Where Your AI Marketing Budget Actually Goes (Spoiler: Not Where You Think)
90% of organizations increase AI marketing spend, but only 12% can prove it works. The gap between adoption and accountability is growing wider.
Dellon S.
June 2, 2026 • 8 min read
The Spending-to-Proof Disconnect
You'd think the logic was simple. Buy more AI tools. Run more AI campaigns. Measure the results. Get paid. But reality has fractured somewhere in the middle.
Comviva just published numbers that should terrify every CMO: 90% of organizations increased their AI marketing budget in 2025-2026. Yet only 12% can prove their AI investments actually moved the needle on revenue. That's not a rounding error. That's structural.
The adoption wave is real. It's just not connected to any proof of return.
90%
Orgs increased AI spend
12%
Can prove it works
78%
Proof gap
70%
Attribution systems broken
Why This Happens
Here's what makes this different from past marketing failures. When you deployed a new ad platform in 2020, at least the platform itself told you click rates, conversion rates, spend per acquisition. You had baseline metrics. They might not tell the whole story. But they told some story.
AI marketing tools don't work that way.
When you deploy an agentic AI system to optimize bids across 15 channels, rewrite creative in real-time, and adjust audience targeting by the hour, you get a black box that can't explain its own decisions. The output is optimized, sure. But optimized for what? Attribution? Brand safety? Margin? Nobody knows. The tool learned to do something. Whether that something makes money is a different question entirely.
Add multi-touch attribution (already broken for 70% of digital marketers). Layer in incrementality measurement that requires holdout groups nobody wants to actually hold out. Then ask your VP of Marketing to explain why CAC went up 8% despite a 30% AI spend increase. The data exists. The truth doesn't.
The Vicious Cycle
Here's how the proof gap becomes systemic:
- Finance pressures marketing to hit ROI targets.
- Marketing buys AI tools to automate and optimize faster.
- Those tools don't integrate with attribution systems built for human-decided campaigns.
- Nobody measures incrementality because it requires turning off AI for 20% of traffic for a month.
- Marketing reports "AI is running X% of our campaigns" without saying "we have no idea if those campaigns work."
- Finance doesn't see proof. Marketing buys more AI tools to close the gap, hoping scale fixes the problem.
- The cycle repeats, now at 2x budget.
By the time you realize you've been optimizing toward a metric that doesn't connect to revenue, you're committed. Your entire martech stack depends on it. Your team is trained on it. Your forecast is built on it.
What 12% of Organizations Actually Do
The CMOs who can prove their AI works fall into one of three camps. None of them are secret. All of them are unglamorous.
The first group did the boring thing:
They ran a pilot. Before deploying AI across the entire demand gen program, they isolated it. Set a baseline. Measured incrementally. Then scaled only the tactics that actually moved conversion rates. This takes 3-6 months and nobody blogs about it because it's not a story. It's discipline.
The second group already had their fundamentals right:
They had a first-party data strategy. Their attribution model was reliable enough for decisions (not perfect, just reliable). They onboarded AI on top of infrastructure that could actually measure it. When new AI tools came in, they knew how to ask: does this improve our metric or just make our dashboards more complex? Most AI tools failed that question.
The third group is luck:
They work in a vertical where last-click attribution actually works. SaaS with a 7-day CTA, maybe. Or e-commerce with a single checkout flow. Verticals where the customer journey doesn't involve 12 touchpoints and 40 days of consideration.
The 88% majority? No infrastructure. No discipline. No luck. Just faster spending on tools that produce nice dashboards.
The Real Cost
Here's what the proof gap actually costs you:
Budget allocation is flying blind:
You're distributing AI spend based on vendor pitches and peer pressure, not results. If Channel A's AI-driven campaigns are actually neutral and Channel B's are +15% ROAS, you'd never know. You might be starving the one that works to fund the one that doesn't.
Brand damage risk is invisible:
When your AI system makes a bad targeting decision (exclusion list too aggressive, lookalike audience too broad, creative copy offensive to a demographic), you don't see it until it shows up on Twitter. You can't audit what you can't measure.
Talent drain is real but slow:
Your analytics team watches dashboards update in real time, knowing none of it connects to actual revenue. Your data engineers build pipelines for tools that output false confidence. Smart people leave. Remaining team says this is why everyone hates martech.
Vendor lock-in hardens:
By the time you might ask "is this AI tool actually valuable," you've integrated it with four other tools, trained your team on its interface, and built your forecast around it. Exiting costs more than staying.
The Escape Hatch (If You Want It)
You can't wait for the industry to fix this. Attribution is getting worse, not better. More channels means more unknowns. More tools means more noise.
But you can build a decision framework that doesn't depend on perfect measurement:
Define what working actually means before you deploy:
Not impressions. Not AI-optimized CTR. Not brand awareness lift studies. Real proxy: does this AI system improve the metric we actually care about (ROAS, CAC, LTV, margin?) by a measurable amount, within 90 days?
Run one AI deployment at a time:
Freeze everything else. This is painful. This is also the only way you'll ever know what's causing a change. If you roll out three AI tools, three martech upgrades, and a new creative strategy in the same quarter, you've built a system that guarantees ignorance.
Use holdouts ruthlessly:
Set aside 5% of traffic for a control group. Run your best non-AI strategy there. Measure what happens. It costs you a small amount of revenue to gain certainty about the rest. Most organizations call this impossible. High-performing ones call it basic hygiene.
Stop hiring for AI integration and start hiring for skepticism:
You need someone whose job is asking: does this work? And you need to empower them to kill projects based on that answer. This person will be unpopular at vendor meetings. Hire them anyway.
"Speed is easy. Proof is hard. That's why 90% of organizations can increase spending but only 12% can prove it works."
The Uncomfortable Question
The math says: 88% of AI marketing budgets are producing unproven returns. Not bad returns. Unproven returns. The difference matters because you can at least optimize bad returns. Unproven returns just sit there, consuming budget and confidence.
Your competitor might be in that 88%. Your competitor might also be in the 12%. You won't know until one of you stops pretending that speed and automation equal results.
The question isn't whether your AI tools are smart. It's whether you're being smart about measuring them.
Related thinking: When AI agents fail, customer retention fails. Or why ROI measurement becomes impossible with agentic AI.