There is a Gartner finding that every marketing leader with an AI implementation on their roadmap should read carefully. According to their research, more than 40% of AI automation projects will have failed by 2027.
The number is striking, but the reason behind it is more important than the statistic itself. Most of these projects don’t fail because the AI tools are bad. They fail because the data the tools are running on is bad, and nobody noticed until the automation was moving fast enough to make the problem impossible to ignore.
“Automating broken data doesn’t fix the data,” says Cristiano Winckler, Director of Digital Operations at Somebody Digital. “It makes the mistakes faster. You’re basically going to fail a lot faster without even knowing, and it’s downhill from there.”
This is the foundational tension sitting underneath a significant portion of the AI marketing conversation right now. Organizations are under pressure to show AI investment and AI results. Vendors are promising efficiency, personalization, and insight at scale. And many teams are layering sophisticated tools on top of analytics infrastructure that was already producing unreliable numbers, just more slowly.
The Data Problem Nobody Talks About in the AI Briefing
Spend a week attending B2B marketing events, and you will hear a great deal about AI’s potential to unlock insight, automate workflows, and personalize experiences at scale. You will hear rather less about what it takes to build a data foundation that makes any of that actually work.
“The most important thing 20 years ago is still the most important thing now, even with AI, and that is good analytics and data setup,” says John Wilkes, Head of Strategy and Co-Founder at Somebody Digital. “The robustness of the data, the granularity of the things that you’re tracking, the data governance you have around it is so incredibly important. I think the point here is that we’re talking about really cool data enrichment things to do, but it all boils down to the fundamentals of having a strong analytics setup. Not just slapping the latest AI tool on top of data that is fundamentally unreliable from the start.”
The irony is that this isn’t a new problem. The challenges of data consistency, attribution, and governance have been central concerns in digital marketing for more than a decade. What AI has done is raise the stakes dramatically. A tracking gap that used to produce a misleading report now produces a misleading model that makes autonomous decisions. A misattributed conversion that used to throw off a slide deck now throws off an algorithm that is spending your budget in real time.
The Three-System Problem
Most B2B marketing organizations operate with at least three sources of conversion truth: the ad platforms, the web analytics tool, and the CRM. In a well-configured environment, these three systems broadly agree. In reality, they rarely do.
“This is quite a common problem we see,” says Wilkes. “Paid media reporting one number, analytics reporting another, and CRM reporting another. This is difficult for the marketing team, but it also destroys credibility with the board.”
Each system counts conversions by different rules, uses different attribution windows, and has different opinions about what a conversion actually is. Google Ads counts a view-through and a click-through the same way it counts a direct conversion. Analytics counts the session that ended in a form fill. The CRM counts the record that was created. When a buyer visits your site three times across two weeks before converting, each system will tell a different story about what happened.
The divergence itself is not the core problem. The core problem is when teams try to make budget decisions, campaign optimizations, or AI-driven automations based on data they haven’t verified against a single source of truth. The AI will optimize faithfully toward whatever signal you give it. If that signal is noisy, the optimization will faithfully accelerate the noise.
“Data consistency (being able to trust the numbers) is paramount,” says Winckler. “It is the single most important thing you need to do to ensure you can make decisions based on the right data. Those will be the best decisions. Those will be the right decisions.”
What Outcome Enrichment Actually Fixes
The framework Somebody Digital uses to address this problem starts with what they call outcome enrichment: taking the conversion data that already exists in your CRM and feeding it back into the systems that are optimizing your campaigns.
Most teams currently measure to the lead, i.e., the form fill, the demo request, and the whitepaper download. The lead is easy to track because it happens entirely within the marketing platform. What happens after the lead, whether it becomes an MQL, gets accepted by sales, progresses to pipeline, closes as revenue, lives in a different system, and in many organizations, never makes its way back.
“Stop looking at what happened and start feeding the next decision,” says Wilkes. “Reports do not make money on their own. They’re just standalone data. It’s the enriched data that does the work.”
Outcome enrichment closes the loop. By pushing CRM data back to the ad platforms by flagging which leads became MQLs, which MQLs progressed to pipeline, which deals closed, you give the algorithm a signal that is correlated with actual business outcomes rather than with form-filling behavior. The platform learns to find more buyers who close, not just more visitors who click.
This matters more, not less, in an AI-optimized campaign environment. Platforms like Google’s AI Max and Performance Max are now making autonomous decisions about where to show ads, which creative to serve, and which audiences to prioritize. The quality of those decisions is entirely determined by the quality of the signal they’re trained on.
“Good news is you make that change on a landing page, or in this case, in your data setup, not only is it benefiting Google paid search, but all other AI engines as well,” says Winckler. “If it’s not triggering the right types of ads based on the right context, that means your content or your data is probably off. That means you have work to do.”
When AI Amplifies Bad Decisions
The mechanism by which bad data and AI interact is worth spelling out, because it explains why so many AI marketing implementations are underperforming expectations.
AI-driven campaign tools learn from conversion signals. When you set up a campaign to optimize for leads, and that campaign runs for 30 days, the platform builds a model of what a “converting” user looks like based on the users who converted. If those conversions include a significant proportion of students, competitors, and non-ICP visitors who happened to fill out a form, the model learns to find more of them.
The optimization then accelerates. The platform gets better and better at finding people who fill out forms. Lead volume may go up. Cost per lead may go down. The dashboard looks healthy. And sales is increasingly frustrated because the leads aren’t converting.
“When you’re running a Google Ads campaign and optimizing for lead volume,” says Winckler, “we want the highest possible number of leads for this budget. Google finds specific keywords or specific channels that get a lot of leads and doubles down on those. This is when you get a lot of what we call noise leads, because they’re not going to help you with MQLs. Maybe we’re talking about students searching for specific information that is related to your product or service, but they’re not going to buy anything from you.”
The automation has made the mistake faster. What used to be a low-grade campaign quality problem manageable, visible in quarterly reports, is now a high-speed optimization engine running in entirely the wrong direction.
The Practical Checklist
Fixing the data foundation before layering AI on top of it is not a glamorous project. It does not make for compelling product demos or conference slides. But Gartner’s research is consistent on this point: 84% of companies are stuck in a measurement “doom loop” where underfunded measurement leads to unclear impact, which leads to rising board skepticism, which leads to tighter measurement budgets. The cycle is self-reinforcing, and the only way out is through it.
The audit should cover at a minimum these areas.
Conversion consistency. Are the conversions you’re counting in the ad platforms the same conversions you’re counting in analytics and CRM? Spend a week comparing the numbers and tracing the discrepancies. The goal is a single definition of conversion that all systems use.
Funnel stage visibility. Can you see, from a single dashboard, how many leads became MQLs, how many MQLs became SQLs, and how many SQLs became pipeline? If those numbers only exist in the CRM and never appear in marketing reporting, the disconnect between marketing and sales metrics is structural.
Signal depth. What are your ad platforms actually optimizing toward? If it’s the form fill, you are optimizing for form fills. To move to MQL or pipeline optimization, you need to push CRM data back to the platforms. Most major CRM systems (Salesforce, HubSpot) have native integrations with Google Ads and LinkedIn that make this technically straightforward.
Attribution model. Are you running last-click attribution? “Last click attribution is dead. It’s been dead for a very long time,” says Winckler. “You need to measure every single touch, everything that happens across all campaigns, and try to find correlations.” The multi-touch view reveals which channels are genuinely contributing to pipeline and which are collecting credit for conversions that would have happened anyway.
Data governance. Who owns each tracking implementation? Who is responsible for ensuring that tracking parameters are consistent across campaigns, that UTMs are applied correctly, and that form submission events fire reliably? In many organizations, this is nobody’s explicit job, which means nobody is systematically catching the drift that accumulates over time.
The Foundation Makes Everything Else Work
One data point from a Somebody Digital client engagement makes the case better than any framework: after conducting a path-to-conversion analysis, the team identified that users who had visited a documentation and help subdomain, informational content, were nine times more likely to request a demo. Nine times. That insight was sitting in the data. It only became visible because the measurement was set up to see it.
“If you can have a view of every single touch and how the different channels and campaigns interact, and how user behavior when they interact with those ads and channels is influencing the MQL and pipeline,” says Winckler, “that’s when data actually starts creating revenue.”
The teams racing to implement AI marketing tools before their data is clean are not wrong to want the capability. They are wrong about the order of operations. The AI tools will be there in six months. The data quality work is what makes them worth using.
“You don’t need more data or AI,” says Wilkes. “You need your data to work harder than it is right now.”
Somebody Digital runs Marketing IQ Live every two weeks — a practical workshop for B2B marketing leaders covering the strategies, tools, and frameworks driving modern marketing programs. Join the next session.
Most projects don’t fail due to faulty AI tools, but rather because the underlying data is unreliable. Automating broken data does not fix it; it simply accelerates the errors, making mistakes faster and more difficult to ignore.
This refers to the common disconnect where ad platforms, web analytics, and CRMs all track and measure conversions using different rules and attribution windows. This discrepancy makes it difficult for teams to verify data against a single source of truth, often leading to conflicting reporting and lost credibility with the board.
AI optimization is driven by the signals it receives. If a campaign is trained on “noisy” data (such as non-qualified leads), the AI will faithfully learn to identify and attract more of those same unqualified users. This causes the automation to scale the mistakes rapidly rather than optimizing for business outcomes.
Outcome enrichment involves pushing CRM data (such as MQL status, pipeline progression, and closed revenue) back into ad platforms. Instead of optimizing for easy-to-track metrics like form fills, this trains the algorithms to identify users who are likely to become actual customers.
To build a robust foundation, audits should cover:
- Conversion consistency: Ensuring all systems use a single, unified definition of a conversion.
- Funnel stage visibility: Establishing a dashboard that connects marketing and sales metrics from lead to revenue.
- Signal depth: Moving beyond optimizing for simple form fills by integrating CRM data.
- Attribution model: Moving away from last-click attribution toward multi-touch models that reveal true channel contribution.
- Data governance: Assigning clear ownership for tracking parameters and ensuring consistency across campaigns.
