AI Success Isn't a Modeling Problem. It's a Data Problem.

As my team works through thousands of potential duplicates for a client, the issue become straightforward. We don't have a data modeling issue. We have a data coherence issue.

The same contact exists in multiple forms. Context is split across systems. Key fields disagree just enough to undermine confidence in the data. Nothing is completely wrong, but the overall picture is not reliable.

And at an executive level, that shows up in one place: performance.

The gap isn't AI adoption. It's impact.

Most organizations have already invested in AI. Nearly all are experimenting with it.

But very few are seeing meaningful returns. BCG puts that number at approximately 4-5% of companies actually realizing real value.

That tells you something important.

This isn't about access to technology. It's about the conditions required for that technology to work.

The constraint is the data layer

In practice, AI doesn't break because it's incapable. It breaks because the underlying data is inconsistent.

You see it immediately when outputs are questioned:

• Accounts don't match across systems

• Contacts are duplicated or outdated

• Activity and attribution don't line up

The issue isn't intelligence. It's reliability.

Salesforce's data reinforces this: most leaders recognize that AI outcomes depend on data quality, yet many acknowledge that they don't have the quality required to support it.

At that point, adoption slows. Not because AI isn't useful, but because it isn't trusted.

Duplicates are a structural sign

Duplicate records are a visible entry point into the issue.

They typically exist because systems aren't integrated, identities aren't resolved, and updates don't propagate. Over time, each platform builds its own partial version of the customer.

What that creates is not just redundancy, it creates fragmentation.

From a revenue perspective, this affects everything:

• Pipeline accuracy

• Segmentation and targeting

• Attribution

• Forecasting

AI simply amplifies these issues. It doesn't fix them.

Scale requires consistency, not more data

Adding more data doesn't solve this issue. It compounds it.

What matters is whether the data behaves like a system:

• A single, consistent representation of accounts and contacts

• Current information, not historical artifacts

• Alignment across tools and workflows

Without that, increased data volume leads to increased noise, and AI outputs become harder (not easier!) to use.

Where leading organizations are different

The organizations getting value from AI have made a structural shift. They've invested in data environments that are connected, continuously updated, and governed at the ecosystem level.

BCG highlights the impact clearly: leaders in data and AI scale more use cases and generate significantly greater returns.

The different isn't better models. It's a stable, unified data foundation that those models can rely on.

This is a revenue issue

Framing this as a data quality initiatives understates the impact. When data isn't aligned, it limits:

• Forecast accuracy

• Go-to-market coordination

• Personalization

• AI deployment

And it carries real financial consequences. Organizations report millions in losses tied directly to poor data quality, with broader estimates reaching into the tens of millions annually. (Source: Forrester)

Final Thought

Working through duplicate records makes one thing clear: the issue is not the records themselves. It is the system producing them.

AI is now mature enough to create real advantages. But it requires a data environment that is consistent, current, and connected.

Most organizations are not there yet. The model is not the constraint. The data is.