Notes · The numbers · July 9, 2026

Forecasting, explained for founders.

Most brand forecasts are a wish typed into a spreadsheet. Someone picks a revenue number that would feel good, spreads it across twelve months, and calls it a plan. Then the year happens, the number doesn't, and everyone acts surprised.

A forecast isn't a wish. It's a model of what's likely, built from what your business actually does, that tells you what has to be true for the target to happen. Get it right and it becomes the most useful document in the company. Get it wrong and you spend the year explaining variances to a fantasy.

The best public thinking on this comes from Taylor Holiday and the team at Common Thread Collective, who forecast for hundreds of DTC brands and hold themselves to landing within a few percent of target. What follows is that school of forecasting as I run it, folded into the model I build for every brand I work with. Three paths: an existing brand with trading history, a brand launching cold, and an expansion into a new market. The principles are the same. The inputs are not.

Three principles before any math.

1. Marketing owns the forecast, not finance. This is the contrarian one, and it's right. Finance can't change the actions that create revenue, so when the number slips, all finance can do is edit the spreadsheet. You build the forecast through the marketing calendar: you tell the model what you're going to do, and it tells you what you can expect. Revenue doesn't show up in the spreadsheet. It shows up because you made it show up.

2. Steer on contribution margin, not revenue. Revenue is vanity, contribution margin is sanity. It should be tattooed somewhere. A forecast that hits the revenue target by overspending on acquisition isn't a plan, it's a way to lose money on schedule. Every forecast gets read from the margin line down.

3. Likely before aspirational. Don't start with what you'd like to happen. Start with what's likely to happen, based on what the business has actually done, then build the bridge from likely to wanted. The gap between the two is your real to-do list: the launches, the creative, the retention work that has to close it. If nothing on the calendar closes the gap, the target is fiction and the forecast just told you so. That's it working.

The machinery all three paths share.

New revenue and returning revenue are different animals. Forecast them separately. This is the structural heart of the whole method. Returning revenue is the predictable part: it's a lagging output of everything you've already done, and it decays on a curve you can measure from your own cohorts. New revenue is the volatile part: it's a function of what you spend and how efficiently it converts. Blend them into one line and you can't see which engine is broken when you miss. Split them and the forecast diagnoses itself.

You cannot forecast what you refuse to measure. Years ago I pulled returning revenue out of the blend and started reporting what a new customer actually cost to acquire, and what those new customers returned per dollar of ad spend. I was told, in writing, that the lifetime value metric was flawed, because "LTV is included in daily MER." It wasn't an argument about the math. Nobody there had seen the two engines apart, so a number with returning customers taken out of it looked broken to them. A big business, and not one person in it could tell me what a new customer cost. There was no forecast in that building, only an average that felt good. I wrote about it in the second P&L.

New revenue = spend × efficiency. The efficiency number that matters is aMER: new-customer revenue divided by the acquisition spend behind it. Not platform ROAS, which we've covered elsewhere (the platforms grade their own homework). Your history tells you what aMER you actually run at a given spend level, by month, because it moves with seasonality.

Efficiency falls as spend rises. Build that in. The single most common forecasting lie is straight-lining today's efficiency across double the spend. Your first dollars hit your warmest audiences and your best creative; the next ones reach colder. So the question isn't "what's our aMER," it's "what's our aMER on the next dollar", the marginal read. Work out the extra revenue each additional slice of spend actually generates, because a brand can look fine on the blended number while the last chunk of spend is quietly losing money. One published example from the Common Thread team: an apparel brand spending over $2M a month at a blended efficiency that looked survivable, where the marginal math showed the last half-million of monthly spend was generating negative contribution. The blended number never shows you that. The marginal one does.

Know your break-even before you argue about targets. This takes one input: contribution margin before ad spend. If your margin before marketing is 50%, revenue has to come in at 2x spend just to break even, so break-even aMER is 2.0. At 70% margin it's about 1.4. Everything above break-even is contribution; everything below it is paying customers to take the product. This math needs zero trading history, which is why it turns up again in the launch path below.

The sanity frame: four quarters. Restructure the P&L into four buckets as a share of revenue: acquisition spend, cost of delivery (COGS, shipping, fulfillment, fees), operating expenses, and profit. The aspirational shape is 25% each. Nobody sits exactly there, and the best-run DTC P&Ls actually push acquisition below 20% of sales and let profit take the difference. The frame's real job is catching distress early: if delivery is eating 40% and acquisition another 40%, no forecast fixes that, the unit economics have to change first.

One model, three numbers. The newest piece of the method, and a genuinely useful one: don't publish one forecast, publish three off the same model. A conservative one for the board (under-promise), the likely one for cash planning (the 50/50 case, this is the real forecast), and a stretch one for the team's bonus targets. Same machine, three read-outs, and everyone stops arguing about whose number is "the" number.

Path one: the existing brand.

You have history. The job is to make the history confess, then decide what you'll do differently.

Step 1: Pull 12 to 24 months of actuals. Monthly net sales split new vs returning, ad spend, the full delivery-cost stack (not just COGS: shipping, fulfillment, payment fees), AOV, new-customer counts, repeat rate, and cohort revenue curves. If the data's a mess, fix the data first. A forecast built on dirty actuals is dirty at birth.

Step 2: Build the seasonality index. Each month's share of the year, from your own history. Most consumer brands live somewhere around 1.3 to 1.5x in November and 0.8 in the winter trough. This is what stops the forecast being twelve identical months, which no real brand has ever had.

Step 3: Forecast returning revenue first. It's the predictable engine. Take your cohorts, watch how each month's customers actually repeat over time, and roll the existing customer file forward. No ambition allowed in this line: it's arithmetic on people you've already acquired. If the business needs this line to be bigger, that's a retention project on the calendar, not a bigger number in the cell.

Step 4: Forecast new revenue from the spend curve. Set spend month by month, apply the aMER your history says that spend level and season actually delivers, and let the model give you new revenue. Then apply the decay honestly: if the plan scales spend 40%, the efficiency assumption comes down, and if you don't know how much, that itself is worth a test before you bet the year on it.

Step 5: Put the calendar under it. Every launch, promo, and campaign moment gets a line and an expected lift based on what similar actions actually did before. This is what makes it marketing's forecast: the calendar is the list of actions that create the revenue. A month with a gap between target and model is a month where the calendar owes you an action.

Step 6: Read it top-down and iterate. Start at contribution margin. Check the bands: margin healthy, marketing as a share of sales sane, payback on acquisition inside a horizon your cash can carry. If a line breaks, change an assumption or change the target. Don't ship a plan that only works if everything goes right.

Step 7: Lock it, then track daily. More on this at the end, it's where forecasts earn their keep or die.

Path two: the brand launching cold.

No history, so no curve to read, and anyone who hands you a precise twelve-month revenue model for an unlaunched brand is decorating a guess. The launch forecast has a different job. It's not there to predict the year. It's there to prove the economics, price the learning, and tell you when the real forecast starts.

Step 1: Prove the unit economics on paper. This is where the break-even math carries the whole plan, because it needs no history. Take your margin before marketing and get your break-even aMER. Then ask what customer-acquisition efficiency brands in your category, at your price point, actually achieve, benchmark ranges, not hopes. If the category norm can't clear your break-even, you don't have a forecasting problem, you have a product or pricing problem, and it's cheaper to learn that in a spreadsheet.

Step 2: Judge it on the first order, at the modal order value. Unless you have real evidence of extraordinary repeat behavior, the first order has to carry itself. "We'll make it up on lifetime value" is the most expensive sentence in DTC when the lifetime value is imaginary. And use your modal order value, the order size customers most commonly place, not the average, which one big gift order can drag upward. New brands with thin data get fooled by averages fastest.

Step 3: Fund a 90-day experiment, not a year. You're buying answers: what does a customer actually cost, what do they actually spend, do they come back. Set spend at a level where losing it entirely doesn't kill you, define what you need to see by day 90, and write those pass marks down before launch, so nobody moves the goalposts after.

Step 4: Budget creative volume as a plan input. Courtesy of how Meta works now: the platform's targeting runs on your creative, and a new brand needs a steady supply of distinct concepts just to find out what converts. That's a real line in the launch budget, not a nice-to-have.

Step 5: Re-forecast monthly, and graduate. Every month of trading replaces a benchmark with an actual. Re-forecast monthly for the first two quarters, no history means short cadences, and label every assumption as benchmark or actual so you can watch the model become yours. Somewhere around month six you usually have enough of a curve to switch to the existing-brand path. That switch, benchmarks out, your own cohorts in, is the moment the brand has a forecast instead of a thesis.

Path three: the expansion (same brand, new market).

This one looks easier than the cold launch and trips more people, because you have history. Just from the wrong place. The trap is treating the new market like an existing brand ("we know our numbers") and copying the model across. The opposite trap is treating it like a cold launch and throwing away everything you've learned. The truth sits in between: the old market is a benchmark library, not a forecast.

I've run this one twice, taking Beast from the US into the UK, and helping take a well-known US bedding brand, silver-infused sheets, a genuine DTC success story at home, across the same water. The second one taught me the most expensive lesson in this note, so let's start there.

The bedsheet problem. On paper it was the easy kind of expansion. Proven product, proven ads, years of trading data, and a mountain of customer content, thousands of real people filming themselves making their beds. What tripped it wasn't the media buying or the margin math. It was the bed.

Because Britain doesn't sleep the way America sleeps, and nobody had done the due diligence to check. A California King doesn't exist in the UK. The sizes that sound the same aren't: a UK King is a different bed from a US King, whose nearest British match is called a Super King. So the size chart, the thing a bedding brand is literally built on, didn't map name-for-name or dimension-for-dimension. Then the deeper one: Americans make a bed with a fitted sheet, a flat sheet on top, and a comforter over that. Brits skip the top sheet entirely: fitted sheet, duvet inside a washable cover, done. The flagship sheet set, whose hero piece is the top sheet, was selling a configuration most British bedrooms simply don't use, while the product a British customer actually reaches for, the duvet cover, wasn't the lead. Even the mountain of customer content, the asset that should have traveled best, was full of American bedrooms: wrong size names on the boxes, beds made a way the customer watching doesn't make hers.

And here's what that configuration break actually cost, because to me the most important thing in a brand is the offer. An offer isn't an ad. It's years of compounding: landing pages that had been through round after round of testing and optimization, post-purchase flows tuned on real behavior, the bundles, the pricing, the mountain of customer content, the ads that had beaten every other ad. All of it built on, and inseparable from, the American product configuration. The moment the configuration didn't map, we couldn't carry that proven offer into the UK. The thing that made the brand a success at home, the compounded, optimized offer, stayed home. We weren't launching with a head start. We were rebuilding the most valuable asset the brand had, from scratch, in a market that had never heard of us.

None of that shows up in a spreadsheet until you know to look. The forecast wasn't wrong because the math was wrong. It was wrong before the math started, because the catalog didn't map to the market, and the offer, the most compounded asset the brand had, couldn't follow it across.

Step 1: Audit the product, and the offer built on it. Before a media plan exists, SKU by SKU: does this exist here, under this name, at these dimensions, used this way, at this price position against the local competitive set? Then the same pass on the offer: which landing pages, bundles, flows, and creative still match this market's reality, and what has to be remade? Get a local into the room, someone who actually lives the category in that market, before a dollar of the launch budget moves. It's the cheapest hour of the whole expansion.

Step 2: Sort what travels from what resets. What travels reasonably well is the relationships. How your order value responds to bundles. The slope of your retention curve, how a customer's value builds after the first order. Which messages and angles convert, directionally. Your margin structure before delivery. These carry because they're properties of the product and the customer's relationship with it. Creative concepts travel too, with the asterisk the bedsheets just taught you: the idea travels, but the footage only travels if the product in the frame matches the market's reality.

What does not travel is the levels, and sometimes the product itself. Your acquisition cost resets, because in the new market you have zero brand equity, a different auction, and a different competitive set: expect early numbers that look like a cold launch, not like home. Your delivery stack gets rebuilt, new 3PL, new shipping rates, duties, VAT, so cost of delivery is recalculated, never copied. Conversion rate, channel mix, and even the seasonality: the shape of your year partially carries, but the peaks land differently, and some of your home-market moments don't exist there at all.

Step 3: Build the plan like a launch. Benchmark the new market cold: break-even math first, a 90-day learning budget, pass marks written down before you spend. The reset levels get launch-path treatment, no exceptions, however good the home numbers look.

Step 4: Steer with the home market. This is the one way home is genuinely valuable: your home numbers tell you what good eventually looks like for this product, so when the new market's actuals come in, you're not reading them blind. If home runs a 2.5 aMER at steady state and the new market is at 1.4 after ninety days, the question isn't panic, it's trajectory: is the gap closing month over month, and what on the calendar closes it faster?

Step 5: Let the new market overrule home. If a number refuses to converge after two or three quarters, believe the new market, not the old one. It's telling you something structural: different competition, different category maturity, different customer. The brands that get expansion wrong are almost always the ones that kept forecasting the old market onto the new one and called the gap "early days" for a year.

The one-line version: audit the product first, benchmark like a launch, steer like a veteran, and let the new market's own data overrule home the moment they disagree.

The part everyone skips: running it.

Forecasting is an exercise in execution more than planning. The forecast is not a January artifact. The monthly targets break into weekly and daily expectations, and actuals get read against them every day, contribution margin first, then spend and revenue, then new versus returning, then channels.

The payoff is speed. A January that's going to miss shows up in the dailies by January 3rd. Without the cascade, you find out on January 28th, after the month is gone and the cash with it. When a metric runs more than about 10% off plan, that's a flag on the day it happens: diagnose whether it's a volume problem or an efficiency problem, then act. That's the whole discipline. The forecast tells you where to look and when; the operating work is still on you.

And this is where a forecast becomes accountability instead of theater. Your target and your model are written down. Every week, reality either agrees with you or it doesn't, and either way you learn something specific. Miss by 30% and the postmortem writes itself: which assumption lied? That's a better conversation than any monthly report.

The short list of ways forecasts lie.

• Revenue forecast, margin ignored. Read from contribution margin down, always.
• One blended revenue line. Split new and returning or you can't see which engine broke.
• Today's efficiency straight-lined across double the spend.
• Delivery cost set to COGS alone, missing shipping, fulfillment, and fees.
• Repeat rate or order value assumed to rise with nothing on the calendar to raise it.
• A launch plan judged on year-one revenue instead of unit economics and cost-to-learn.
• An expansion forecast copied from the home market, with the gap called "early days" for a year.
• No seasonality: twelve equal months, a thing that has never happened.
• Built in January, opened in December.

← Back to Notes