TL;DR: Automating existing processes with AI without rethinking them is like running faster on a treadmill. You burn more energy but go nowhere. Real progress starts with reimagining workflows, not speeding up flawed ones. Trust, delivered at run time through cohesive data modeling, governance, and quality capabilities is what makes AI outputs reliable and AI initiatives scalable.
I’ve spent enough time in data kitchens to know a five-star setup from a fire hazard.
In the past year, I’ve sat with teams at some of the largest organizations in financial services, healthcare, and manufacturing. Organizations’ AI ambition is enormous. The data foundation underneath it is not. Teams are skipping the metadata, skipping the data modeling, skipping the governance, all in a race to keep up with what the business is demanding from AI.
Here’s the problem with that approach. If you’re just using AI on your current processes, you’re just speeding up what you’re delivering today. You’re not reinventing anything. You’re not looking at the use cases and outcomes you’re trying to achieve in a new way. Speed without reimagination isn’t transformation.
You can’t sacrifice quality and safety for speed. And yet that is precisely what’s happening. You wouldn’t do that in your own kitchen – you’d lose a finger chopping carrots too fast.
As Stephen M. R. Covey stated in The Speed of Trust, “When trust goes down, speed will also go down and costs will go up. When trust goes up, speed will also go up and costs will go down.”1 That equation describes exactly what I see playing out in AI programs right now. The organizations investing in data trust are moving faster and spending less. The ones skipping it are paying for that shortcut on the back end.
” When your data lacks context, governance, and a verifiable trust level, AI acts out.”
Sue Laine, Global Data CTO at Quest Software, on why skipping data foundations to chase AI speed is a bet organizations keep losing.
Why are AI projects still failing at such high rates?
The data is stark. Gartner predicts that “through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data,”2 and organizations with “successful AI initiatives invest up to four times more (as a percentage of revenue) in foundational areas, such as data quality, governance, AI-ready people and change management.”2 The organizations pulling ahead aren’t buying better models. They’re building better foundations.
Look at what it takes today. A data modeling team—anywhere from five to fifty people, depending on the size of the organization—defines a new data domain. A separate metadata team then spends another four to eight weeks tagging, cataloging, and tracing lineage for that same domain. Your data quality group profiles it with its own tools and its own definition of “good,” rarely comparing notes with the other two teams. Finally, business stewards step in to define what it all means in plain language. All that effort, spread across four disconnected teams, and the question at the end of the day is still the simplest one: do you trust the data?
When your data lacks context, governance, and a verifiable trust level, AI acts out. It looks at raw numbers and makes pattern-based decisions with no grounding in your business reality. Legal and compliance teams are right to push back. Your end users are right not to trust the outputs. The system isn’t broken because the technology failed. It’s broken because the data foundation was never built.
Gartner’s 2026 Data & Analytics Summit reinforced this directly: the essential ingredients of trust must be built into AI systems from the start, not added as an afterthought. The organizations that treat trust as a design principle rather than a retrofit are the ones with real AI results to show.
“The point-to-point tool model that has governed data management for 20-plus years was never designed to deliver cohesive, trusted, AI-ready data at the pace AI now demands.”
Why siloed data teams can’t close the gap on their own
Siloed teams keep working in isolation because the tools they have been handed were purpose built for their modeling, quality, MDM, integration, and governance practice. The mismatch is not just organizational. It is generational. A new generation of data practitioners, analytics engineers, data engineers, product-minded builders, expects modeling to work the way the rest of their stack works: workflow-first, embedded in the development lifecycle, and iterated on at the speed of dbt and CI/CD pipelines.
This is not a people problem. It is a structural one. The point-to-point tool model that has governed data management for 20-plus years was never designed to deliver cohesive, trusted AI-ready data at the pace AI now demands.
The answer is not more teams or more tools. It is a unified practice, where modeling, contextualizing, cataloging, governance, lineage, and quality operate in concert, and where the output of that practice is not a report or a dashboard but a trusted, reusable data product.
“Without a shared, governed definition layer, consistency erodes the moment a new tool, team, or AI system touches the data, and the trust score becomes the symptom rather than the cause.”
What does a data trust score actually measure?
For a long time, trust in data meant one thing: did it pass the data quality profiling run? That definition is too narrow for what AI requires.
For years, protecting that quality required grinding manual effort: stewards tagging metadata by hand, catalogs lagging behind reality, and quality checks running in silos no one else could see. That work is now shifting to agents that continuously trace and observe the data, flagging drift the moment it appears. The agents do not make the final call. A person still has to accept or reject what gets flagged—because that human judgment is what the trust score ultimately rests on. Underneath all of it is a simpler question: does “customer” or “revenue” mean the same thing in this data product as it does everywhere else in the organization? Without a shared, governed definition layer, consistency erodes the moment a new tool, team, or AI system touches the data, and the trust score becomes the symptom rather than the cause.
A practical data trust score model lets you define what trust means for your organization, apply weights to each dimension, and classify data assets into tiers: think gold, silver, and bronze. Consider it a nutrition label for your data: you always know what you are working with before you commit to using it. Gold means high trust, high reuse, well-governed. Silver means useful but with caveats. Bronze means proceed with caution.
This kind of explicit, visible trust scoring is what makes AI outputs defensible. It is the difference between an organization that can scale AI confidently and one that keeps restarting pilots that never reach production. If you have not yet defined what trust looks like for your own data, that is the starting point. The question that follows it is how you build at the pace that level of trust demands.
How does a data product factory change delivery timelines?
A data product is a curated, self-contained collection of data assets wrapped in semantic context, governance guardrails, and a trust score. It is ready for consumption. You do not need to hunt for it, verify it manually, or rebuild it from scratch the next time someone needs it. A trust score tells you what to trust. A data product factory tells you how to build it at scale. And if you can find a way to create an automated data product factory, then you’re working at the speed and scale that AI deserves, with trust as the underlying foundation to accelerate your work.
There are four maturity levels to work toward.
- Raw data products are foundational feeds serving as inputs to other products.
- Master data products consolidate datasets around core domains such as customer, product, or revenue into a reusable source of truth.
- Insight-based data products match specific outcomes to specific datasets.
- Composite data products let you assemble multiple trusted products into a curated package purpose-built for a new AI model or use case.
From prompt to trusted data product, Sue Laine discusses how data product development is automated.
Picture requesting a new data product for fraud detection, tied to a specific regulatory compliance initiative. The system first searches your enterprise logical model for the relevant entities and attributes. If it finds no match, it discovers the logical definition on its own before creating the specification—so you are no longer forced to treat enterprise data modeling as a continuous catch-up exercise. Once the specification is in place, the system locates the physical data behind it and wraps the entire package in your governance guardrails. What remains is a secure, trusted, reusable data product, delivered in days rather than the weeks it once required.
Why reuse is the compounding advantage most teams are leaving behind
There’s a saying in our industry that the fastest path from data to dollars is a data product. That point is easy to underestimate. A trusted data product is reusable, so teams stop recreating the same work across the organization. It is measured and monitored for value, so it does not quietly go stale on a shelf. It earns a trust score the moment people begin using and rating it, and it is accessible and composable enough to be found in a marketplace and assembled into something new. Most of the organizations I work with are already putting that logic into production, deploying these data products directly into their Snowflake, Databricks, or Microsoft environments.
The data product does not sit on a shelf and go stale. It lives in a data marketplace where your teams can find it, compare alternatives, interact, and understand its trust level before committing. That kind of transparent, accessible collaboration is what gets end users to engage with governed data instead of going around it.
When you are assembling a new AI model that needs five different data domains, you are creating them from trusted data products that already carry their context, governance, and trust scores with them.
What the semantic layer makes possible
Get specific about what that semantic foundation actually contains. First is the semantic layer itself—the logical representation of your data. Sitting on top of it is an ontology layer that captures the business context and vocabulary unique to your organization. Finally, there is a contextual data layer that supplies the capacity to generate the data product itself on demand. When you run your LLMs against all three together, they stop guessing at what “customer” or “revenue” means and begin answering in your organization’s own language.
When your AI has access to a semantic layer reflecting your enterprise’s logical definitions and data product catalog, it can filter for authoritative sources, apply privacy controls, and align outputs to your business objectives. Gartner identifies advancements in semantics as one of the top three forces reshaping the industry this year.3 The semantic foundation is not optional. It is the infrastructure that makes AI outputs trustworthy at scale.
The path forward
The organizations that close the trust gap first will scale AI confidently while others are still debugging outputs and managing skeptical stakeholders.
Three things will move you forward.
- Stop treating data management as a pre-condition you satisfy once. It is a continuous capability that needs to be embedded into every AI initiative from the start which is the creation of data.
- Invest in a unified data management platform that connects data modeling, cataloging, governance, quality, and lineage in a single operating environment. Each of those capabilities in isolation gives you a partial picture. Connected, they give you a trust score you can act on. The fragmented tool stack was never designed to deliver that, and it never will.
- Build with an automated data product factory and treat trusted data products as the fundamental unit of AI delivery. Teams that start now will have a meaningful head start.
The data you need is almost certainly already in your organization. The question is whether you can find it, trust it, and use it fast enough to matter.
Sources:
- Covey, Stephen M. R., The Speed of Trust: The One Thing That Changes Everything, Free Press, 2006
- Gartner, “Lack of AI-Ready Data Puts AI Projects at Risk,” February 2025
- Gartner, “Top Trends in Data and Analytics for 2026,” February 2026
GARTNER is a trademark of Gartner, Inc. and/or its affiliates.
Gartner does not endorse any company, vendor, product or service depicted in its publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner publications consist of the opinions of Gartner’s business and technology insights organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this publication, including any warranties of merchantability or fitness for a particular purpose.
