The Most Durable AI Investment: A Strong Data Foundation
The future of AI is uncertain. The value of understanding your data isn’t.
Jasmine Libert
Senior VP Growth and GTM
"I like to think about this stuff in my free time. "
It’s an interesting time to be making business decisions about AI.
The technology is advancing quickly, and AI capabilities are appearing in tools many of us, especially in marketing, already use every day. At the same time, the conversation around AI is becoming more complicated. Communities are questioning the energy, water and infrastructure demands of new data centers. Governments are working through how AI should be regulated. Businesses are dealing with cost implications, privacy, governance and security questions. Recent safety research has documented AI models behaving in concerning ways under experimental conditions, including deception, blackmail and attempts to circumvent controls. There have also been real-world incidents during cybersecurity testing where models gained unauthorized access to systems outside their intended environments. And some of the people leading the development of frontier AI are now publicly arguing that the industry needs to slow down enough for safety and oversight to keep pace.
None of this tells us exactly where AI is headed. It does make one thing clear: organizations have good reason to look carefully at where AI creates meaningful value, what they're comfortable asking it to do, and what needs to be in place before they do more.
We don’t know exactly how those concerns will play out. Regulation will evolve. Technology will change. Some AI applications will prove enormously useful, while others will probably have a much shorter shelf life.
But organizations still have decisions to make now.
One of the more useful ways to think about readiness in a period like this is to invest in things that will remain valuable across different versions of the future.
Data is one of them.
What makes data ready for AI?
AI-ready data is accurate, accessible, connected and supported by enough context and governance to be used appropriately for a specific AI application. It includes clear definitions, known ownership, reliable relationships between systems and people who can evaluate whether the resulting output makes sense.
Good data gives you options
Data strength isn’t particularly exciting.
It takes meticulous QA. It takes health checks, documentation and governance. It takes an understanding of what a business cares about, how its customers behave and how its teams operate. And it takes technical expertise to make sure the signals you think you’re collecting are the signals you’re really collecting.
There isn’t a plug-and-play tool that makes all of that disappear.
There is something reassuring about this work, though: its value doesn’t depend on correctly predicting what happens next.
High-quality data improves measurement and customer understanding. It supports more reliable reporting and better activation. It gives people stronger evidence for decisions. And yes, it creates a stronger foundation for AI.
If AI capabilities look very different five years from now, that work still has value. If regulation changes how AI can be used, it still has value. If privacy expectations or the signals available to marketers change again, it still has value. If the next important shift in marketing technology has very little to do with generative AI, it still has value.
Good data gives organizations options.
What counts as “good data” is changing
For years, much of the conversation about data quality in marketing has centred on accuracy, completeness and consistency. Those things still matter enormously.
But they aren’t the whole story anymore.
A perfectly maintained Google Analytics implementation is useful. It can give you reliable information about what people are doing across your digital properties. Your CRM might contain an equally reliable view of customer relationships. Media platforms understand campaign exposure and performance. Commerce systems know about transactions. Product and service systems contain even more of the story.
Each of those systems can contain perfectly good data while the organization as a whole still sees only fragments of the customer and the business.
That has always created limitations, and AI makes those limitations harder to ignore.
When we ask AI to identify patterns, explain performance, personalize an experience or recommend an action, we’re asking it to do more than retrieve a value from a database. It needs enough information to understand what those values mean, how they relate to one another and which parts of the picture may be missing.
Adobe’s 2026 research illustrates the readiness gap. Only 44% of surveyed organizations said their data quality and accessibility were adequate for AI. More than half said their current data unification and structure limited their ability to advance AI initiatives, while 75% identified data integration and quality as a leading challenge for implementing agentic AI.
In some ways, AI is exposing weaknesses in our data foundations that were already there.
Fragmented customer data was a problem before AI. So were inconsistent definitions, gaps in measurement, unclear ownership and systems that couldn’t easily share information. Organizations have learned to work around many of these limitations. People supply missing context, reconcile conflicting numbers, know which reports to trust and understand the business well enough to recognize when something doesn’t look right.
As we ask AI to do more of the interpreting, analyzing and recommending, those workarounds become harder to rely on. The quality of the underlying data matters, but so does the context around it: how information relates, what it means, where it came from and whether it can appropriately be used.
That makes AI readiness a useful lens for looking at the health of the broader data environment. The work that makes data more useful to AI, improving its quality, connecting relevant signals, clarifying definitions and strengthening governance, also makes that data more useful to the people trying to understand customers and make decisions.
And that’s an important distinction. If the value of this work depended entirely on what happens with AI, investing in it right now might feel like a bigger bet. It doesn’t. These are capabilities organizations need anyway.
Connection changes what data can tell us
Imagine a customer visits a website after seeing an ad. They research a product, leave, return later through email, make a purchase and eventually contact customer service.
There may be accurate information about every one of those interactions. But if those signals live in separate systems with no reliable way to relate them, each system knows only one part of what happened.
That matters for ordinary analysis. It matters even more when AI is being asked to make sense of the journey.
Connected data doesn’t mean every piece of information an organization owns needs to be moved into one enormous system. It means the relevant signals can be brought together when there is a reason to do so, with enough shared identity, definitions and business context to understand how they relate.
It also means knowing whether the data can appropriately be used in the first place.
Consent, permissions, ownership, privacy and governance are part of the foundation. So are the systems and workflows that make information available where it is useful. And so are the people responsible for deciding whether an output makes sense.
That last part deserves more attention than it sometimes gets.
Readiness includes knowing when to question the answer
As AI becomes easier to use, it becomes easier to get an answer.
That doesn’t necessarily make it easier to know whether the answer is a good one.
People still need enough understanding of their customers, their business and their data to recognize when something looks wrong. They need to know what information is represented, what might be missing and whether two metrics that sound similar actually mean the same thing. They need enough context to decide whether an AI-generated insight is useful, irrelevant or simply incorrect.
Those capabilities don’t come from installing a platform.
They come from understanding how the organization works and making that understanding available to the people and technology making decisions.
This is also why enablement matters. Expertise is more useful when it spreads. Documentation, shared definitions, training and knowledge transfer can feel secondary to the technical work, but they determine whether people can use the foundation well after it has been built.
The strongest data environment in the world has limited value if only a handful of people understand what’s in it.
Start with what you want AI to do
There’s another risk in the current environment: trying to become generically “AI ready”: Readiness depends on the intended outcome.
Using AI to help a marketing team explore campaign performance requires one set of conditions. Personalizing customer experiences requires another. An AI-supported analysis may depend heavily on reliable source data, consistent metrics and enough business context to interpret them. Personalization may place much greater demands on identity, connected customer signals, consent and real-time activation.
That makes the intended use a practical place to start.
What are we trying to accomplish? What information would be required to do it well? Can we trust that information? Can the relevant signals be connected? Do we understand what they mean? Are we permitted to use them this way? Can our technology make them available at the right moment? Do the people involved know how to evaluate the result?
Those questions tend to produce a much more useful picture of readiness than an inventory of AI tools.
They also help organizations distinguish between opportunities they can pursue now and those that need more groundwork. Sometimes the answer will be a new technology investment. Sometimes it will be an integration, a governance decision, better measurement or some decidedly unglamorous QA.
That’s useful information too.
Steady is a strategy
There is understandable pressure to move quickly when technology is changing this fast. Waiting for complete certainty isn’t realistic either. Organizations need room to experiment, learn and find out where AI is genuinely useful to them.
But speed and readiness aren’t the same thing.
There is value in doing the work that makes more possibilities available later: improving data quality, connecting the information that matters, clarifying definitions, strengthening governance, improving measurement and making sure people understand the systems they rely on.
None of this is particularly new. Much of it isn’t particularly exciting. It also doesn’t become obsolete every time the technology landscape changes.
AI gives us new reasons to care about the quality and context of our data. It raises the stakes when that foundation is weak. But the value of getting the fundamentals right extends well beyond AI.
Whatever comes next, organizations will still need to understand their customers. People will still need reliable information to make good decisions. And technology will still only be as useful as the understanding we’re able to build around it.
That seems like a pretty good place to invest.
Is your data ready for what you want AI to do?
Napkyn helps organizations assess the quality, connection, governance and usability of their marketing data. We can identify the gaps that matter to your intended AI use case and help you prioritize what to address first.
More Insights
Understanding the IAB TCF v2.3 Update: New Vendor Disclosure Requirements Explained
Rob English
Director, Data Solutions
Jun 17, 2026
Read More
How to Choose the Right ETL Tool for Your Data Pipeline
Cem Bakar
Cloud Architect
Jun 10, 2026
Read More
CM360 Path to Conversion Reporting: Full Customer Journey Attribution
Monika Boldak
Associate Director, Marketing
Jun 3, 2026
Read More
More Insights
Sign Up For Our Newsletter
Napkyn Inc.
204-78 George Street, Ottawa, Ontario, K1N 5W1, Canada
Napkyn US
6 East 32nd Street, 9th Floor, New York, NY 10016, USA
212-247-0800 | info@napkyn.com
The Most Durable AI Investment: A Strong Data Foundation
The future of AI is uncertain. The value of understanding your data isn’t.
Jasmine Libert
Senior VP Growth and GTM
September 28, 2026
"I like to think about this stuff in my free time. "
It’s an interesting time to be making business decisions about AI.
The technology is advancing quickly, and AI capabilities are appearing in tools many of us, especially in marketing, already use every day. At the same time, the conversation around AI is becoming more complicated. Communities are questioning the energy, water and infrastructure demands of new data centers. Governments are working through how AI should be regulated. Businesses are dealing with cost implications, privacy, governance and security questions. Recent safety research has documented AI models behaving in concerning ways under experimental conditions, including deception, blackmail and attempts to circumvent controls. There have also been real-world incidents during cybersecurity testing where models gained unauthorized access to systems outside their intended environments. And some of the people leading the development of frontier AI are now publicly arguing that the industry needs to slow down enough for safety and oversight to keep pace.
None of this tells us exactly where AI is headed. It does make one thing clear: organizations have good reason to look carefully at where AI creates meaningful value, what they're comfortable asking it to do, and what needs to be in place before they do more.
We don’t know exactly how those concerns will play out. Regulation will evolve. Technology will change. Some AI applications will prove enormously useful, while others will probably have a much shorter shelf life.
But organizations still have decisions to make now.
One of the more useful ways to think about readiness in a period like this is to invest in things that will remain valuable across different versions of the future.
Data is one of them.
What makes data ready for AI?
AI-ready data is accurate, accessible, connected and supported by enough context and governance to be used appropriately for a specific AI application. It includes clear definitions, known ownership, reliable relationships between systems and people who can evaluate whether the resulting output makes sense.
Good data gives you options
Data strength isn’t particularly exciting.
It takes meticulous QA. It takes health checks, documentation and governance. It takes an understanding of what a business cares about, how its customers behave and how its teams operate. And it takes technical expertise to make sure the signals you think you’re collecting are the signals you’re really collecting.
There isn’t a plug-and-play tool that makes all of that disappear.
There is something reassuring about this work, though: its value doesn’t depend on correctly predicting what happens next.
High-quality data improves measurement and customer understanding. It supports more reliable reporting and better activation. It gives people stronger evidence for decisions. And yes, it creates a stronger foundation for AI.
If AI capabilities look very different five years from now, that work still has value. If regulation changes how AI can be used, it still has value. If privacy expectations or the signals available to marketers change again, it still has value. If the next important shift in marketing technology has very little to do with generative AI, it still has value.
Good data gives organizations options.
What counts as “good data” is changing
For years, much of the conversation about data quality in marketing has centred on accuracy, completeness and consistency. Those things still matter enormously.
But they aren’t the whole story anymore.
A perfectly maintained Google Analytics implementation is useful. It can give you reliable information about what people are doing across your digital properties. Your CRM might contain an equally reliable view of customer relationships. Media platforms understand campaign exposure and performance. Commerce systems know about transactions. Product and service systems contain even more of the story.
Each of those systems can contain perfectly good data while the organization as a whole still sees only fragments of the customer and the business.
That has always created limitations, and AI makes those limitations harder to ignore.
When we ask AI to identify patterns, explain performance, personalize an experience or recommend an action, we’re asking it to do more than retrieve a value from a database. It needs enough information to understand what those values mean, how they relate to one another and which parts of the picture may be missing.
Adobe’s 2026 research illustrates the readiness gap. Only 44% of surveyed organizations said their data quality and accessibility were adequate for AI. More than half said their current data unification and structure limited their ability to advance AI initiatives, while 75% identified data integration and quality as a leading challenge for implementing agentic AI.
In some ways, AI is exposing weaknesses in our data foundations that were already there.
Fragmented customer data was a problem before AI. So were inconsistent definitions, gaps in measurement, unclear ownership and systems that couldn’t easily share information. Organizations have learned to work around many of these limitations. People supply missing context, reconcile conflicting numbers, know which reports to trust and understand the business well enough to recognize when something doesn’t look right.
As we ask AI to do more of the interpreting, analyzing and recommending, those workarounds become harder to rely on. The quality of the underlying data matters, but so does the context around it: how information relates, what it means, where it came from and whether it can appropriately be used.
That makes AI readiness a useful lens for looking at the health of the broader data environment. The work that makes data more useful to AI, improving its quality, connecting relevant signals, clarifying definitions and strengthening governance, also makes that data more useful to the people trying to understand customers and make decisions.
And that’s an important distinction. If the value of this work depended entirely on what happens with AI, investing in it right now might feel like a bigger bet. It doesn’t. These are capabilities organizations need anyway.
Connection changes what data can tell us
Imagine a customer visits a website after seeing an ad. They research a product, leave, return later through email, make a purchase and eventually contact customer service.
There may be accurate information about every one of those interactions. But if those signals live in separate systems with no reliable way to relate them, each system knows only one part of what happened.
That matters for ordinary analysis. It matters even more when AI is being asked to make sense of the journey.
Connected data doesn’t mean every piece of information an organization owns needs to be moved into one enormous system. It means the relevant signals can be brought together when there is a reason to do so, with enough shared identity, definitions and business context to understand how they relate.
It also means knowing whether the data can appropriately be used in the first place.
Consent, permissions, ownership, privacy and governance are part of the foundation. So are the systems and workflows that make information available where it is useful. And so are the people responsible for deciding whether an output makes sense.
That last part deserves more attention than it sometimes gets.
Readiness includes knowing when to question the answer
As AI becomes easier to use, it becomes easier to get an answer.
That doesn’t necessarily make it easier to know whether the answer is a good one.
People still need enough understanding of their customers, their business and their data to recognize when something looks wrong. They need to know what information is represented, what might be missing and whether two metrics that sound similar actually mean the same thing. They need enough context to decide whether an AI-generated insight is useful, irrelevant or simply incorrect.
Those capabilities don’t come from installing a platform.
They come from understanding how the organization works and making that understanding available to the people and technology making decisions.
This is also why enablement matters. Expertise is more useful when it spreads. Documentation, shared definitions, training and knowledge transfer can feel secondary to the technical work, but they determine whether people can use the foundation well after it has been built.
The strongest data environment in the world has limited value if only a handful of people understand what’s in it.
Start with what you want AI to do
There’s another risk in the current environment: trying to become generically “AI ready”: Readiness depends on the intended outcome.
Using AI to help a marketing team explore campaign performance requires one set of conditions. Personalizing customer experiences requires another. An AI-supported analysis may depend heavily on reliable source data, consistent metrics and enough business context to interpret them. Personalization may place much greater demands on identity, connected customer signals, consent and real-time activation.
That makes the intended use a practical place to start.
What are we trying to accomplish? What information would be required to do it well? Can we trust that information? Can the relevant signals be connected? Do we understand what they mean? Are we permitted to use them this way? Can our technology make them available at the right moment? Do the people involved know how to evaluate the result?
Those questions tend to produce a much more useful picture of readiness than an inventory of AI tools.
They also help organizations distinguish between opportunities they can pursue now and those that need more groundwork. Sometimes the answer will be a new technology investment. Sometimes it will be an integration, a governance decision, better measurement or some decidedly unglamorous QA.
That’s useful information too.
Steady is a strategy
There is understandable pressure to move quickly when technology is changing this fast. Waiting for complete certainty isn’t realistic either. Organizations need room to experiment, learn and find out where AI is genuinely useful to them.
But speed and readiness aren’t the same thing.
There is value in doing the work that makes more possibilities available later: improving data quality, connecting the information that matters, clarifying definitions, strengthening governance, improving measurement and making sure people understand the systems they rely on.
None of this is particularly new. Much of it isn’t particularly exciting. It also doesn’t become obsolete every time the technology landscape changes.
AI gives us new reasons to care about the quality and context of our data. It raises the stakes when that foundation is weak. But the value of getting the fundamentals right extends well beyond AI.
Whatever comes next, organizations will still need to understand their customers. People will still need reliable information to make good decisions. And technology will still only be as useful as the understanding we’re able to build around it.
That seems like a pretty good place to invest.
Is your data ready for what you want AI to do?
Napkyn helps organizations assess the quality, connection, governance and usability of their marketing data. We can identify the gaps that matter to your intended AI use case and help you prioritize what to address first.
More Insights
The Most Durable AI Investment: A Strong Data Foundation
Jasmine Libert
Senior VP Growth and GTM
Sep 28, 2026
Read More
How to Detect Google Analytics Tracking Issues Early with Automated QA
Skylar van Dalen-Flude
Senior Data Analyst
Aug 26, 2026
Read More
Why GA4, Google Ads, and BigQuery Numbers Don’t Match
Chris Wright
Associate Director, Training and Development
Jul 15, 2026
Read More
More Insights
Sign Up For Our Newsletter