Notes from the operator's seat.

The growth talent market has a hole shaped exactly like a portfolio company

There are three ways a PE-backed company can buy growth expertise.

All three were designed for someone else.

The performance agency was designed for brands that need hands on keyboards. It sells channel management, prices on spend or retainers, and reports in platform metrics. The incentive structure is the problem: an agency paid on media spend has never once recommended spending less. The work can be competent and the model still fails, because the agency optimizes the channels it was hired to run, not the P&L it was never shown.

The full-time CMO was designed for companies with a decade of runway. A real one costs $400K to $700K all-in, takes six months to hire, six more to evaluate, and the median tenure is barely two years. On a five-year hold, a wrong CMO hire burns 40% of the clock. Look at lululemon if you want the expensive version of this lesson: four consecutive failed CEO successions while comps went flat for eight straight quarters and roughly $17 billion in value walked out the door. Hiring senior operators is the highest-variance bet in the building, and most portcos are structurally unable to underwrite it.

The strategy consultant was designed for the board, not the engine room. The deck is often right. Then it gets handed to a team that was never going to be able to execute it, and everyone agrees the strategy "didn't land."

What the hold period actually requires is a fourth thing: an operator with executive judgment who still does the work. Someone who can sit in the board meeting and defend the CAC math, then open the ad account and see why the math moved. Someone whose engagement is scoped against a number, EBITDA, contribution margin, ARR, and who builds the system so it runs after they leave.

This isn't a novel idea. Finance figured it out years ago: the fractional CFO is a standard play in every operating partner's contact list. Marketing is a decade behind, partly because the discipline let itself be defined by its vendors, and vendors sell what they have.

Consider what the spread looks like when the seat is filled versus empty. Vuori sold a majority stake to General Atlantic at $5.5 billion on roughly $400 million of revenue: a brand with founder presence, a real wholesale book, and unit economics that survived diligence. Everlane, the poster child of the same DTC generation, just sold to Shein for $100 million against a debt load that consumed nearly all of it. Common stockholders got a note, not a check. Same decade. Same category. The difference was never the ad accounts. It was whether anyone was operating the whole system against a number.

The white space, then, is not "better agencies" or "faster executive search." It is the seat between them: senior enough to be accountable for outcomes, hands-on enough to produce them, and structured to match the one resource a portfolio company never has enough of, which is time.

The market will fill this hole. The only question is whether your portfolio gets the benefit before your competitors' portfolios do.

An execution problem wearing a strategy costume

Most portfolio companies don't have a growth problem. They have an execution problem that has learned to dress like a strategy problem.

The tell: the company has had three strategies in four years and none of them changed the trajectory.

When the strategy changes that often and the results never do, the strategy was never the variable. The variable is the system underneath it: who owns the number, whether the measurement can be trusted, whether anything that works gets written down and repeated.

You can watch this at any scale. lululemon ran through four CEO successions while same-store sales sat flat for eight quarters. Each new leader, a new strategy. The pattern underneath, a board allergic to the actual fix, never moved.

New strategy decks are cheap. Operating systems are not. That price difference explains most of what's in your portfolio review.

Why performance agencies systematically fail PE-backed companies

It's not the people. Agency teams are full of smart, hardworking operators.

It's the model, and the model fails PE-backed companies in three specific ways.

First, the incentive mismatch. Agencies price on spend or scope. Sponsors buy outcomes. When the fee grows with the budget, every recommendation tilts toward more spend, more channels, more scope. Nobody at the agency is paid to find the $80K a month that shouldn't be spent at all. In a hold-period context where every dollar of EBITDA is worth 8 to 12 dollars at exit, that blind spot is expensive.

Second, the altitude mismatch. The agency sees the ad accounts. The sponsor sees the P&L. Between them sits a gap where the actual questions live: is growth coming at the expense of contribution margin? Is the channel mix building enterprise value or renting revenue? Adidas gave the industry its most honest confession on this. Their own global media director admitted publicly in 2019 that last-click attribution had pushed them to massively over-invest in performance channels, and that when they finally ran the econometrics, brand investment was driving the majority of sales. A multi-billion-dollar advertiser, steered wrong for years, by dashboards everyone knew were biased and nobody was accountable for questioning.

Third, the time mismatch. Agencies are built for continuity; their economics depend on the retainer renewing. A hold period is built for transformation on a clock. "Maintain and optimize" is rational behavior for the agency and a slow leak for the sponsor.

None of this means fire your agencies. Good ones are force multipliers, and the right answer is usually to keep them and manage them better. It means the missing layer is above the agency: someone inside the business who owns the number, sets the standard, translates between the ad account and the board, and makes the agency's output answer to the P&L.

Agencies execute. Someone still has to govern. That seat being empty is the single most common finding when you audit a portco's growth function, and the agency is the wrong party to blame for it.

The hold period is a forcing function. Use it.

Marketing academia says brand takes ten years. Sponsors have five, sometimes four by the time anyone looks at the growth engine seriously.

Heritage brands get to play a different game. J. Press has been making roughly the same suit since 1902 and doesn't have to grow to survive. Your portco does not have that option. It has an exit clock, and pretending otherwise is how value-creation plans die of romance.

The wrong conclusion is "PE can't build brand." The right conclusion is that the clock forces a sequencing discipline most marketing organizations never develop.

Year one: fix measurement and capture existing demand efficiently. Fund the engine with found money.

Years two and three: scale what's proven, build demand-creation muscle, take the unit economics up a weight class.

Final years: optimize for what a buyer will diligence. Durable revenue, diversified channels, a growth system that doesn't depend on any one person.

A clock concentrates the mind. Most marketing teams have never had one. The good news for the sponsor: imposing a clock is free.

The 100-day growth diagnostic every new platform deserves

Every new platform investment gets a finance audit in the first hundred days. Almost none get a real growth audit. Here's the one worth running, four workstreams, board-ready output.

Weeks 1-2, Measurement integrity. Can revenue be traced to its sources, or is everyone reading platform-attributed numbers at face value? Audit tracking, attribution windows, the gap between platform-reported and Shopify/CRM truth. Nothing else matters until this does, because every downstream decision inherits whatever lies the measurement tells.

Weeks 3-5, Unit economics. New customer CAC (not blended), contribution margin after marketing, payback period, cohort repeat behavior. Most portcos can produce revenue numbers on demand and not one of these four.

Weeks 6-8, Channel and concentration risk. What share of revenue depends on one channel, one algorithm, one keyword class? Branded search dependency and single-channel concentration are valuation discounts that show up in diligence. Better to find them as the owner than have them found by your buyer.

Weeks 9-12, Team and system. Who owns the growth number? What happens when the one person who knows the ad account leaves? Is anything documented, repeatable, transferable?

Output: a fact pattern on paper, a prioritized fix list sized against the P&L, and a 90-day execution plan with owners. The whole thing costs a rounding error against the deal fees and routinely finds the first two points of EBITDA.

Marketing diligence is still stuck in 2015

Buy-side diligence has gotten ruthless about everything except marketing. Quality of earnings gets a team of accountants. Quality of revenue, the machinery that produced those earnings and is supposed to keep producing them, gets a management presentation and a nod.

Here's what 2015-era marketing diligence checks: total revenue growth, blended CAC if you're lucky, a list of channels, the size of the email file.

Here's what actually predicts whether the growth survives the hold.

Attribution quality. If the target measures itself on platform-reported ROAS, the growth numbers are softer than they look. Platforms grade their own homework, and they are generous graders.

Channel concentration. Forty percent of revenue from branded search is not a growth engine. It's a tollbooth on demand that something else created, and the something else is usually decaying.

New versus returning mix. Revenue held flat while new customer acquisition declined means the company is harvesting its base. That's a melting ice cube presented as stability.

Incrementality evidence. Has anyone ever run a holdout? A company that has never tested incrementality does not know its own CAC. Neither does its buyer.

And the new one: ask the answer engines. Open ChatGPT, Claude, and Perplexity and ask what they say about the target's brand and category. An increasing share of discovery now routes through models that have already formed a view, assembled from the company's own copy, its reviews, its press, its community. A brand the models describe accurately and favorably has an asset no balance sheet captures. A brand the models can't describe at all has a marketing problem that hasn't hit the revenue line yet. Web Smith's work at 2PM on this point is worth every operating partner's time: the brand that controls the vocabulary the answer engines use for a category will own that category for years.

None of this requires a data room miracle. It requires asking for five things sellers rarely volunteer. The spread between what marketing diligence checks and what actually matters is one of the last cheap edges in mid-market PE, and it will not stay cheap long.

The cheapest EBITDA in the portfolio is inside the media budget

Buying EBITDA through ops improvements is hard. Renegotiate the supply chain, consolidate facilities, eighteen months of pain.

There is usually an easier first dollar, and the precedents are brutal.

Uber discovered that roughly $100 million of its app-install spend was doing nothing: they shut two-thirds of the budget off and installs didn't move. P&G cut $200 million of digital spend and saw no change in business outcomes, then said so out loud at an industry conference, to applause from nobody who sells ads.

These were sophisticated advertisers with measurement teams. Your $30M portco spending $5M a year on media, with no incrementality testing and platform dashboards as the source of truth, is not beating their base rate. Expect 15 to 25 percent of that budget to be non-working.

A holdout test costs a few weeks and some forgone retargeting spend. Finding $1M of non-working media inside a $5M budget is a straight EBITDA transfer at zero operational risk. At a 10x multiple, that's $10M of enterprise value for the price of one experiment nobody had asked for.

Audit the media before you audit the factory. The factory is harder.

What "we need better marketing" actually means

When an operating partner says a portco "needs better marketing," they are almost never talking about marketing.

Listen closely and it decodes to one of four sentences.

"I can't trust the forecast." Marketing committed to numbers that didn't land, twice, and now every projection from that team gets a mental haircut. The issue isn't creativity. It's credibility.

"I can't see where the money goes." Spend went up, revenue went up less, and nobody can explain the gap in a way that survives questioning. The board sees a black box with a budget line.

"I can't tell what would happen if we cut it." The classic test. Ask what happens to revenue if marketing spend drops 30%, and watch whether the answer is analysis or theology. Most teams answer with theology. The few who answer with holdout data run the meeting.

"I don't know if this scales." The current numbers might be fine. The question is whether the engine produces the back half of the value-creation plan or whether it's already at the flat part of its curve.

Notice that all four are finance questions wearing marketing costumes. They're about trust, traceability, causality, and capacity. A team can be excellent at campaigns and fail all four. Most do, because nobody ever told them these were the questions.

The fix is not more marketing. It's making the growth function answer the way the CFO's function answers: numbers that reconcile, forecasts with error bars and a track record, and a stated, tested position on what's incremental. Do that and the phrase "better marketing" quietly disappears from the board agenda.

The CAC payback test

A simple test for whether a company has a marketing executive or a brand manager with a big title: ask them to walk the board through CAC payback by cohort. No deck. No analyst feeding numbers.

The real ones do it from memory, because they think in that math all day. They know payback moved from month 4 to month 6, they know why, and they know what they're doing about it.

This isn't gatekeeping. It's the job. Marketing at a PE-backed company is capital allocation with a creative department attached. The executives who get treated as peers by the CFO are the ones who do the CFO's kind of homework.

Board-ready marketing reporting, the one-page standard

Most marketing board updates are activity reports: campaigns launched, impressions served, awards almost won. Boards want one page, and the page is this.

The number. Contribution margin after marketing, in dollars, against plan. One line. This is the headline; everything else is supporting detail.

Acquisition economics. New customer CAC (never blended), CAC payback, and LTV:CAC by cohort. Trend, not snapshot: three data points minimum so the board sees direction.

Efficiency. MER or blended ROAS, with the platform-versus-actual reconciliation shown. The act of showing the reconciliation is what builds trust. Boards forgive bad numbers and punish surprising ones.

Forecast versus actual. Last quarter's projection next to what happened, with variance explained in one sentence each. This line item, sustained for four quarters, is what turns a marketing leader from a cost center into a colleague.

Concentration. Share of revenue by channel, flagged when any single source exceeds a third. The board is underwriting risk. Give them the risk.

One decision. What you're changing next quarter and what it does to the number. Not five initiatives. One decision with a dollar attached.

That's the page. No platform screenshots, no funnel diagrams, no word "engagement." A marketing leader who ships this page every quarter, accurately, will get every budget they ever ask for. That's not idealism. It's how capital allocators respond to people who speak capital.

The forecast is the product

Marketing teams think their product is campaigns. From the board's chair, the product is the forecast.

Consider what the board actually consumes from the growth function. They don't see the ads. They see commitments: revenue will be X, CAC will hold at Y, the new channel reaches breakeven by Q3. Then they watch whether reality matches. That delta, forecast versus actual, sustained over quarters, is the entire reputation of the marketing organization. It determines whether budget requests get approved in five minutes or litigated for five weeks.

The cautionary tale is Peloton, which built a multi-billion-dollar valuation on a demand forecast that treated a pandemic as a permanent condition, then spent two years and several CEOs discovering it wasn't. The forecast error did more damage than any campaign ever could have. That's the asymmetry: good ads make quarters, bad forecasts break companies.

The teams that understand this operate differently. They forecast in ranges with stated assumptions, not point estimates pulled from hope. They build the contribution curve for each channel so "what happens at $200K more spend" has an actual answer with diminishing returns built in. They track their own forecast error and report it before being asked, which sounds like exposing weakness and is in fact the fastest credibility play available, because it signals the numbers are real.

And the compounding works both directions. Two missed forecasts and the haircut begins: the board discounts marketing's numbers, which shrinks budgets, which guarantees the next miss, which deepens the discount. Competent teams get defunded not because the marketing was bad but because the forecasting was, and the board had no way to tell the difference.

Run the engine well, obviously. But report it like a CFO would, because the audience is people who think like CFOs. The campaign wins the customer. The forecast wins the budget that funds the next ten thousand customers.

Channel concentration is a valuation discount

A brand doing $40M with 45% of revenue attributable to Meta is not a $40M brand. It's a $40M revenue stream with a single point of failure, and the buyer's diligence team will price it exactly that way.

Operators learned this lesson in other domains years ago: customer concentration gets flagged in every QoE, supplier concentration gets a risk memo. Demand concentration somehow escaped the checklist, even though an algorithm change can do to revenue what losing a top customer does. Faster. With no contract to renegotiate.

The Everlane post-mortem had a line in it that should hang in every DTC boardroom: the brand was not enough of a moat for products that could be replicated anywhere. Concentrated demand plus replicable product equals a comp you don't want. That one settled at $100M against a debt load that ate it.

Diversification costs efficiency in the short run. The CTV dollar will not perform like the retargeting dollar this quarter. But the spread between a one-channel multiple and a diversified-engine multiple pays for years of that inefficiency. You're not buying media. You're buying a better exit narrative.

The org chart matters more than the budget

Show me a portco's marketing org chart and I'll tell you more about next year's growth than the budget will.

The pattern that fails is everywhere: one overworked generalist VP, two coordinators, and seven vendors, each vendor owning a channel, no one owning the number. Money flows out monthly, reports flow in monthly, and the reports do not reconcile with each other because each vendor grades its own channel with its own attribution. The VP spends their week refereeing vendor claims instead of running a system. Headcount looks lean. Output is leaner.

What works in the mid-market is a small core that owns three things in-house, no exceptions: measurement, forecasting, and strategy. The truth has to live inside the company. Around that core, vendors and agencies execute, and they execute well precisely because someone inside can evaluate the work, set the standard, and kill what isn't incremental.

The hiring implication runs against instinct. The first senior hire is not a CMO, and it is certainly not another channel manager. It's the analytical operator who builds the measurement layer and the forecast, because that person makes every other dollar smarter, including the agency dollars already going out the door. Strategy without measurement is opinion, and the org chart should be built in the order that turns opinion into math.

Budgets are easy to approve and easy to waste. Structure is what decides which one happens.

Over-tooled, under-instrumented

The standard mid-market martech audit finds 23 tools and no answer to "what was new customer CAC last month."

There's a CDP nobody finished implementing, an attribution platform nobody believes, two analytics suites that disagree with each other, and a BI dashboard built by someone who left. Annual cost: mid six figures. Questions it can answer reliably: roughly zero.

The instinct is to buy tool 24 to fix it. The fix is the opposite direction. Instrumentation before tooling: clean conversion tracking, one source of revenue truth, new-versus-returning customers separated, channels tagged consistently. That's not software, that's plumbing, and plumbing is unglamorous right up until you need water.

Tools are purchased. Instrumentation is built. Companies that confuse the two end up with the 23 tools, and the question still open.

Triangulation, or how to stop arguing about attribution

Every attribution method is a liar. The discipline is in knowing how each one lies, and making them check each other.

Multi-touch attribution lies by overcounting the bottom of the funnel. It can only see clicks, so it hands credit to the last things clicked: branded search, retargeting, email. The channels that created the demand upstream, the CTV spot, the creator video watched on a lock screen, are invisible to it. MTA is a security camera pointed at the cash register, swearing the cashier is responsible for all the sales.

The canonical demonstration is a decade old and still under-read. eBay's economists shut off paid search in a controlled experiment and found that the bulk of it, especially brand keywords, was paying for clicks from people who would have arrived anyway. Tens of millions of dollars, annually, buying traffic the company already owned. The dashboards had called that spend efficient for years. The experiment called it what it was.

Marketing mix modeling lies in the other direction. It sees everything but slowly and fuzzily: aggregate spend in, aggregate revenue out, correlations estimated over years. It will tell you TV "works" with a confidence interval wide enough to drive a media plan through, and it can be tuned, consciously or not, to say what its sponsor hoped.

Incrementality testing tells the truth, but only about the exact thing tested, in the exact period tested. A geo-holdout proves what Meta prospecting did in March. It says nothing about September, or about the next channel over.

So the answer isn't picking a winner. Single-method measurement is single-point-of-failure measurement. The answer is triangulation: run all three, let each correct the others' known biases, and treat disagreement between them as information rather than annoyance. The modern measurement stack, and credit where due, the Triple Whale and Common Thread Collective generation of operators has dragged this thinking from enterprise theory into mid-market practice, works roughly like this: MTA for daily steering, because it's fast and its bias is at least consistent. MMM for quarterly allocation, because only it sees the whole system. Incrementality tests as the periodic audit that recalibrates both, because only experiments establish causality.

In practice the triangulated answer is unglamorous: platform-reported ROAS is usually 1.5 to 3x too generous, retargeting is far less incremental than it looks, upper-funnel is more. Which is precisely why the triangulation matters; every bias in the naive setup flatters the same direction, toward spending more on what's easiest to measure.

One more thing, because it's the question CFOs ask: this is not an enterprise luxury. A $20M brand can run geo-holdouts with free tooling and a competent analyst. A lightweight MMM is a few weeks of work, not a seven-figure engagement. The constraint was never budget. It was the willingness to find out that some of the budget never needed to be spent.

Stop litigating which dashboard is "right." They're all wrong, usefully, in different directions. The job is to sit at the intersection.

A receipt written by the salesman

Platform-reported ROAS is a sales receipt written by the salesman.

Meta reports what Meta touched. Google reports what Google touched. Add up the platform dashboards and most brands "earn" 130 to 180 percent of their actual revenue. Every platform is telling a version of the truth in which it is the hero.

This is not fraud, it's structure. Each platform attributes within its own window, on its own logic, with no knowledge of the others, and every incentive to be generous to itself. Expecting the ad platform to referee its own incrementality is expecting the casino to audit its own odds.

The dashboards are useful for one thing: relative, within-platform decisions. Which creative, which audience, which campaign. The moment the question becomes "how much did this channel really add," the dashboard stops being evidence and starts being marketing.

MMM is back, but not for the reason you think

Marketing mix modeling spent two decades as enterprise furniture: million-dollar engagements, twelve-week refreshes, a consultancy priesthood interpreting the output. Then privacy broke click-tracking, and suddenly MMM is on every mid-market roadmap.

The revival is real but widely misread. The case for MMM was never that it's precise. It isn't; the confidence intervals are honest about that. The case is that it's the only method that sees the whole system at once: every channel including the unclickable ones, price, promotion, seasonality, competitors, all in one model with revenue on the left side of the equation. It's the only tool that can even attempt the question boards actually ask, which is "what is the shape of the whole portfolio, and where is the next dollar best spent?"

Remember what happened to the advertisers who didn't have that view. Adidas ran years of budget allocation on last-click dashboards and later admitted the dashboards had pushed them to drastically over-fund performance channels while the econometrics said brand was doing the heavy lifting. The cost of not having a whole-system model wasn't measurement vanity. It was years of misallocated nine-figure budgets.

What changed is the cost of asking. Open-source frameworks, modern data infrastructure, and a generation of tooling aimed squarely at the mid-market collapsed the price from seven figures to a competent analyst and a few weeks. A $30M brand can now afford the worldview that used to require a Fortune 500 budget.

What didn't change is the failure mode. An MMM built on bad spend data, too few observations, or wishful priors will produce confident nonsense, and it will be believed precisely because it looks like science. The model is also purely correlational; it proposes, and only experiments dispose. The brands doing this well treat the MMM as a hypothesis generator and run incrementality tests on its biggest claims before moving real money.

Used that way, with humility and a testing budget, MMM is the closest thing marketing has to a balance sheet: imprecise line by line, indispensable in aggregate. Used as an oracle, it's astrology with better fonts.

An incrementality calendar for a $5-50M brand

Incrementality testing fails in most organizations not from complexity but from formlessness: everyone agrees it matters, no one schedules it. The fix is a calendar. Here's the one that works at mid-market scale.

Quarter 1: test the biggest line item. Geo-holdout on your largest prospecting channel, usually Meta. Match markets, hold out 20-30% of geography, run four to six weeks. This single test typically re-prices 40% of the budget and pays for the entire year's program.

Quarter 2: test the most suspicious line item. Retargeting and branded search, the channels MTA loves most. Audience holdout or geo split. This is where eBay found its tens of millions of non-incremental brand-search spend, and where Uber found the $100 million of app-install budget it could shut off without losing a single install. Brace yourself. This is where the found money lives, and it is rarely small.

Quarter 3: test the thing you want to scale. Before CTV or another upper-funnel bet gets real budget, structure the launch itself as an experiment: matched-market rollout, not national splatter. You get the growth attempt and the proof in the same spend.

Quarter 4: re-test the winner. Incrementality decays. The Meta result from Q1 is not permanently true; auctions change, creative fatigues, baselines move. Annual recalibration on the biggest channel keeps the whole measurement stack honest.

Rules of the road: one variable per test, pre-registered success criteria so nobody re-litigates after the fact, and a standing agreement with finance on what happens to freed-up budget, otherwise the savings evaporate into "more of the same."

Four tests a year, each a few weeks long. That's the entire program, and it puts a brand ahead of 90% of its competitors on the only measurement question that matters: what would have happened anyway?

New customer CAC is the only CAC

Blended CAC is the most popular metric in e-commerce and the most reliably misleading. It divides total spend by total customers, new and returning alike, and in doing so it lets returning customers subsidize bad acquisition math.

Here's the mechanism. A brand with a strong repeat base can watch blended CAC hold steady for a year while new customer CAC quietly doubles. The repeat purchases, earned by past acquisition and product quality, keep flowing into the denominator and flattering the average. The dashboard says stable. The engine says decay. By the time blended CAC finally moves, the new-customer problem is six months old and the cheap fixes are gone.

I've watched this from inside more than once. Revenue flat, board calm, blended metrics serene, and underneath it the new customer count down 30% year over year. That's not stability. That's a brand harvesting its own base and calling it efficiency. Buyers' diligence teams have learned to spot it, which means the metric that flattered the operating review becomes the finding that cuts the multiple.

The discipline is to split everything. New customer CAC against first-order contribution margin tells you whether acquisition stands on its own feet or borrows against the future. Cohort repeat rates tell you whether the future will actually pay. Blended MER stays useful as a cash-efficiency gauge, but it's a thermostat reading, not a diagnosis.

The infrastructure ask is modest: tag new versus returning at the conversion level, push it into the ad platforms and the reporting. A week of tracking work. The alternative is running a company on an average of two populations moving in opposite directions, which is how brands get surprised by problems their own data had been reporting for two quarters, just in a column nobody graphed.

Your MER went up. That might be bad news.

MER improving is the metric equivalent of good weather: pleasant, and not necessarily informative.

Marketing efficiency ratio is revenue over spend, all-in. It goes up when media gets genuinely better. It also goes up when you underspend against available demand, when organic and email carry more of the load, when a strong product month flatters the denominator. Two of those four are problems wearing a green number.

Real example from a recent mandate: MER hit its best month in a year, 7.9x, and the honest read was negative. Revenue grew because the brand was coasting on equity and retention while a broken campaign quietly cut paid volume. Efficiency was up because investment was accidentally down, and the growth account underneath was thinning.

A ratio can't tell you which of its two numbers moved. Always decompose: spend, revenue, and the new-customer count behind the revenue. MER is the smile on the face. You still have to take the pulse.

Forecast media like a CFO forecasts cash

A CFO would be fired for forecasting cash the way most teams forecast media. "We'll spend $300K and revenue should be around $2M" is not a forecast. It's a mood.

The CFO's discipline translates almost one to one. Cash forecasting works because it's built from drivers, stated assumptions, scenario ranges, and a feedback loop on its own error. Media forecasting works the same way or doesn't work at all.

Drivers first. Revenue from paid is spend times efficiency, and efficiency is a curve, not a constant. Every channel has a contribution curve with a steep part, a productive middle, and a flat expensive tail. The single most valuable analytical artifact a growth team can build is that curve per channel, estimated from spend variation history and checked against incrementality tests. With curves in hand, "what happens at $200K more" stops being a debate and becomes arithmetic with error bars.

Scenarios second. One number is always wrong. Three numbers, a base case on current trajectory, an upside with assumptions named, a downside with triggers named, give the board something it can actually plan against, because that's the shape capital allocators already think in.

Error tracking last, and this is the one almost nobody does. Log every forecast, compare against actuals monthly, publish your own miss rate. Forecast error that's measured shrinks; the act of scoring it changes the inputs people are willing to write down. Within a few quarters the range narrows and something more valuable happens: the board starts treating marketing's numbers like finance's numbers.

That's the actual prize. The forecast isn't paperwork that follows strategy. In a capital-allocation business, the forecast is how strategy gets funded.

Measurement debt compounds like technical debt

Engineers know the pattern: skip the refactor, ship the feature, pay interest forever. Marketing has the identical disease with worse accounting.

Broken conversion tracking, attribution windows set once in 2021, UTM chaos, platforms double-counting each other. Each defect feels small. But every optimization, every budget shift, every creative readout, every forecast inherits the errors. Bad measurement doesn't sit still; it compounds, because decisions made on wrong numbers create new wrong numbers, and the algorithms training on corrupted conversion data automate the error at scale.

The interest payments hide in plain sight: a bidding algorithm fed phantom conversions, a "winning" creative that won a tracking artifact, a channel cut because its credit was assigned elsewhere.

The payoff structure is also identical to tech debt: paying it down is boring, unglamorous, and the highest-ROI work available. Two weeks of tracking remediation routinely outperforms a quarter of optimization built on top of the debt. Fix the books before reading them.

Ads don't 5x companies. Systems do.

Somewhere in the last decade, an entire industry convinced itself, and its clients, that the path from $10M to $50M runs through an ad account.

It's a comfortable belief. It's also arithmetic illiterate.

Run the math on what 5x actually requires. Take a $10M brand spending $2M on media at a 5x MER. To reach $50M on ads alone, with the diminishing returns every channel exhibits, you'd need spend to grow 8 to 10x while efficiency politely declined only modestly. No brand's contribution curve survives that. The flat tail arrives long before the finish line, and the brand drowns in its own CAC somewhere around year two.

The auction itself enforces this. Common Thread Collective's team has a framing I think about often: at $500 a day you're bidding against small businesses that don't know their own unit economics, and at $5,000 a day you're bidding against operators with better margins, better offers, and better conversion rates, some of whom can profitably pay more for a customer than your entire average order value. They tell the story of a jewelry brand that wanted more creative volume to scale, and the industry data showed competitors' average CAC was higher than the brand's AOV. No quantity of ads closes that gap. The auction isn't a media problem at that point. It's a business model problem with a media bill.

CTC's Joy Sharma has a line for the general case: solving a business problem with a marketing solution is where businesses go to die.

Now look at what actually moves when companies 5x. Offer and price architecture: AOV up 25% reprices every channel simultaneously, paid and organic alike. Conversion rate: a point of CVR improvement reprices every click you'll ever buy. Repeat economics: LTV expansion is what lets you outbid competitors for the same customer and still make money. Distribution breadth: retail, marketplaces, partnerships, channels that compound instead of saturating. Product velocity: the spring line that nearly doubled new customer counts in one of my current mandates did more than any campaign optimization that quarter. The media amplified it.

And the creative data confirms the hierarchy. Across CTC's client base, brands without offer-market fit see creative hit rates around 2%, and brands with it see 7%. Same talent, same formats, same platform. When the offer is right, a flat product shot beats the hundred videos that failed before it. Creative isn't the input that was broken.

The two cleanest case studies of the decade say the same thing from opposite directions. Vuori took General Atlantic capital at $5.5 billion on roughly $400 million of revenue: founder present, wholesale book deep, unit economics that survived diligence, ads as an amplifier on a working system. Everlane, with a decade of brand heat and as much performance-marketing talent as money could rent, sold to Shein for $100 million that its debt mostly consumed. The difference was never who had better ads.

Ads have a precise and important job: they are the amplifier and the accelerant. They take a working machine and feed it faster. What they cannot do is be the machine. An amplifier with nothing upstream just makes the silence louder and more expensive.

This is also why "performance agency promises 5x" engagements end the way they end. The agency controls one input, media, and gets judged on an output, growth, that five other inputs determine. When growth stalls, the agency optimizes harder on its one lever, efficiency erodes, trust follows, and eighteen months later there's a new agency making the same promise. The carousel isn't a talent problem. It's a scope problem: nobody was ever hired to fix the system, so the system stayed broken while the amplifier got replaced annually.

First principles, then. Before any conversation about scaling spend: is the offer right, is the margin structure sound, does the conversion path work, do customers come back, does the measurement tell the truth? Each yes makes every media dollar more powerful. Each no makes media spend a tax on hope.

Ads are important. I've spent fifteen years buying them and still do. That's exactly how I know they're the last 30% of the answer, and why the first 70% is where every engagement worth taking begins.

The edge moved upstream

Being excellent at Google and Meta in 2026 is being excellent at email in 2010: required, and worth nothing at the margin. The tooling matured, the automation absorbed the tricks, and everyone competent converged on the same playbook. Tablestakes.

The differentiation moved upstream, into the channels performance marketers avoid because measurement is hard: CTV, programmatic, linear, OOH, audio.

The arbitrage is structural. Performance buyers won't touch what they can't track click-through. Traditional buyers can't hold what they buy to performance standards. The operator who can plan reach media like a brand strategist and measure it like a performance buyer sits alone in the intersection, buying attention at prices the competition has rationalized away.

Saturated channels reward incumbency. Hard-to-measure channels reward capability. Pick your scarcity.

Buy CTV like a performance marketer, plan it like a brand strategist

CTV is where media budgets go to be misunderstood. Brand teams buy it like television, reach and frequency and a prayer. Performance teams either avoid it entirely or buy it like a giant Facebook campaign and quit when the last-click numbers disappoint. Both are wasting the most interesting channel of the decade.

The honest frame: CTV is a demand-creation channel with performance-grade accountability available to anyone willing to do the work. That means two disciplines at once.

Plan it like a brand strategist. CTV's job is to make people want the thing before they search for the thing. That means planning against audiences and household penetration, not CPMs alone. It means creative built for a living-room screen and a ten-foot viewing distance, not a repurposed vertical video. It means committing to flight lengths long enough for frequency to work, eight weeks minimum, because demand creation is cumulative and a two-week test of a brand channel is a test of nothing.

Buy it like a performance marketer. Programmatic pipes mean you control targeting, frequency caps, and dayparting at a granularity linear never offered. Measure with the tools that actually fit the channel: matched-market holdouts on the rollout, branded search and direct traffic lift in exposed geos, halo effects on paid search efficiency, post-exposure conversion through household graphs where available. None of this is exotic anymore; it's a competent analyst and a willingness to define success before launch.

The brands getting CTV right report the same pattern: weak last-click numbers, strong geo-lift numbers, and a paid search account that gets measurably cheaper in exposed markets because demand arrives pre-warmed. That last effect is the entire point. Upper funnel done right doesn't compete with performance channels. It subsidizes them.

The algorithm changed again. Good.

Every quarter, the same ritual: platform ships an update, ad accounts wobble, panic threads bloom. And every quarter, a quiet transfer of market share to the operators who treat volatility as the business model rather than an interruption to it.

Stable systems get arbitraged to zero. When nothing changes, every advantage gets copied and competed away, and the spoils go to whoever has the most budget. Volatility resets the table. Each shift briefly misprices something: an ad format, an audience, a bidding approach. The operators in the accounts daily find the mispricing first. The ones managing from a monthly dashboard read about it later, in a case study, written by the people who took their share.

This is why media buying is a practiced craft rather than a solved problem, and why "we set it up and it runs itself" should worry you in any pitch. The machine runs itself the way a portfolio runs itself: fine on average, expensively at the tails.

The algorithm changing is the moat. It just only moats the people still doing the work.

Creative is the targeting now

For a decade, media buying was a targeting craft. Audiences, lookalikes, interest stacks, exclusion architecture: the skill was telling the machine who to find. Privacy changes and platform automation ended that era with almost no funeral.

What most organizations haven't internalized is how literal the replacement is. Meta's delivery system is now a stack of interlocking AI models, the GEM, Lattice, Andromeda, UTIS generation of infrastructure, that decides who sees an ad based on how people respond to the ad itself. The measurement community, Triple Whale's research team among the clearest on this, has documented the consequence: delivery flipped from audience-led to creative-led, and the old trick of running fifty micro-variants of one concept now collapses into a single signal. The machine wants structurally different concepts, not the same concept in fifty hats.

So each ad functions as its own audience definition. A video about fit and sizing finds the customer worried about fit. A price-anchored carousel finds the deal seeker. Ship five genuinely different concepts and you've shipped five targeting strategies; the machine sorts the rest. The media buyer's job didn't disappear, it moved: from picking audiences to engineering the portfolio of messages that lets the machine pick well.

That reframe has operational teeth. Creative volume and variance become media decisions, not brand-team afterthoughts; an account with two concepts running is an account targeting two segments, no matter what the settings claim. Testing structure becomes the strategy: clean concept-level reads, ruthless kill criteria, and a production pipeline that treats winning angles as something to scale, not celebrate. And the analytics layer has to grade creative on contribution, not clicks, because the algorithm will happily find a million people who watch and never buy.

The practical evidence shows up fast in any account run this way. In one current mandate, pushing a proven video concept from $10 a day to $135 a day improved ROAS while scaling. That's not a paradox. The message matched a market, and the machine, given a winner, knew exactly where to take it.

Targeting didn't die. It got promoted into the creative department. Staff accordingly.

The saturation audit: knowing when a channel is done giving

Every channel has a curve, and every team eventually spends past the bend while the dashboard insists things are fine. Averages hide the bend: ROAS is the average of all spend, and the average stays decent long after the marginal dollar went bad. Here's the audit that finds the bend before the P&L does.

Check marginal, not average, efficiency. What did the last 20% of spend return, versus the first 80%? Platform data can approximate this through spend-tier analysis or simple before/after reads on budget changes. A channel at 4x average and 1.5x marginal is a channel telling you it's full, politely.

Read the auction signals. CPMs rising faster than the market, frequency creeping past 3-4 on prospecting, audience overlap expanding, CPC inflation on stable keywords. The platform charges more when it's running out of the people you want. Price is information.

Know who you're bidding against now. This is the one almost everyone misses: scaling spend changes your competitive set. The brands at your target spend level have different margins, different AOVs, different offers. Pull industry AOV-to-CAC benchmarks and check whether your unit economics can even compete at the next tier. If competitors can pay more per customer than you charge per order, the channel isn't saturated. Your offer is.

Test the ceiling directly. Push spend 30% for three weeks in a matched-geo split. If revenue follows at acceptable efficiency, the curve has room. If spend absorbs and revenue shrugs, you've met the ceiling and it cost you three weeks instead of three quarters.

The decision rule: marginal efficiency below your contribution-margin breakeven means the next dollar belongs somewhere else. A new channel, creative variance, offer work, or back in the EBITDA line. Channels aren't loyal and budgets shouldn't be either.

Brand harvesting looks like found money for exactly two quarters

There's a name for the most reliable trick in the quarterly-pressure playbook, and Chip Wilson just spent a proxy fight teaching it to everyone: brand harvesting.

Cut the upper-funnel spend. Add the promotional credit-card discounts. License the mass-market collab. Lean on markdowns and outlet volume. Every one of these moves works immediately: performance metrics hold, revenue holds, the savings and promo volume drop straight into the quarter. Whoever made the call looks smart.

For about two quarters. Demand creation works on a lag; the equity built last year is still converting this year. Harvest it and you're draining a reservoir you've stopped refilling. lululemon ran this play long enough that the bill came due in public: eight quarters of flat comps, markdown levels analysts called damaging to premium positioning, and a $17 billion hole that an activist founder is now litigating in front of the whole market.

The insidious part is attribution. The costs land later, on someone else's metrics, far from the decision that caused them. By the time "efficiency is slipping" reaches the board deck, the answer is six months old.

Some cuts are right. Make them with the lag priced in, and with geo-level evidence of what the upper funnel was actually carrying. Harvest knowingly or don't harvest at all.

Consumers buy in moments. B2B buys in committees. Stop running the same funnel math.

The most expensive copy-paste in marketing is taking the DTC playbook into B2B, or the reverse, and assuming the math transfers. It doesn't, because the buyer physics are different in kind.

Consumer purchase is individual, fast, and emotional with rational garnish. One person sees the thing, wants the thing, and the window between desire and decision is measured in minutes or days. This is why creative velocity, offer structure, and friction removal dominate consumer outcomes, and why consumer attribution, for all its flaws, at least chases a traceable event: there was a moment of decision, recently, near a click.

B2B purchase is collective, slow, and political. The "buyer" is six to ten people with different incentives: the champion who wants the problem solved, the CFO who wants the spend justified, the IT lead who wants no risk, the end users who want no change. The cycle runs months to years. And the decisive marketing moments mostly happen where no pixel fires: the buying committee meeting, the Slack message asking "anyone used these guys?", the shortlist formed from memory before any vendor knew the deal existed.

That last point carries the strategic weight. By the time a B2B buyer fills out a form, the shortlist is usually set and the form is confirmation, not discovery. Which means B2B marketing's real job is being remembered into the shortlist: brand, in the unfashionable sense, built through consistent presence long before any intent signal exists. Measuring it on lead-gen math, cost per MQL and last-touch credit, systematically defunds the activity that fills next quarter's pipeline in favor of harvesting this quarter's.

The funnel math follows the physics. Consumer: optimize the moment, measure in days, scale creative. B2B: optimize the memory, measure in quarters, scale presence and proof. Two different machines. Companies that run both, and plenty of PE portfolios do, need two different dashboards, two different patience settings, and a leadership team that knows which game each dollar is playing.

95% of your B2B market isn't in-market. Spend like it.

The Ehrenberg-Bass number that should be taped to every B2B budget meeting: at any given moment, roughly 95% of your addressable buyers are not buying. Not this quarter, often not this year.

Now look at the standard B2B budget: nearly all of it aimed at the 5%. Intent data, bottom-funnel search, retargeting, SDR sequences against "hot" accounts. An entire industry competing in an auction for the same sliver, bidding the CAC up annually, while the 95% who will buy eventually hear nothing from anyone.

The brands that own a category do the opposite math. They maintain cheap, consistent presence against the 95%, so when a buyer finally enters the market, the shortlist is already written and they're on it. The 5% spend then converts at multiples of category average, and everyone credits the bottom-funnel tactics for what the patient money did.

Capture demand efficiently, yes. But demand capture is a tollbooth. Somebody upstream has to build the traffic.

The DTC playbook is dead. The DTC math isn't.

Read the obituaries in order. Bonobos: bought by Walmart for $310 million, sold six years later for $75 million. Casper: IPO'd, taken private at a fraction, sold again. Allbirds: from $4 billion to delisting-watch. And now Everlane, the brand that defined millennial transparency, sold to Shein, the precise opposite of everything it claimed to stand for, for $100 million that its debt mostly consumed. Common shareholders, the employees who took equity instead of salary, got zero.

The autopsies keep blaming "DTC." Wrong organ.

What died was a media arbitrage. Cheap Meta CPMs, venture subsidies for growth at any margin, and a press cycle that treated a Shopify store and a sans-serif logo as a business model. The brands built purely on that arbitrage were renting their growth, and when the rent went up, the model went with it. As the Everlane post-mortem put it, the brand was never enough of a moat for products that could be made anywhere.

But look at what survived, and what it kept. Vuori sold a majority stake to General Atlantic at $5.5 billion on roughly $400 million in revenue, with a founder still answering Instagram comments, a wholesale book deep enough to absorb DTC volatility, and unit economics that held up in diligence. The heritage houses, the J. Press and Drake's tier that Web Smith keeps pointing the cohort toward, never took the arbitrage at all: slower cadence, owned customer relationships, no exit clock, compounding quietly for decades.

What the DTC era actually contributed, and what survives it, is a body of math every consumer business should run regardless of channel: contribution margin per order as the unit of truth, cohort-based LTV instead of averaged wishes, payback windows set by cash reality rather than faith, incrementality over platform-reported applause. The Common Thread Collective school of operators has pushed this into something like financial planning for demand: forecasting revenue from spend curves, managing the P&L at the cohort level, treating media as capital deployment with expected returns and error bars.

That discipline is channel-agnostic. Direct when direct wins, wholesale when wholesale wins, and the spreadsheet doesn't have a favorite.

DTC was never the strategy. It was a distribution choice that happened to come with unusually good data. The playbook was a moment. The math is the inheritance. Take the inheritance and skip the funeral.

AOV is a strategy, not an outcome

Most teams treat average order value like weather: report it, mention it moved, move on. Run the sensitivity and that indifference looks expensive.

At typical DTC margins, a 15% AOV lift does more for contribution than a 15% CAC reduction, and unlike the CAC fight, it isn't waged against an auction full of competitors. It's waged against your own offer architecture, where you're the only bidder.

There's a second-order effect most teams miss, and CTC's data makes it concrete: AOV and conversion rate sit on a curve. A $40 product needs something like a 6% conversion rate to compete; a $1,000 product can be perfectly healthy at 1%. Your offer's position on that curve, relative to your category, determines what CAC you can afford, which determines what spend levels you can survive. AOV isn't a reporting line. It's auction admission price.

The levers are unglamorous and proven: bundles built from what customers already co-buy, thresholds set just above current AOV, post-purchase offers riding on a transaction already won, a premium anchor that makes the core product feel reasonable. A product launch that pulls AOV up 14% reprices every click in the account that month. Every channel at once. No platform's permission required.

The auction sets your CPMs. You set your offer.

Demand capture vs. demand creation: most budgets are 90/10 when the business needs 60/40

Audit a typical mid-market budget and the split is unmistakable: 90% capture, 10% creation, if creation gets funded at all. Branded search, shopping, retargeting, intent keywords, the spend that shows up where demand already exists and collects it. It's the rational result of measurement bias. Capture channels produce beautiful attribution. Creation channels produce awkward silences in the ROAS review.

The problem is arithmetic, not philosophy. Capture spend is bounded by the demand that exists. There are only so many people searching for the category, only so many warm visitors to retarget. Past that ceiling, incremental capture budget doesn't buy growth; it buys higher CPCs on the same demand and, increasingly, credit for conversions that were coming anyway. eBay proved the extreme case years ago: shut off brand search experimentally and the traffic mostly showed up anyway.

The research base here is older and sturdier than most of the industry admits. Binet and Field's work across hundreds of campaign databases landed on roughly a 60/40 split between long-term brand building and short-term activation as the profile that maximizes growth. The exact ratio shifts by category, share position, and margin structure. The direction of the standard error never does: almost every account I've ever opened was over-captured and under-created, harvesting aggressively from a pool nobody was refilling, with a dashboard that called the harvesting "performance."

And when someone finally rebalances, the results have a way of being public. Airbnb cut performance marketing dramatically in 2021, shifted to brand, and told investors traffic held so well they never went back. P&G found $200 million of digital spend it could cut with no business impact, then reinvested where reach was real. Neither company concluded ads don't work. Both concluded that paying full price to interrupt people who were already coming is a strange use of shareholder money.

Creation fills the reservoir. Capture drains it efficiently. The measurement layer, geo-lift on branded search, halo on paid efficiency, cohort source analysis, audits the flow between them. Brands that only capture compete on bid strategy. Brands that create get searched for by name, which is the cheapest click in the auction and the closest thing media buying has to a dividend.

AI won't replace your marketing team. It will expose which half was already replaceable.

The honest version of the AI-and-marketing conversation is uncomfortable and clarifying.

A large share of marketing work was always assembly: pulling numbers into decks, reformatting reports, writing variations, summarizing calls, moving data between tools that don't talk. AI does assembly now, at near-zero marginal cost, a day's work in an hour. Pretending otherwise is a payroll decision, not an opinion.

What AI cannot do is the part that was always the actual job: deciding what the numbers mean, what the brand should say, which risk to take, what not to do. Judgment, taste, and accountability don't compress. They get more valuable as everything around them gets cheaper, the way a good editor matters more when everyone can write.

There's a second front opening at the same time: your brand is now being described to your customers by models you don't control. Ask ChatGPT or Perplexity about your category and an answer comes back, assembled from your copy, your reviews, your community, your press. The brands winning the next cycle write every product page and founder note as if a language model will ingest it and repeat it to a future customer. Because it will.

So the org math changes twice. Teams shrink in the middle, leverage concentrates at the top, and the editorial function, the people who decide what the company says, quietly becomes a revenue function. For a portfolio company, that's not a threat. It's the cheapest capability upgrade on the board: same headcount, twice the throughput, if someone teaches the team to work this way. Most won't. That's the opening.