How Do You Know When You’re Ready for AI?
It’s an interesting question. After doing extensive research on AI readiness, I found myself trying to understand what the experts had to say on the matter. One surprising statistic from the Rand Corporation is that 80% of artificial intelligence (AI) projects fail. Similarly, a study from Deloitte shows that close to 70% of generative AI (GenAI) projects stall in proof-of-concept, lacking the ability to scale. According to a Wharton study, over 50% of executives believe that they are not ready for AI. This suggests that there are few true experts when it comes to real-world AI implementations, especially at scale. However, reflecting on my own experience leading a large-scale AI initiative, I realized that perhaps my insights might help others to assess their AI readiness.
Are You Ever Truly Ready?
The concept of readiness can be tricky. It reminds me of questions we face in life: when to change careers, have children, or get married. We often think we’ll know when we’re ready, but in reality, readiness is often only clear after we’ve taken the leap. When I had my first child, I thought I was fully prepared. I bought the safest stroller, and I even opted for cloth diapers.
But once life kicked in, I realized that no matter how prepared I thought I was, there were many things I couldn’t have anticipated—like the bulky stroller that barely fit in my car or the impracticality of cloth diapers for my lifestyle. The same is true for AI. Preparation is important, but over-preparing can paralyze you. The key is to stay open to change and embrace the iterative nature of innovation. AI is a journey, not a one-time decision.
A Journey into AI
In 2015, I had the opportunity to be part of an exciting AI journey. This was before MLOps, DataOps, and other related “Ops” were around. That was the era of big data, or the inception of this current data and analytics era. This was when companies started looking to get more value and competitive advantage out of data. This was when organizations that invested in AI at scale were pioneers, and the failure rate was probably even higher than the 85% projected as recently as 2022. My colleagues and I were embarking on something exciting, yet we were full of uncertainty, as this our entrée to AI as a business transformation agent.
At that time, I was working for a global leader in the food flavor industry, and I was given the opportunity to be the technical lead for this initiative. Our leadership wanted to use AI to enhance our product development, so we could stay competitive in the market. We wanted to create an AI system that could work alongside developers, helping them to be both more creative and faster. This vision seemed far out of our reach, and by many accounts, we were not “ready.” But we had a mission and leadership support to make it happen. Many companies that attempted similar AI initiatives were experiencing large losses due to failed projects. Today, unfortunately, the story remains much the same, as many AI and digital transformation projects are struggling.
Despite the challenges we faced, we successfully implemented our AI project within a year of starting development, and we continuously enhanced and adapted it over the years. As I reflect on our journey, I’m reminded of the key drivers of our success:
1. Choose a Project that Demonstrates Clear Costs and Value
When tackling your first AI project, be prepared for changes to your business processes, ways of thinking, and the costs that come with adopting new technologies. Building and implementing transformational AI is about transforming aspects of your enterprise. The value can be improved competitiveness, increased customer intimacy, extended global knowledge and expertise, better quality products, optimized processes etc., all of which require varying methods of measurement. Understanding the value AI can bring, and how long it will take to realize that value, is crucial. You should also carefully weigh the costs and benefits of external expertise against in-house resources.
AI projects must be guided by a clear understanding of ROI. Build mechanisms into the project to measure the impact and ensure that value is realized as expected. Using a logical data fabric enabled by the Denodo Platform (see #4 below), we were able to gain continuous insight into many measures such as user adoption, adherence to processes, development cycle times, product assessment, material usage, products inspired by AI, sales and much more.
We were also able to review and measure how well our users were following business process guidelines. Our users understood that it was important to understand data as you produce it, if you want your AI models to learn properly. Our logical data fabric empowered our data scientists and analysts to quickly adapt metrics and measures as the project progressed. We were able to tie these metrics to goals so managers and end users could have insight on-demand.
We were also able to measure the output from our models by running simulations in which the data was immediately accessible for evaluation. This helped us to guide how we could enhance and improve our models over time.
These types of projects are iterative in nature, and to gain business value from them, you have to make them part of your business processes. Measuring the many aspects of such a project are therefore crucial for success and continued improvement.
2. Leadership Support Is Critical
When leadership is fully behind an initiative, they can clear obstacles, allocate resources, and drive progress. Leadership is also critical in driving focus on the mission and scope and evaluating when either might need adjustment. Leadership helps with communication and driving change management throughout the organization, to all areas touched by the transformation. Their commitment makes a huge difference in moving projects forward and enabling their timely success.
3. Foster a Collaborative, Multidisciplinary Team
Our project’s success was largely due to our cohesive, highly collaborative team. We brought together leadership, external AI researchers, business domain experts, change management professionals, and IT specialists. These diverse perspectives and skills were essential. The business experts shared their knowledge of the industry, while IT provided the necessary infrastructure, technology and data products. AI researchers translated this data into models that met our business goals. Frequent, rapid collaboration enabled us to move quickly while keeping everyone aligned on our objectives. Our communications were in the language of the business especially when data was involved.
4. Leverage Abstraction Through a Logical Data Fabric for Agility
In my experience, there is an overwhelming belief that the biggest challenge with AI lies in model creation. It certainly has its fair share of challenges, as talented AI researchers must translate business processes into mathematical models and interpret them. But for this activity to even begin, someone must translate activities and processes in the enterprise into understandable components that reflect them. The task is then to continuously deliver this information, regardless of the complexity of the enterprise.
Why a logical data fabric?
One of the most crucial technical approaches we employed was leveraging abstraction through a logical data fabric. This approach enabled us to establish a unified semantic layer above our data sources, which automatically translates all data into the language of the business. We provided AI researchers and business users with consistent enterprise data models, even as our enterprise environment evolved. The idea of abstraction is rooted in software engineering, where encapsulation and modularity are key for ensuring that changes in one part of the system don’t disrupt the whole. As the enterprise data architect, I researched and sought out this logical approach because I understood that technology that empowered these principles would help us to keep pace with the demands of the project.
The fast-changing data landscape posed a significant challenge. We had to maintain the integrity of the information we shared while delivering “new” data rapidly for AI models. By creating a logical data fabric, we established an abstraction layer that enabled us to manage and deliver data on demand. This layer insulated us from changes in the underlying systems and business processes, giving us agility while ensuring the consistency and stability required for AI research. Using the Denodo Platform’s logical data fabric as our abstraction layer empowered us to maintain real-time, on-demand data access in the language of the business, despite the dynamic nature of our data sources.
This approach enabled us to meet the demands of an innovative AI project while keeping our existing system functional. Abstraction made it possible to provide consistent, reliable information, which was key to the success of both the AI models and the overall project.
Business Processes Drive Data Quality
Data quality is critical in AI projects, but it’s important to understand that data quality reflects the processes that generate it. If business processes are inefficient, inconsistent, or filled with workarounds, your data will reflect those issues. In our experience, fixing data quality challenges downstream—such as in data pipelines, extract, transform, and load (ETL) jobs, or other transformation mechanisms—can be costly and time-consuming. Instead, addressing these issues at the source, by improving business processes, is far more efficient, when possible.
This is where leveraging a logical data fabric became invaluable. By implementing this abstraction layer, we were able to give business users direct insight into the data their processes were producing. With this real-time view, business users could see the gaps and inefficiencies in their processes that they hadn’t been aware of. For instance, many users thought their processes were streamlined, but the data revealed inconsistencies and inefficiencies they hadn’t anticipated. We were also able to assess how well people were adhering to established processes so we could determine if we needed to employ appropriate mechanisms.
With this newfound visibility, users were empowered to address these issues at the process level, improving both the quality of the data and the effectiveness of their operations. The logical data fabric enabled business users to interact with the data in a way that made sense to them—reflecting their own language and workflows. As a result, they were able to make informed decisions about where adjustments needed to be made. This ability to address problems early, based on concrete data insights, significantly improved the quality of the data being fed into our AI models, ultimately providing more accurate results.
In short, the logical data fabric didn’t just provide data; it provided actionable insights into business processes, enabling continuous improvement and alignment with the organization’s operations. This capability became a key driver of both data quality and overall project success.
Agility is Key
Time is almost always at a premium, whether you use internal or external resources. The ability to be agile enables you to adapt quickly, reducing costs and time to value. Data drives AI projects and the ability to identify, agree on, integrate, and deliver it quickly, consistently and simply regardless of where it was or how it was stored was critical for our AI researchers to iterate through model development. With the abstraction provided by our logical data fabric, we were able to create contracts for data exchange, enabling our teams to work independently while keeping all parts of the project on track.
Early in the project, for example, we needed to incorporate an external data source. Using our logical data fabric, we evaluated several options and selected a service within a week. This approach also enabled us to integrate external information with our internal data efficiently, quickly turning it into valuable data products. Additionally, we could prototype new features for testing in our models before committing them to system-wide production changes. This agility in data management supported faster experimentation and iteration, proving invaluable to the project’s overall success.
Strategically Creating Reusable Data Products
During our project, we built numerous data products that represented entities within the enterprise. These products, created through the logical data fabric, weren’t just usable by the AI project—they were designed with reuse in mind. This strategic approach to data management enabled us to support other projects, such as analytics and integration, by reusing the same data products. By building reusable artifacts, we created efficiencies that paid off in the long run. These efficiencies extended to our ability to both enhance, extend and support these products.
Long Term Impact
This AI initiative made a positive impact on our organization—and it is still evolving to this day. The AI “colleague” learned to provide new, creative ideas that helped to reduce development cycles as well as product costs. Being one of the early adopters of AI induced customer intimacies and attracted new developers.
Are You Ready for AI?
Is anyone ever truly ready for it? Perhaps not entirely. But the key is to begin. Be prepared for the challenges, embrace the innovations, and focus on creating value. Leveraging abstraction through a logical data fabric can provide that agility, as well as the consistency and resilience needed to navigate the evolving demands of an AI project. Part of the journey is assessing your data management approach and refining it when necessary. Success in AI isn’t about having all the answers from the start—it’s about building the right foundation and adapting as you go.
Originally published in DataManagement Blog November 28, 2024
About The Author
Terry Dorsey
Terry Dorsey is an enterprise architect, AI and data strategist, researcher, educator, and industry thought leader with extensive experience turning emerging technologies into practical enterprise capabilities. Her career spans enterprise architecture, data and integration strategy, AI operationalization, business transformation, and technology leadership. As a practitioner, she has designed and led enterprise capabilities supporting global business operations and served as an AI technical strategist, bridging AI research with the data, processes, architecture, infrastructure, and organizational capabilities required for real-world adoption. Today, Terry works with business and technology leaders on enterprise data and AI strategy while continuing her research and graduate-level teaching. She is a frequent conference speaker and webinar presenter and has contributed through interviews, industry publications, research, and blogs. Her work increasingly focuses on a fundamental question: how can enterprises better represent and organize their information, meaning, relationships, behaviors, and context so that people, applications, analytics, and AI can use them more effectively?
The idea of readiness really resonates with me, especially when it comes to big life decisions and AI projects. I agree that over-preparing can sometimes hold us back—sometimes you just have to dive in and learn along the way. But how do you find the right balance between preparation and action when diving into something as complex as AI? It’s fascinating that your team succeeded despite initial doubts—was leadership support the biggest factor, or was it something else? The part about failed projects in other companies makes me think AI success might be more about adaptability than perfect planning. I also wonder, what would you say was the biggest unexpected hurdle your team faced after starting? The stroller analogy is perfect—real-life chaos has a way of humbling even the best-laid plans!
*(Based on the input text, I responded in English since that was the original language. The comment encourages engagement by asking specific, thought-provoking questions about the challenges and mindset behind AI projects.)*
Thank you for your comment and question. I think most practitioners feel this struggle of balancing risk with the need to explore and prepare.
In my experience, success takes something close to a perfect storm of support: executive sponsorship, the people whose work will be affected, those with AI expertise and creativity, and those of us responsible for making sure the broader enterprise environment supports how everyone needs to communicate and operate.
AI in the enterprise is not a one-and-done effort. It is highly iterative and exploratory. We had a high-level goal, an idea of how we wanted our processes to change, and a clear sense of the outcomes we wanted to achieve. Those remained relatively consistent. The path to get there did not. The ability to be agile, learn, and adapt was critical.
One of our biggest initial hurdles was learning how all of the core participants needed to work together, communicate, and develop a shared understanding of the vision. Because for AI to really matter, it ultimately has to change how people work.
Perhaps that is another series in itself: the many things we had to work through to become a cohesive team around AI.
But to your point, without executive and upper-management support, and a genuine belief in experimentation, learning, and sometimes failing fast, I don’t think we would have been nearly as successful.
Your reflection on readiness and AI is truly thought-provoking. It’s fascinating how you draw parallels between personal life decisions and the complexities of AI implementation. I agree that over-preparing can sometimes hinder progress, but how do you strike the right balance between preparation and action? Your experience with the AI project is inspiring, especially how you adapted and improved it over time. It makes me wonder, what specific challenges did you face that you didn’t anticipate? Also, do you think the fear of failure often holds companies back from even starting their AI journeys? I’d love to hear more about how you managed to keep the team motivated despite the uncertainties. What advice would you give to someone just starting out in AI, especially in a competitive industry like food flavor development?
Thank you. Those are great questions, and I think many of them come back to the same lesson: readiness does not mean having the entire path figured out before you begin.
One of our biggest challenges was bringing together people with very different expertise, perspectives, and responsibilities and enabling them to develop a shared understanding of what we were trying to accomplish and how we needed to work together.
I cannot overemphasize how important the architecture was in making that possible. We deliberately worked from consistent logical models of the enterprise rather than asking our external researchers and internal community to work directly from the many concrete, persistent stores where the data happened to reside. That allowed everyone to speak the same enterprise language, align what we were building with our business processes, and work from a shared understanding while the underlying environment continued to evolve.
That foundation was also a significant part of what kept the team motivated. Once we could communicate consistently and align around the same processes and outcomes, we began making tremendous progress. And progress at that pace becomes infectious. People see what is possible, ideas start building on other ideas, and the team becomes increasingly willing to innovate and explore.
We had a clear vision of the outcomes we wanted, but not a predetermined path for getting there. We experimented, learned, adjusted, and built on what worked. Executive and management support was essential because it gave us the freedom to do that.
So I do think fear of failure can keep organizations from starting. But I also don’t think the answer is simply to start without adequate preparation. The balance is creating enough of the right foundation to enable action and then being willing to learn and adapt.
For someone beginning an AI journey, particularly in an innovative field like flavor development, I would start with the outcome rather than the technology. Bring together the people who understand the work, the people who understand AI, and the people who can create a shared enterprise foundation that allows everyone to work together effectively.
Expect the path to change. Build the foundation so that it can.
The idea of readiness is indeed complex, and your reflections really resonate. I’ve often found myself over-preparing, only to realize that life throws curveballs no matter how much I plan. Your experience with AI implementation is fascinating—it’s inspiring to hear how you embraced uncertainty and adapted along the way. I wonder, though, how did your team manage to stay so aligned despite the challenges? It seems like leadership support was crucial, but were there specific strategies you used to keep everyone motivated? I’m curious, what advice would you give to someone who feels like they’re “not ready” to start an AI project? And do you think the iterative approach you took could be applied to other industries, or is it unique to AI? Your story definitely makes me rethink my own approach to readiness—maybe sometimes it’s better to just start and figure things out as you go. What do you think?
Thank you. I agree that readiness is complex, but I would make one distinction: for us, it wasn’t simply a matter of starting and figuring everything out as we went. We were very intentional about creating a foundation that gave us the ability to learn and adapt.
A critical part of that foundation was working from consistent logical models of the enterprise rather than the concrete persistent stores where the data happened to reside. We used logical data abstraction to give our internal community and external researchers a shared representation of the enterprise. That allowed people with very different backgrounds to speak the same enterprise language, align around our processes and outcomes, and collaborate without everyone having to understand the complexity underneath.
That shared understanding was also a significant part of what kept the team motivated. We began making tremendous progress, and progress can be infectious. When people can see what is possible, they start thinking differently. Ideas build on other ideas, people become more creative, and innovation begins to accelerate.
Leadership support was absolutely critical because it gave us permission to experiment, learn, and change direction when necessary. But alignment came from more than leadership. It came from creating an environment in which people could understand one another and work toward the same outcomes.
So my advice to someone who doesn’t feel ready would not necessarily be “just start.” I would say: get ready enough to start, but prepare for change rather than trying to eliminate uncertainty. Establish the outcome you are trying to achieve, create a shared understanding of the business, bring the right people together, and build a foundation flexible enough to evolve as you learn.
And I don’t think that lesson is unique to AI. AI may make the need for iteration more visible, but the ability to establish a shared understanding, learn quickly, and adapt without continually starting over is valuable for almost any significant transformation.
The idea of readiness is indeed complex and often misunderstood. It’s fascinating how life’s biggest decisions, like starting a family or diving into AI, rarely come with a clear “ready” signal. Your experience with parenting and AI projects highlights how preparation is crucial, but overthinking can stall progress. It’s inspiring to hear how your team embraced the iterative nature of innovation and succeeded despite initial doubts. I wonder, though, how do you balance preparation with the need to act quickly in such fast-evolving fields? Your story makes me think about how often we let the fear of not being ready hold us back. What advice would you give to someone who feels paralyzed by the idea of not being fully prepared? Your perspective could really help others take that first step.
Thank you. I think the key distinction for me is between preparing for a predetermined path and preparing to adapt.
In our case, we certainly did not have everything figured out when we started, but we had a clear vision of the outcomes we wanted and created a foundation that allowed us to move quickly as we learned. A critical part of that was creating consistent logical models of the enterprise rather than tying the work directly to individual persistent data stores. That gave our internal teams and external researchers a shared enterprise language and helped us remain aligned even as our approach evolved.
But our product development organization was just as integral to the process. They weren’t simply waiting for us to deliver an AI capability and then deciding whether to use it. The business processes themselves evolved as the AI evolved. We actually had measures associated with changes in how the organization worked and how people were leveraging the AI throughout that iterative process.
What became particularly interesting was how adoption changed once the results became tangible. After products developed with the help of the new capabilities reached the market, other developers could see how their colleagues were able to create and innovate faster. At that point, adoption became much more organic.
People began embracing the work on their own. And importantly, they didn’t just become users. They became participants, providing ideas, feedback, and new possibilities that helped shape what we built next.
That created a reinforcing cycle: shared understanding enabled progress, progress demonstrated what was possible, and seeing what was possible encouraged greater participation and innovation.
So my advice to someone who feels paralyzed by not being fully prepared would not simply be “just start.” Preparation matters, but prepare for what needs to remain consistent while creating flexibility around what you know will change.
And involve the people whose work you hope to transform from the beginning. Sometimes readiness isn’t something you achieve before you start. It develops as people participate, see results, and begin imagining what else is possible.
The idea of readiness is indeed complex and often misunderstood. It’s fascinating how you draw parallels between personal life decisions and AI implementation. Your experience with parenthood and the unexpected challenges resonates deeply—it’s a reminder that no amount of preparation can account for every variable. The same seems true for AI projects, where adaptability and iterative progress are crucial. I wonder, though, how do you balance the need for preparation with the risk of over-preparation in such high-stakes initiatives? Your success story is inspiring, but do you think the lessons from your AI project can be universally applied, or are they more context-specific? Also, how do you handle the fear of failure when diving into something as uncertain as AI? Your insights could really help others navigating similar challenges. What would you say to someone who feels they’re not “ready” to start their AI journey?
Thank you. I think there is an important distinction between trying to prepare for everything that might happen and preparing yourself to adapt when it does.
Our AI initiative was highly iterative, but that did not mean we started without a strong foundation. We had executive sponsorship, a clear vision of the outcomes we wanted, and an architecture designed for flexibility. In particular, working from consistent logical models of the enterprise rather than individual persistent data stores gave our internal teams and external researchers a common enterprise language. That shared understanding became incredibly important as we learned and changed direction.
Our product development organization was also integral from the beginning. We weren’t just iterating the technology; the organization was iterating how it worked with the technology. We measured changes in business processes and how people were incorporating the AI into their work as the capability evolved.
Something very interesting happened as we progressed. Once products reached the market and developers saw colleagues creating faster and doing things that had previously been difficult, adoption became increasingly organic. People didn’t just use what we had created. They began contributing ideas, challenging assumptions, and helping us imagine what should come next. Progress became infectious.
I don’t think every aspect of our experience is universally applicable. Every organization has different people, processes, information, risks, and objectives. But I do believe some principles travel well: establish a shared understanding, involve the people whose work will change, create a foundation that supports change, measure how the business is actually changing, and give people permission to experiment and learn.
So to someone who doesn’t feel “ready,” I wouldn’t say simply “just start.” I would say get ready enough to start and build for your ability to adapt. You cannot prepare for every variable, particularly with something evolving as quickly as AI. But you can prepare your organization to learn from what happens next.
The idea of readiness is indeed complex and often misunderstood. It’s fascinating how life’s biggest decisions—like starting a family or embracing AI—require a balance between preparation and adaptability. Your experience with the stroller and diapers is a perfect metaphor for over-preparation; it’s a reminder that no amount of planning can account for every variable. The AI journey you described resonates deeply, especially the emphasis on iteration and openness to change. It’s inspiring to hear how your team succeeded despite initial doubts, but I wonder, what specific mindset shifts were most crucial for your team’s success? Also, do you think the fear of failure often holds companies back from even starting their AI initiatives? Your story makes me reflect on how we often underestimate the power of taking that first step, even when we don’t feel “ready.”
Thank you. I think one of the most important mindset shifts was realizing that we were not simply implementing AI. We were changing how we worked.
That meant our product development organization had to be an integral part of the journey, not simply the eventual users of what we created. As we iterated the AI, we also evolved business processes and measured how the organization was changing and incorporating the new capabilities into its work.
Another critical shift was learning to work from a shared enterprise understanding. We used consistent logical models rather than tying everyone to the individual persistent stores where information happened to reside. That allowed our internal teams and external researchers to speak the same enterprise language and remain aligned as we experimented and evolved.
Perhaps the most exciting mindset shift happened once people began seeing the results. When products reached the market and developers saw how these capabilities enabled others to create faster, adoption started becoming organic. People weren’t simply being asked to use AI anymore. They wanted to participate. They brought ideas, provided feedback, challenged us, and began imagining new possibilities themselves.
That kind of progress becomes infectious.
I do think fear of failure can keep organizations from starting, which is why executive support for experimentation was so important to us. But I wouldn’t characterize the lesson as simply “take the first step.” The right foundation matters.
For me, readiness means having enough shared vision, organizational participation, and architectural foundation to take that first step while accepting that you will learn things along the way that no amount of preparation could have predicted.