Is Your Marketing Budget Working? How to Measure and Improve ROI

1. Blog
2. Measurement & Budget

By Saroj Jha · January 15, 2026 · 9 min read

ROI, ROAS and CAC payback each answer a different question about the budget you already have. Here is how to count what your marketing caused, judge the next dollar rather than the average, and move money every month.

From this article

A working budget changes monthly; an inert one changes annually.

Most marketing budgets are built by taking last year’s number and adding a little. That settles how much you can spend. It says nothing about whether the spending works, and that is the question this article answers: how to measure the return on the budget you already have, and how to improve it without spending more.

If you are still deciding how big the budget should be, start with how much a startup should spend on marketing and the marketing budget benchmarks by stage. Here the size is fixed and the job is to make it earn.

A budget is working when you can show three things for each channel: the gross profit it returns, not the revenue; what it caused, not what it touched; and what the next dollar would buy, not the average one. Measure those, then move money every month in small steps.

Why the question is getting harder to dodge

Budgets aren’t growing, so the return has to. Gartner’s 2026 CMO Spend Survey of 401 marketing leaders puts marketing at 7.8% of company revenue, up from 7.7% in 2025, and 56% say they lack the budget to deliver this year’s strategy^[3]. The CMO Survey’s 2026 edition, 308 marketing leaders at US companies, found a mean of 9.0% of revenue and a median of 5%. In the same survey, 55.5% said they feel increasing pressure from their CFO to prove the value of marketing, and 58.8% from their CEO^[1].

The common response is the trap. Asked how marketing copes with that pressure, 70.6% of those who answered said by focusing on short-term impact over long-run gains. Gartner adds that less mature organizations may be over-indexing on short-term optimization and on the channels that are easiest to measure and automate^[4]. Both surveys mostly describe large companies, but a startup’s board and runway apply the same pressure. Proving return is not the same as funding whatever is easiest to count.

Running case: Example Co. Fictional

Example Co. is a Series A startup selling membership and class-booking software to independent fitness studios with one to five locations, at $250 a month per studio. At 80% gross margin, each customer brings in $200 of gross profit a month. It spends $60,000 a month across five channels: branded search, non-brand search, Meta ads, review-site listings and fitness-industry trade shows. It wins 30 new customers a month, a blended CAC of $2,000 and a payback of 10 months. All its numbers are illustrative.

ROI, ROAS and CAC payback answer different questions

Each of the three common return metrics is right for one question and wrong for the others.

• ROAS is the revenue an ad platform credits to your ads divided by what you spent on them. Google’s own example: $5 in sales for every $1 of ad spend is a 500% ROAS^[5]. It is fast and available per campaign, but it counts revenue rather than profit and relies on the platform’s own attribution.
• Marketing ROI is the profit marketing caused, net of the spend, divided by the spend. The marketing dictionary of the Marketing Accountability Standards Board (MASB) writes it as incremental revenue attributable to marketing, times contribution margin, minus marketing spending, all divided by marketing spending^[6]. Revenue minus cost over cost, the formula many dashboards use, overstates the return by the whole cost of delivering what you sold.
• CAC payback is the number of months of gross profit it takes to earn back what you spent to win a customer. Bessemer measures it against gross-margin-adjusted revenue and sets targets of under 12 months for SMB customers, 18 for mid-market and 24 for enterprise^[7]. It answers the cash question: how long the budget is tied up before it comes back.

Example Co., step 1 Fictional

Non-brand search costs $20,000 a month. Google reports 10 customers and, because Example Co. passes each one’s first-year contract value of $3,000, a ROAS of 150%. A conversion lift test says 8 of those customers were caused by the ads. Their first-year gross profit is 8 × $3,000 × 80% = $19,200, a first-year ROI of −4%. If the average customer stays three years, the same 8 bring in $57,600 of gross profit, an ROI of 188%. CAC payback is $2,500 per customer ÷ $200 a month = 12.5 months.

None of these numbers is wrong. ROAS is the right tool for daily bidding inside a channel. ROI on incremental gross profit, over a stated window, is the right tool for moving money between channels. Payback is the number your board and your cash plan care about. For a subscription business the window decides the ROI verdict, so always say which window you used.

Do this

Write down one definition for each metric and use only that one: ROAS from the platform, for bidding only; ROI on incremental gross profit over a first-year window and a lifetime window; CAC payback in months of gross profit. Then recompute your three biggest channels on all three. The free Unit Economics Calculator does the payback and lifetime arithmetic.

Count what your marketing caused, not what it touched

Every ad platform grades its own homework, and each one counts a customer it touched as a customer it won. Add up the platform reports and you usually get more customers than you actually signed.

The best-known evidence comes from eBay. When it switched off ads on searches for its own brand name, natural search links brought back 99.5% of the click traffic it had been paying for. For non-brand keywords, the experiment put the average short-term return at −63%, where standard regressions on the same data had suggested more than 1,400%^[2]. eBay is a household name, so a startup whose competitors bid on its brand will lose more when it stops, but you only learn how much by testing. A study of 15 large advertising experiments at Facebook reached a similar conclusion: the observational methods most advertisers rely on often failed to reproduce the effects the randomized tests measured^[8].

Testing has become cheaper. In November 2025 Google said an incrementality experiment that might once have cost more than $100,000 can now be run for $5,000^[9]. How to design and read a geo holdout is in incrementality testing for startups.

Example Co., step 2 Fictional

The platforms and the CRM claim 40 customers in a month when Example Co. signed 30. A four-week holdout, with branded search switched off in half its regions, shows only 2 of the 12 customers credited to branded search were caused by the ads: the rest searched its name and clicked the organic link. Branded search looked like the cheapest channel at $500 a customer. Counted by what it caused, it is the most expensive, at $3,000. Trade shows go the other way: the CRM credits them with 3 customers, but 5 a month name the show on the sign-up form.

Do this

Run one holdout test a quarter on the channel you would most hate to be wrong about. For most startups that is branded search. Use only the customers a test or a model says were caused when you compute CAC and ROI, and for channels you can’t test yet, such as events, compare the CRM with what customers say on a required “How did you hear about us?” field. The free Self-Reported Attribution Coder sorts the answers into channels.

Judge the next dollar, not the average

Every channel runs into diminishing returns: each further dollar reaches buyers less ready to act. An average return describes money you have already spent. The budget decision is about the next dollar. Google’s documentation for Meridian, its open-source marketing mix model, draws the same line: ROI is the incremental outcome at your current spend divided by that spend, marginal ROI is the return on the next dollar, and the channels with the highest marginal ROI are the best place for additional money^[10].

You don’t need a model to get a first read. Raise a channel’s spend for four weeks, ideally in only some regions, and divide the extra spend by the extra customers. That is the marginal CAC. The MMM-Lite guide shows how to fit full response curves, and the Paid Media Budget Allocation Playbook turns marginal CAC into a channel split.

Example Co., step 3 Fictional

Non-brand search has the best volume, so the instinct is to feed it. A step test raises it from $20,000 to $30,000 a month: caused customers go from 8 to 10. The extra $10,000 bought 2 customers, a marginal CAC of $5,000, twice the $2,500 average. Review-site listings go from $8,000 to $12,000 and caused customers from 6 to 8: a marginal CAC of $2,000. That is worse than the channel’s own $1,333 average, as diminishing returns predict, but far better than search’s $5,000 and branded search’s $3,000. The next dollar belongs in the smaller channel.

Do this

Before you add money to a channel, raise its spend by 20 to 50% for four weeks and compute marginal CAC as extra spend divided by extra customers caused. Fund the channel with the lowest marginal CAC, not the one with the best average, and log each test in the free Paid Media Test Log.

Free tool

Split the budget on marginal CAC

The free Paid Media Budget Allocator takes your channels, costs and a CAC target, and splits paid spend with diminishing returns built in.

Move money every month, in small steps

A working budget changes monthly; an inert one changes once a year. Reallocation is where measurement pays off, but only if the moves are small enough to read and the rules are set before the results arrive. These are the rules I use:

• Rank on marginal incremental CAC. Take money from the channel whose next dollar buys least and give it to the one whose next dollar buys most.
• Move 10 to 20% of a channel’s budget at a time. One change per channel per month, so you can tell what each move did.
• Give slow channels a longer window. Events, content and brand work pay back over months. Asked how long the effect of their marketing spending lasts, The CMO Survey’s respondents gave a median answer of six months, and 54% said a year or more^[1]. Judge them over a quarter or two, on what customers say as well as what the CRM credits.
• Decide before you test. Write down what result will make you cut, hold or scale, so the decision isn’t argued after the fact.

Example Co., step 4 Fictional

Example Co. cuts branded search from $6,000 to $2,000 a month, keeping bids on the searches where competitors buy its name, and moves the $4,000 to review-site listings. Non-brand search and Meta hold. Trade shows hold too, and get judged over two quarters on the sign-up form. The next month it loses one customer from branded search and gains two from review sites: 31 new customers on the same $60,000. Blended CAC falls from $2,000 to about $1,935 and payback from 10 months to about 9.7. That is 12 more customers a year for no extra spend.

Do this

Book a 30-minute budget review on the same day each month. Bring one table: spend, customers caused and marginal CAC by channel. Make at most one move per channel, of 10 to 20% of its budget, and write down what you expect it to do.

Report return the way a CFO reads it

A board needs to see what the money bought, how sure you are, and what you are changing. I keep the marketing page to five lines: total spend, new customers, blended CAC and payback, incremental CAC for every channel you have tested, and the moves made this month with the reason for each. Cost per lead, conversion rates and pipeline sit underneath as leading indicators, not as the headline.

Leave platform-reported ROAS and conversions out of the board pack. Added up, they claim more customers than you won, and a CFO who spots that once will discount every number after it. Report the slow channels on their own line, over their own window, so a quiet month doesn’t get them cut. The Marketing KPI Board Report template lays this out.

Do this

Replace any slide that sums platform ROAS with one table of spend, customers, blended CAC, payback and incremental CAC for tested channels. Add one line on what you moved, why, and which test runs next quarter.

What changes by stage

• Pre-seed — There is little budget to optimize. Track cost per conversation with a qualified buyer and ask every customer how they found you.
• Seed — Blended CAC and payback, a required “How did you hear about us?” field, and one holdout on branded search once spend allows.
• Series A–B — Incremental CAC by channel, one test a quarter, step tests before scaling any channel, and a mix model once you have the history for it.

If you want me to look at your own funnel, positioning and channels, my Growth Audit ends with three prioritized fixes and a 30-day action plan.