CMO playbook & advanced tactics for budget negotiation

The hardest call in this job is not defending a number in front of the CFO. It is deciding which of your own lines to kill when the evidence says the money does nothing, while eleven people have built their year around that line and your read of the evidence is wrong more often than you would like. Cutting spend that works and cutting spend that only looks like it works produce an identical saving this quarter and opposite results six quarters out. Arbitrating between those two, at a scale where one call moves tens of millions, is the part of the role that does not delegate.

Where the kill list comes from

By the time you are arbitrating you already have the cost lines the foundations lesson sets out and the zero-based build the frameworks lesson walks through. Neither gives you permission to cut. A zero-based exercise ranks spend by justified need; it says nothing about whether the money moves anything, because most marketing reporting measures spend that sits near an outcome rather than spend that caused one. Branded search, retargetingretargetingShowing ads to users who have previously visited your site or interacted with your brand, to bring them back and drive conversion.View full definition → and app-install campaigns post the highest attributed returns in almost every stack, and they are the three lines most likely to be buying customers who were already on their way.

The arbitration rule is narrow. Nothing gets killed on attributed ROIROIReturn on Investment: the ratio of net profit to the cost of an investment. A 300% ROI means each dollar invested returns $3.View full definition →, and nothing survives on it either. You kill on the gap between a market that had the spend and a comparable market that did not.

1. Prove the null before you kill

eBay ran the cleanest public version of this. With economists Thomas Blake, Chris Nosko and Steve Tadelis, it switched off paid search across a set of US markets and watched what happened to traffic and sales. Branded keywords returned close to nothing: people who typed "eBay" into Google clicked the organic result instead, so the company had been paying for clicks it already owned. Non-brand search did generate incremental sales, but the effect sat almost entirely with people new to eBay or who used it rarely. Averaged across all users, returns were negative by a wide margin. The study ran in Econometrica in 2015.

The edge case matters more than the headline. The honest conclusion was not "search advertising does not work." It was that the same keyword was pure substitution for one slice of the audience and genuinely incremental for another, and eBay had been paying an identical price for both. A blanket kill would have thrown away the working half. Worth noting that eBay today sells promoted listings to its own sellers, so its current published position on ad effectiveness is not disinterested.

2. Grade every cut by how reversible it is

Media can be switched off on Monday and back on the following Monday, at a price. Nothing else on your sheet behaves that way. Committed upfront rates, sponsorship terms, agency capability and, above all, people take quarters to rebuild and rebuild at a worse price. Uber cut roughly 400 marketing roles in mid-2019, close to a third of its marketing organisation, folding brand and performance teams together. Whatever the merits, that is not a decision you reverse in a planning cycle.

Before any kill list goes to the CFO, mark each line reversible, slow to reverse, or gone for good. Take the reversible savings first even when they are smaller. A CMO who delivers the same number using only reversible cuts has bought the organisation an option; one who hits it by dismantling capability has spent one.

3. The two mistakes are not symmetrical

Cut spend that was not working and you see the saving inside a month with no visible damage. Cut spend that was working and, if it was brand rather than activation, the damage surfaces over six to eighteen months as base demand drifts down, price sensitivity rises and your acquisition costs quietly climb. Les Binet and Peter Field's IPA work is the standard reference for that lag. The feedback loop therefore rewards the wrong cut on the timescale a leader is actually judged on, which is why brand lines get killed disproportionately.

Procter & Gamble's 2012 cost programme, announced at $10 billion across cost of goods, overhead and marketing, ran alongside a stall in organic growth. A. G. Lafley's return in 2013 shifted the arbitration from how much to what: P&G divested or discontinued around 100 brands to concentrate spend behind roughly 65 that produced almost all the profit. The instructive part is that the answer to weak returns was a smaller portfolio at full weight rather than a full portfolio at reduced weight.

4. Name who eats the cut, in writing

An equal-percentage reduction across every line is the default an organisation reaches for under pressure, and it guarantees the wrong outcome: it protects the largest weak line and starves the smallest strong one. If you do not publish the criteria, the cut gets allocated by whoever negotiates hardest internally, and that is rarely correlated with contribution. State the threshold, apply it to your own favourite programme first, and let people see you do it.

What the big kills actually taught

Uber in 2017 shut off around two thirds of its digital advertising, on the order of $100 million a year, after an ad fraud investigation. App installs barely moved. Kevin Frisch, then running performance marketing, discussed it publicly; Uber went on to sue its mobile agency Fetch Media over the spend. The finding that travelled around the industry ("$100m of ads did nothing") is the wrong lesson. Uber had enormous unaided demand in its core markets, so a large share of its install spend was buying downloads that were coming anyway. A challenger with a tenth of that awareness would get a different answer from the identical test. Uber's total sales and marketing expense still fell from roughly $4.6 billion in 2019 to about $3.1 billion in 2020, which tells you the arbitration continued well past the fraud.

P&G took the opposite route through the same problem. Marc Pritchard cut a reported $200 million of digital spend in 2017 over viewability and placement quality, and said reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → went up rather than down, then attacked the fee stack: over $750 million in agency and production savings, with a further tranche targeted after that. What P&G did not do was reduce total advertising weight, which has stayed in the region of $7 billion a year. Kill the waste and the intermediary margin, hold the media pressure.

CMO action items

• Run a rolling holdout schedule so a kill decision never waits on a new test. Two or three matched geographies dark on one channel at a time, rotating quarterly, gives you evidence in hand when the cut request arrives at short notice.
• Attach a restart cost and a restart lead time to every line on the kill list. "Saves $4m, restarts in three weeks" and "saves $4m, restarts in nine months at a worse rate card" are different decisions and should not appear on the same table without that column.
• Write the reinstatement trigger at the moment you cut, not later: the specific metric, the threshold and the date you review it. Cuts made without one become permanent by inertia rather than by decision.
• Segment before you kill. If a channel is incremental for new or lapsed customers and substitutional for your existing base, the answer is an audience exclusion, not a line item removal.

How leaders get this wrong

The commonest error is killing the measurable because it is measurable. Performance channels can be tested and therefore can be proven guilty; sponsorship, brand and anything with a lag cannot be, so they survive audits they would fail. Over three or four cycles this systematically hollows out the half of the budget that compounds.

The second is cutting to a percentage rather than a threshold. A 15% instruction from finance is an allocation problem, not an evidence problem, and it invites you to spread the pain. Come back with a list ranked by measured contribution and let the number fall where the evidence stops.

The third is treating a null result as permanent. Media that did nothing at a bad price, with bad placement, in a saturated market, may work under different conditions; P&G reinvested behind exactly the channels it had cut once the quality problem was fixed. Reopen the question annually.

The last is personal, and rarely said aloud. Median CMO tenure runs at roughly three to four years, the shortest in the C-suite, while the damage from a wrong brand cut takes longer than that to fully surface. The savings are credited to you and the consequences to your successor. That asymmetry is precisely why the reinstatement trigger and the written criteria matter: they are the only record that the call was reasoned rather than convenient.