AI Is the Tool. Data Is the Fuel.

The conversation around AI is moving extraordinarily fast. Every week brings new models, agents and capabilities promising to analyze more, automate more and make organizations smarter.

Naturally, that has organizations asking: What can AI do for us?

But there is another question that may ultimately matter more: What does our AI actually know?

AI does not automatically understand your customers, your history, your transactions or the relationships between different parts of your organization. It works with the information and context available to it. That means the quality of AI output depends on much more than which model or AI tool you choose. It depends on the data and technology architecture underneath it.

As Salesforce CEO Marc Benioff put it:

“You have to get your data right to get your AI right.”

That sounds simple. But “getting your data right” means much more than cleaning up a database.

It means asking whether your systems actually give AI enough connected, structured and accessible information to understand your organization in the first place.

And increasingly, the industry is recognizing that this foundation is becoming one of the biggest determinants of AI success.

Gartner reports that more than 75% of organizations are prioritizing investments in AI-ready data, while 90% of chief data and analytics officers say their existing data architecture needs an overhaul to support new AI use cases.

The AI layer matters.

But so does everything underneath it.

The Same AI Can Produce Very Different Intelligence

Imagine two organizations using the exact same underlying AI model.

At the first organization, AI can access registration transactions. It knows that Jane registered for an event.

At the second organization, AI can see that Jane has participated for five consecutive years, registered for several different events, donated to another participant, fundraised in the past, volunteered last season, purchased merchandise, regularly engages with marketing communications and has not yet registered for next year's event.

Now ask both systems the same question:

Which customers are most at risk of not returning, and what should we do about it?

The difference in the answers may have very little to do with the intelligence of the AI model. The difference is context.

One system knows that Jane registered. The other understands Jane's relationship with the organization.

As AI models become increasingly accessible, that distinction becomes more important. The model itself may become less differentiating while the proprietary context available to it becomes more valuable.

Gartner's 2026 research reinforces this point. Organizations with the most mature AI-ready data and analytics capabilities reported up to 65% greater business outcomes, including revenue growth and cost optimization. Gartner increasingly describes context — including the semantics, metadata and relationships that help systems understand what data means — as critical infrastructure for AI.

In other words, AI does not create organizational context from nothing. It inherits the context your technology has already created.

Data Quality Matters. But It Isn't Enough.

For decades, technologists have warned about “garbage in, garbage out.” That principle certainly applies to AI. Duplicate records, incorrect information, inconsistent fields and missing transactions can all undermine the quality of AI output.

But focusing exclusively on data quality misses a much bigger part of AI readiness.

Your data can be completely accurate and still tell an incomplete story.

Imagine that your registration platform contains perfect information about every registration you've processed. The data is clean, accurate and current. But fundraising lives in another system. Volunteer activity lives somewhere else. Marketing engagement is stored in another tool. Merchandise purchases sit in an ecommerce platform. Customer history from a previous registration system wasn't fully migrated.

Nothing in those systems is necessarily wrong. The problem is that no single system understands the entire relationship.

This is why Gartner specifically cautions that traditionally defined “high-quality” data does not automatically equal AI-ready data. AI readiness also depends on whether the right data is available for the use case, how it is structured, whether it can be accessed and how much context accompanies it.

Accuracy tells AI whether a data point is correct.

Architecture helps AI understand what that data point means.

AI Needs Relationships, Not Just Records

Consider a $250 donation.

On its own, it tells you something happened: someone gave $250.

Now imagine the system also knows that the donor is a four-time participant, previously fundraised for the organization, volunteered last year, belongs to a membership program and recently engaged with three marketing campaigns.

The transaction hasn't changed. The meaning has.

That is what context does.

For AI to meaningfully understand an organization, it increasingly needs to understand the relationships between people, transactions, events, campaigns, teams, organizations, communications and historical activity.

That puts new importance on the architecture underneath the applications organizations use every day.

For years, software evaluations have focused heavily on functionality. Can the platform process registrations? Send emails? Accept donations? Manage volunteers? Sell merchandise?

Those are still important questions. But AI introduces another one:

Does the technology understand how all of those activities relate to the same customer?

The answer can dramatically change what AI is capable of doing with the information.

Moving Data Is Not the Same as Connecting It

Most organizations use multiple systems. That may be unavoidable, but it also makes the data problem significantly harder.

Every additional system creates another place where customer identity, history and activity can become disconnected. In theory, integrations can bridge those gaps. In practice, truly connecting systems is difficult.

Robust integration requires more than sending a few fields from one platform to another. It often requires technical resources, ongoing maintenance, data mapping, identity resolution, monitoring, error handling and sometimes custom development. Many organizations simply are not equipped, financially, technically or operationally, to maintain that level of integration across an increasingly complex technology stack.

The fortunate ones have strong technology partners who can help bridge those gaps in designing the integrations, maintaining them over time and helping ensure the data flowing between systems remains usable and trustworthy. 

As a result, what gets described as an “integrated” ecosystem is often much less connected than it appears. One system may receive only a subset of fields from another. Data may move in one direction. Synchronization may happen in batches. Historical context may never transfer. Different platforms may still maintain separate versions of the same customer.

Consider three different technology architectures.

• In the first, a registration platform sends an email address and registration status to a marketing system once each night.
• In the second, registration and marketing platforms maintain a robust two-way integration, continuously sharing relevant participant attributes, engagement and updates.
• In the third, registration, marketing activity, ecommerce, fundraising and CRM all operate around the same underlying customer record.

All three organizations might reasonably say their technology is “integrated.” But those architectures create very different levels of customer context.

A one-way integration moves data. A robust two-way integration allows systems to maintain significantly more shared context. A native data model can go further by removing the need to reconstruct portions of the customer relationship between separate applications in the first place.

That distinction matters enormously for AI because every disconnected system creates another boundary across which identity, history and context have to be preserved.

The more fragmented the technology stack, the more work is required to create a complete, reliable understanding of the customer before AI can do anything meaningful with it.

Fragmented Systems Can Create Fragmented Intelligence

Humans are remarkably good at compensating for fragmented technology.

An experienced team member knows that Jane Smith in the volunteer database is probably the same Jane Smith who participated last year. Someone remembers that a major donor also ran the marathon three years ago. The marketing team knows which spreadsheet contains the previous platform's history. Finance understands which report needs to be reconciled before anyone trusts the number.

Those connections exist inside people's heads.

AI cannot reliably depend on institutional memory that was never represented in the technology.

IBM's Rob Thomas described this challenge particularly well:

“Without that, your AI will just amplify the fragmentation that exists in many organizations.”

That is an important warning.

Putting an intelligent interface over fragmented systems can make fragmented information much easier to query. It does not necessarily make the underlying understanding of the organization more complete.

In fact, the better AI becomes at confidently answering questions and taking actions, the more dangerous incomplete context can become.

Technology debt can become intelligence debt.

History Is Part of the Context Too

The most valuable thing AI can know about a customer may not be what they did today. It may be how today's behavior compares with everything they have done before.

Consider a relatively simple question: Who is unlikely to return next year?

A system with twelve months of participant activity has limited evidence. A system with seven years of history can recognize participation frequency, changes in spend, event preferences, fundraising activity, engagement trends, lapses, returns and progression through different parts of the organization.

History transforms transactions into patterns.

And patterns are where AI becomes considerably more powerful.

That makes historical data migration an AI issue too.

When an organization changes platforms but brings over only contact information and a few summary fields, it hasn't merely left archived records behind. It has potentially reduced the context future AI systems can use to understand those customers.

Organizations evaluating technology should therefore be asking not only whether they own their historical data, but whether their operational platform can actually use it.

Can years of transactions and interactions become part of the persistent customer relationship? Or does meaningful history live somewhere else, disconnected from the systems expected to power future intelligence?

A CRM Helps, But the Architecture Around It Still Matters

CRM is particularly important in this conversation because a legitimate CRM is fundamentally designed around relationships over time rather than individual transactions.

But simply checking a “CRM” box doesn't solve the architecture problem either.

An organization should ask three different questions.

• First, do we have a true CRM? Does the system maintain a persistent customer record across interactions, activity and time, or does the platform primarily provide participant profiles, filters and reporting around registration transactions?
• Second, how much does that CRM actually know? If registrations, merchandise, donations, volunteer activity and other important engagement happen in other systems, which of those interactions actually become part of the customer record?
• Finally, how is the CRM connected to the operational platform? Is data imported manually? Does information flow one way? Is there comprehensive two-way synchronization? Or is CRM native to the same platform and underlying customer data?

HubSpot, Salesforce and other dedicated CRMs can be extremely powerful. But when CRM lives outside the operational platform, the integration between those systems becomes part of the organization's data architecture, and therefore part of its AI architecture.

AI can only reason across the context that architecture makes available.

The Better AI Gets, the More This Matters

It would be easy to assume that increasingly intelligent AI models will eventually overcome these architectural limitations.

Industry research suggests the opposite.

McKinsey argues that as organizations move AI from experimentation into scaled deployment, data is emerging as a constraint. AI systems increasingly retrieve and recombine information across documents, databases, applications and workflows, making the reliability and structure of the underlying data foundation more consequential, not less.

Deloitte has reached a similar conclusion. Its 2026 State of AI research says legacy data and infrastructure architectures cannot support the demands of real-time, autonomous AI. And in a separate 2026 study of organizations already piloting agentic AI, 72% of leaders said they lacked unified, accessible data, while 67% cited the cost and complexity of integration as a barrier.

Why is architecture becoming more important?

Because AI is moving from answering questions toward taking action.

The progression is increasingly:

Ask → Understand → Recommend → Act → Learn

If you ask AI which participants are unlikely to return, incomplete context may produce an incomplete answer.

But imagine giving it permission to identify those participants, build an audience, determine the appropriate message, launch a retention campaign and monitor the outcome.

Incomplete context now leads to incomplete action.

The more authority organizations give AI, the more confidence they need in what it knows.

Look Beyond the AI Demo

Soon, almost every enterprise software platform will have AI.

There will be assistants, agents, conversational analytics, recommendation engines and automated workflows. Many of them will be genuinely impressive.

That means the existence of AI itself will tell buyers less and less.

A better evaluation starts underneath the AI layer.

What data can the AI access? Does the platform understand a persistent customer relationship or primarily transactions? How much historical context is available? Are important activities connected to the same identity? How many systems does the AI need to cross to understand the organization? Are those integrations one-way or two-way? Is data available in real time? Can intelligence lead directly to action?

Those questions expose something an AI demo cannot:

How much of your organization does the AI actually understand?

And organizations are beginning to invest accordingly. Gartner found that organizations reporting successful AI initiatives invest up to four times more in foundational areas including data quality, governance, AI-ready people and change management than organizations reporting poor AI outcomes.

The AI race isn't only a race to deploy the best model.

It is also a race to build the strongest foundation underneath it.

AI Is the Tool. Data Is the Fuel.

Models will keep improving. Agents will become more capable. Features that seem extraordinary today will increasingly become standard.

But AI will always need something to be intelligent about.

Your customers. Their history. Their relationships. Their transactions. Their behavior. Your operations. Your organization.

The data and architecture underneath AI determine how much of that picture it can actually see.

So before asking whether a technology platform has AI, ask what may ultimately be the more important question:

What does its AI actually know and what does the architecture underneath it allow it to understand?

AI is the tool. Data is the fuel. 

And your architecture determines how far it can take you.