Scaling media spend 3x while raising ROAS.

At a glance

Paid media budget	About $100M to $300M+ a year (3x)
Incremental revenue	Up 42%
ROAS	Increased while spend scaled
Credit Karma revenue	$1.8B in fiscal 2022 to $2.3B in fiscal 2025
My role	Led marketing measurement, March 2021 to October 2025
Methods	MMM, geo-lift tests, diff-in-diff, synthetic controls, brand lift

The challenge

Credit Karma wanted to spend more efficiently across all verticals but no one could agree on what was working. Each vertical reported its own numbers against its own KPIs, so Marketing and Finance were looking at different answers to the same question.

• No shared definition of success. Channels were judged on different metrics, with no common view of short-term return versus long-term value.
• Testing was ad hoc. Experiments ran when someone asked for one, and readouts took about six weeks.
• Legacy attribution over-credited the bottom of the funnel. That made it hard to justify TV and other upper-funnel spend.

Scaling a budget on that footing usually means paying more for the same customers.

The approach

I built the program in three phases, each one making the next possible.

1. Align all spend to the same KPIs

Every dollar got measured against one short-term KPI and one long-term KPI. I worked with Finance, Marketing, Engineering and about 30 stakeholders to agree on the definitions before any modeling started. That gave the company one language for comparing a TV buy to a paid search campaign.

2. Build a consistent test cadence

With shared KPIs in place, I set up a standing calendar of incrementality tests across channels, including search, social, programmatic, linear TV and CTV. Automated monitoring in Python and SQL replaced six-week readouts with weekly checkpoints, so the team could run more tests at once and stop the weak ones early.

3. Implement MMM, then let it steer the tests

I selected and onboarded a media mix model vendor, then owned the data pipelines, model validation and calibration against test results. The model became the source of truth that Finance and Marketing used to set budgets. From there it worked as a loop: the MMM showed where the next dollar was least certain, and those questions set the test plan.

The results

Spend tripled and efficiency improved at the same time, which is the opposite of what usually happens when a budget scales.

• Budget grew 3x, from about $100M to more than $300M a year.
• Incremental revenue rose 42%.
• ROAS increased over the same period.
• Upper-funnel spend earned its place. With MMM as the source of truth, TV and brand investment could be defended in the same terms as direct response.

Credit Karma's reported revenue grew over the same period. Intuit's fiscal year ends July 31.

Fiscal year	Credit Karma revenue	Change
2022	$1.81B
2023	$1.63B	Down 9%
2024	$1.71B	Up 5%
2025	$2.26B	Up 32%

Revenue dipped in fiscal 2023 before reaching a record in fiscal 2025, 25% above fiscal 2022. Many things drive segment revenue, so I don't claim that growth for measurement alone. What measurement did was let the company keep investing through the dip with confidence about which dollars were working.

Sources: Intuit's fiscal 2024 Form 10-K and fiscal 2025 Form 10-K. The fiscal 2025 figure is fiscal 2024 revenue plus the reported $555M increase.

Why it worked

The order mattered more than any single method.

• Definitions before models. An MMM is only useful if Finance trusts what it's measuring. Agreeing on KPIs first meant the model's answers were accepted when they arrived.
• Tests and models checked each other. Experiments calibrated the MMM, and the MMM decided which experiments were worth running.
• Speed changed behavior. Weekly checkpoints made testing a habit instead of a special project.

The same sequence works for any brand that's about to scale spend and isn't sure its current numbers will hold up.