One of the fastest ways I've found to improve marketing efficiency is to stop paying Google for customers who were already looking for you. In the past six months, I've helped nearly a dozen companies cut most of their branded paid search and move that money into channels with real incremental return.

At a fintech I advise, we cut branded search spend by about 70% in a set of test markets over six weeks. Total conversions in those markets stayed within about 2% of the control markets, and we moved the freed-up budget into paid social and podcast sponsorships. Two things still surprise me:

The first is how hard that money works once it moves. Dollars that were buying clicks from people already searching for the company by name start reaching buyers who had never heard of it. Huge win. 

The second is how much branded search flatters the numbers around it. Someone who types your company name into Google is usually already on the way to you. Pay for that click, count the conversion, and paid search looks far more efficient than it is. Blend branded and non-branded into one ROAS figure and the strong branded number hides how weak the rest can be. The same click also soaks up credit for demand that a podcast, a webinar or an AI answer created the week before. 

And going to predict the counterargument as I’ve heard it over and over…

If a competitor bids on your name, some branded spend is genuinely incremental, and a small defensive position can make sense. A holdout test tells you how much. Conquesting doesn't justify the level of spend most of these companies were carrying.

Branded search is just the clearest case of a wider problem. A lot of marketing spend gets judged on attribution, which is mostly a record of who was around when the sale happened. I'd rather judge every marketing dollar the way a CFO judges any other use of capital.

I learned this standard at Affirm because I had to build it. We were optimizing for GAAP profitability, and for marketing that came down to one rule: if you couldn't provide incremental measurement for a program, the program didn't exist.

In our consumer business, we backed that rule with universal holdout groups. A slice of customers was held out of our marketing, so our programs could be measured against people who never saw them. The gap between the two groups was the only result that counted.

The metric we measured that gap in was revenue less transaction cost, one step closer to profit than revenue alone. A program could drive plenty of volume and still fall short if the business it brought in cost too much to process.

The accounting raised the stakes. Under GAAP, advertising costs are generally expensed as incurred or when the advertising first runs, so wasted spend is an immediate drag on profit. The accounting can't tell you whether that dollar was incremental. The holdouts did that, which is why the internal standard mattered more than the accounting treatment.

Working that way changed what marketing was inside the company. We stopped arguing for budget and started allocating capital. The best CMOs I know work like this today. They treat the budget as capital, each channel as a position, and their own job as getting the best return on the next dollar.

Our services work at StackedGTM puts me inside a lot of marketing budgets. Founding teams spending their own runway, growth-stage companies with a board asking about CAC payback, larger marketing orgs spread across a dozen channels. I'm still amazed at how much of that money gets allocated on guesswork.

The people running these budgets are smart. The weak part is the process around the money. A budget gets set as last year's split plus a growth rate. Platform-reported numbers get taken at face value. The channel owner with the best story gets the increase. Branded search survives every planning cycle because its numbers look so good.

When I start with a new team, these are the first four questions I ask about the budget:

A CFO would want all four answered before approving a capital project of the same size. Plenty of marketing budgets get approved without anyone asking.

The most useful model I've found for a marketing budget comes from portfolio management. A portfolio manager doesn't ask whether a stock is good. They ask how much of it to own, given everything else they hold, how confident they are, and what it costs to be wrong.

Every channel, program and sponsorship in a marketing budget is a position with a size, an expected return and a level of confidence. I'd size the big positions on proven incrementality and keep unproven channels as smaller, deliberate bets until they earn more. I'd watch concentration too. I wouldn't want a company leaning too hard on one channel, even a profitable one, because auction prices move, algorithms change, and new channels reprice old ones.

Returns drift, so the portfolio needs rebalancing on a schedule, before something breaks. I'd keep a fixed slice of budget as an options book: small experiments with lopsided upside, most of which go nowhere. Not every position needs to pay back in the same quarter, either. Brand and distribution can pay back over years, so I'd want to know each position's time horizon along with its expected return and how confident we are in it.

And I'd be willing to exit. Spend that repeatedly fails to show incremental lift gets cut, however long it's been in the plan and whoever championed it. Spend we haven't been able to measure well yet gets a smaller position and a plan to prove itself. Branded search is usually the first line I'd look at.

Under all of it sits one question I'd ask about last year's results: how much of that growth did marketing actually create, and how much would have happened anyway? Category demand grows, the brand is already known, and some channels mostly catch buyers who were coming regardless. A big share of branded search works that way. Only the growth marketing created should decide where next year's money goes.

The biggest gap between marketing math and finance math is average versus marginal return. Most channel reporting shows averages: total attributed revenue over total spend. A CFO deciding where the next dollar goes cares about what that next dollar will return.

Channels have diminishing returns. The first slice of a budget buys the easiest demand, and each slice after it buys less of it, at higher prices. A channel can be profitable on average and still be the worst place to put your next dollar.

The instinct is to keep feeding a channel like this because the average looks great. A smaller channel with a lower average, still early on its own curve, could return far more on that same $100k. Branded search tends to look like the right side of this curve: a great average built on demand that was already coming.

So I treat allocation as an optimization problem. You keep moving dollars until the return on the next dollar is roughly even across channels, after adjusting for how confident you are in each estimate and how long each one takes to pay back. Platform-reported numbers won't get you there, since each platform counts what it touched.

No single method answers all of this, so a setup finance will trust uses three together.

Incrementality tests are the strongest evidence we have for whether marketing actually caused a result, as long as they're designed well. Universal holdouts like the ones we ran at Affirm, geo holdouts and matched-market tests all fit here. They're slow and narrow, and they're the most trustworthy read on whether a channel causes results.

Marketing mix modeling uses historical spend and outcomes to estimate each channel's contribution and its response curve. It's how you get to marginal returns, and it picks up channels attribution can't see, like podcasts, sponsorships and offline media. Attribution itself still earns its keep for optimizing inside a channel. I wouldn't use it to move money between channels.

The strongest version calibrates the model with test results. The model covers the whole portfolio, and the tests keep it honest. A model is only as good as its data volume, the variation in your spend, and its assumptions, which is why the calibration matters.

For a long time that setup was out of reach for most teams. A credible MMM meant a data science team or an expensive consultancy, months of data cleanup, and a model that was stale by the time it landed.

Here's what I'd actually want from a measurement platform: first-party marketing and sales data in one clean place, a response curve for each channel, incrementality tests running all the time instead of once a year, and a ranked list of budget moves that comes out of all three. Until recently, that took a dedicated team.

BlueAlpha is one implementation of that approach. Full disclosure, I advise the company but I do so because they’re excellent (Peter, their CEO, shares this love/PTSD from having to prove the same thing during his time at Tesla). It combines marketing mix modeling, budget optimization and automated incrementality testing in one platform, takes on the integration and cleanup of first-party data, and ranks possible reallocations by expected lift.

It's built for lean teams where the marketing leader and the CFO are both in the room, which is the setup this whole piece argues for. When both of them read the same model, a branded search debate gets settled with evidence. I wrote recently about BlueAlpha's marketing mix model work with MUBI if you want to see it on a real business.

There's a fair objection to all of this. Some of the most important channels right now are the hardest to measure, and AEO is the obvious one.

When a buyer asks ChatGPT or Perplexity which tool to use and your company shows up in the answer, there's often no click to track. They might type your URL a week later, search your brand name, or come in through a sales rep. Attribution mostly can't see it. Holdouts are hard to design when you don't control the surface, and most mix models don't have enough history to pick it up yet.

I agree that it's messy. I don't think messy means unmeasurable, and I wouldn't use it as a reason to either ignore the channel or fund it on faith. Here's how I'd approach it: a pre/post program analysis built on two numbers.

Capture both for a stretch before the work starts so you have a real baseline, then track them weekly. As the work lifts visibility, the business metric should start moving up and to the right with it, and if it stays flat, that's a finding too. Where you can, stagger the rollout across comparable markets or product areas for a comparison group, and log launches, pricing, seasonality and other spend changes in the same window, knowing that overlapping AI answers keep this less clean than a paid media holdout.

This is how we measure AEO programs at StackedGTM day to day: visibility as the leading indicator, one agreed business metric, and the correlation between them.

Two lines moving together, adjusted for seasonality and everything else that changed, is directional evidence. It's weaker than a controlled test, and I'd say so plainly in the board update. I'd fund AEO, but I wouldn't fund it like a channel I've measured for five years, and I wouldn't leave it at zero because the measurement is young. As history builds up, it moves into the mix model and gets held to the same standard as everything else.

If I stepped into a CMO seat tomorrow, this is the order I'd work in.

When marketing reports in incremental terms, you walk into the budget meeting with evidence of what each dollar produced and a plan for where the next one goes.

The CMOs I'd bet on over the next few years are the ones who can show where the next dollar creates the most value. In a surprising number of the companies we work with at StackedGTM, the money they need to grow is already in the budget, sitting in a line whose numbers look better than its results.

P.S. If you've run a branded search holdout, or built universal holdouts on your own team, reply and tell me what you found. I'll share the most useful ones in a future issue.