-out audience over six to eight weeks, at a fraction of the cost of finding out the way we did.