Brand vs Non-Brand Paid Spend: The ROAS Split That Should Change Your Budget
Blended ROAS lets brand campaigns borrow credit from demand they did not create. Splitting brand from non-brand, then ranking on contribution per new customer, turns the reports you already have into a budget you can defend.

Brand-keyword ads showed no measurable short-term benefit in a landmark NBER study. Non-brand ads worked for some users and not others. Yet most paid budgets are still defended on blended ROAS, a number that averages brand spend, which buys revenue you already had, against non-brand spend, which buys new customers.
Blended ROAS is the Dashboard Illusion in its purest form. It lets brand spend borrow credit from demand it did not create, and it hides the decision you are already making about where incremental growth comes from.
This analysis separates paid search and paid social into brand and non-brand, then re-ranks on contribution per incremental customer, the lens that blended ROAS cannot give you. It shows how to run the split with reports your team already has, the four patterns that tend to show up, and what the incrementality evidence actually says about paid search returns and where it stops. The result is a budget defended on contribution, not on blended ROAS.
Blended ROAS Is Hiding a Decision You Are Already Making
Blended ROAS supports exactly one decision: spend more, or spend less. It cannot tell you which campaign absorbed the last increment of budget, or what kind of demand that increment bought.
Brand keyword clicks and cold prospecting clicks arrive in the same account total. The customer who typed your name and the customer who had never heard of you both count as one attributed order at the same value. The aggregate therefore looks identical whether you are funding demand creation or paying to intercept demand you already had.
That is the Dashboard Illusion in its purest form. A healthy blended number can conceal a prospecting campaign generating every new customer you have, sitting beside a brand campaign buying back orders you were going to receive anyway. Field-experiment research found that conventional estimates of paid search returns vastly overstate the incremental effect.
The cost surfaces later. Under margin pressure, the standard response is a Blunt-Instrument Budget Cut: a flat percentage off every line, which strips money from demand creation while capture spend keeps running untouched.
Before you open a report, state the decision plainly: which lines are buying demand that did not exist, and which are paying to intercept demand that did? A profit-on-ad-spend floor set per product rather than per account is what makes that question answerable. Everything below serves it.
Why Brand Spend Borrows Credit It Did Not Create
Someone who types your brand name, or your product plus your name, has already decided. That click confirms intent; it does not create it. Brand search therefore converts at multiples of non-brand, and the multiple is not efficiency. It is the accumulated result of paid social, organic content, PR, email, word of mouth and prior purchases. The brand keyword is the collection point, not the source.
Blended ROAS credits that collection point with the whole sale. That is why brand campaigns sit at the top of the account and non-brand looks like the expensive problem child. The credit was earned elsewhere and booked here.
Field-experiment research on paid search is blunt. At a large, well-known retailer, brand keyword ads produced no measurable short-term benefit, a result the authors explicitly flagged as an extreme case (NBER working paper). Take the direction, not the magnitude.
The same mechanism runs through paid social: branded search terms, retargeting past purchasers, purchaser-list audiences. Each intercepts demand the business already had. No study here tests that channel, so treat it as reasoning from the same logic, not a proven result.
Splitting capture from creation also means allocating shared spend honestly, the same discipline behind splitting shared costs without distorting SKU margin. Assign credit where it was earned, and the split starts telling the truth.
How to Split Brand and Non-Brand Using Reports You Already Have
The split starts in the search account. Pull a campaign and keyword report for a fixed date range, then tag every keyword and campaign as brand, non-brand, or ambiguous under one written rule: your brand name, common misspellings, and brand plus product terms. Write the rule down so a second person can apply it without asking you.
Then hunt the leaks. Search Partner and Display placements inside non-brand campaigns, dynamic search ads matching branded queries, and Performance Max or Shopping campaigns harvesting branded traffic all move capture spend into the creation column. The search terms report shows what actually triggered the spend, which is where the mislabelling surfaces.
In paid social, split by audience rather than campaign name. Existing-customer lists, retargeting pools, and engaged-view retargeting count as capture; cold prospecting counts as creation. Report them separately even when the platform insists on blending.
Build two columns instead of one: spend, revenue, and orders per channel, split into capture and creation. This is a reporting change, not a new stack. Once it exists, you can rank campaigns by contribution margin rather than headline ROAS.
Run the split on a rolling 90-day window and the same window a year earlier. Seasonality and a shift in spend mix between the two periods are the two things most likely to fake a trend.
Keep it to one spreadsheet tab, updated weekly. If the first pass takes longer than an hour it is too granular to survive a busy month, and it will quietly stop being done.
Reading the Split: Four Patterns That Show Up
With the split built, four shapes recur. Attach a figure to each: a pattern you cannot state in one sentence with a number is not yet a finding. The figures below are illustrative, to be replaced with your own.
Pattern one: capture flatters the blend. Brand at 15% of spend and 45% of attributed revenue is ordinary. Expected behaviour, not efficiency. Your blended ROAS is being carried by spend that intercepted demand it did not create.
Pattern two: non-brand buys less newness. It holds most of the spend and most of the attributed orders, but its share of new customers slips while auction costs rise. More money, fewer genuinely new people.
Pattern three: leakage inverts the picture. Branded queries matched inside non-brand, Performance Max, or Shopping make non-brand look stronger and brand look weaker, the reverse of the usual assumption. A search terms report is what exposes it.
Pattern four: track new customers per £1,000 of non-brand spend, monthly. A flat or falling ratio while spend climbs means you are buying repeat purchase paths and calling it growth.
Read against real margin rather than platform revenue, as the four-level profitability framework sets out, these stop being reporting quirks. They become budget decisions.
Contribution per Incremental Customer: The Lens Blended ROAS Cannot Give You
Those four patterns point to one conclusion: attributed revenue per pound spent is the wrong unit. The unit that matters is contribution margin per new customer acquired: gross margin on the order, less the ad spend it took to win that customer, counted only where the customer would not have bought without the ad.
You cannot measure that exactly without a holdout, so run small ones. Pause brand search across a set of postcodes, or hold back a slice of budget or audience for a fortnight, then compare orders against a matched baseline. Published blackout research puts the overstatement at 2 to 10 times causal impact, which is why the controlled pause does work attribution cannot.
Where a holdout is impractical, use proxies and label them as proxies: the new versus returning customer split, first 90-day margin per new customer, and whether branded search volume held steady when brand spend was cut.
The number to watch is not ROAS. It is: for every extra £1 of non-brand spend, how many first-time customers appeared, and what margin did they deliver in their first 90 days?
Contribution per new customer is Hawk Margin’s framing, not a research finding. Treat it as the proposed lens and pressure-test it against your own holdout data rather than accepting it as settled. It is the same discipline as measuring beyond ROAS, against contribution margin. A first-time customer carries future value a returning customer does not add, and capture spend on someone already coming back cannot manufacture that value however good the ROAS looks.
What the Incrementality Evidence Actually Says, and Where It Stops
Holdouts give you a number. The published evidence tells you why you needed one in the first place.
The strongest study here is a set of large-scale field experiments at a single retailer, later published in Econometrica. Brand keyword ads produced no measurable short-term lift, a result the authors themselves describe as an extreme case. Do not generalise it into “brand spend does not work”.
The transferable finding is about who responds. Non-brand ads moved new and infrequent users. Frequent users whose purchases the ads did not influence carried most of the spend, which is why average returns came out negative.
Then the methodological lesson, the one to keep. Because clicks and purchase intent are correlated, conventional non-experimental estimates overstate paid search returns. Attribution credits correlation. Only a randomised holdout separates cause.
Read that as a warning about your dashboard, not a verdict on your channel. The marketing fundamentals most ecommerce founders get wrong start with treating attribution as proof. Your brand is far less familiar than the retailer in the study, so brand capture may be doing real work in your account.
Know where the evidence stops. Paid search only, US-focused, at a large brand with high repeat frequency. The mechanism may travel to paid social; the measured result does not. Reason from the mechanism, then test on your own account. Evidence published a decade ago should not set a 2026 budget. Use it to justify running the test, not to skip it.
What the Split Does to Your Budget Decisions
The evidence earns you the test. The budget call is what follows.
Reallocate before you cut. If contribution per new customer on non-brand is healthy while brand capture adds nobody new, the money moves from capture to creation, and it stays in the account. That is the line between a surgical decision and a Blunt-Instrument Budget Cut. Ranking campaigns by contribution margin to cut ad spend is the discipline that makes it repeatable.
Set the split as a target. Decide what share of paid budget belongs in creation, write it down, and hold that share when a brand campaign’s headline ROAS makes the easier number tempting.
Fund creation to protect next quarter. Non-brand spend you fund now is what produces branded queries later. Cap brand spend without funding creation and the branded demand base you draw on next year shrinks. Treat that as a hypothesis worth testing against your own branded search volume, not a settled fact.
Fund holdouts as a permanent line item. A small slice of paid budget for controlled pauses and audience holdbacks is the only thing that keeps contribution per new customer honest over time. Without it, you are back to correlating clicks with purchases.
Anchor to real margin. Contribution figures sitting beside actual product-level margin, as they do in HawkScan, are defensible. Attribution pulled from an ad platform is not.
Report one page. Brand spend, non-brand spend, new customers, contribution per new customer, direction of travel. One page, one owner. That page is where the Profit Accountability Gap closes, because the number on it has a name against it.
The Shift: A Budget Defended on Contribution, Not on Blended ROAS
Blended ROAS answers a different question. It confirms the account converted. It cannot tell you which half of the spend created demand and which half merely collected it. That is not a flaw in the metric, it is a mismatch between the metric and the decision.
The action is small and starts this week. Tag brand and non-brand in the search account. Split capture from creation in paid social by audience. Add one line to the report: new customers per £1,000 of non-brand spend. For product-level spend, attributing ad spend to individual product performance closes the same gap one level down.
Then hold spend decisions to that line rather than the account total. Money that buys customers who would never have found you is growth. Money that buys customers already on their way is rent.
What changes is the conversation, not the dashboard. A budget defended on contribution per new customer survives a board meeting. A budget defended on blended ROAS survives only until someone asks which customers it bought.
Conclusion
Blended ROAS confirms conversions. It cannot tell creation from capture. Brand spend often borrows credit it did not create. The split exposes which pound buys new customers and which merely collects existing demand. The metric that matters is contribution per incremental new customer, not account-level ROAS.
Next step: this week, tag brand and non-brand in search, split capture from creation in paid social, and add new customers per £1,000 of non-brand spend to one page. Give it one owner. Review spend against that line, not the blended total. That is how the Profit Accountability Gap closes, because every number has a name against it. A budget defended on contribution survives scrutiny. Defend that page, and let the dashboard follow.



