Issue #42 – Ecosystem Considerations for Data Science
ML & AI don't exist in a vacuum; what you need to know across the data ecosystem as these areas continue to evolve
Read Time: 15 minutes
In this newsletter, I like to focus on the big picture.
This involves making sense of different categories, topics and ideas, taking a holistic view of topics.
And while people think the conversations about AI are big-picture stuff, realistically, they aren't.
It is not that AI is not a big picture idea or concept, it is that most people think of it too narrowly.
Almost every company is rushing to implement some form of Artificial Intelligence, convinced it's the silver bullet that will solve all their business challenges. They hire a team of data scientists and ML/ AI engineers and get building.
But they don’t consider the ‘big picture’. They don’t consider the rest of the Data Ecosystem.
Because here's the reality: ML and AI don't exist in isolation.
Machine Learning and AI need to be deeply embedded within the business, data and technology teams, with untold implications on how the organisation operates. However, few leaders are thinking about them that way. Instead, they are treated as standalone technologies that will replace workers or magically create efficiency.
In the end, your ML and AI solutions are only as good as the data ecosystem they are built on. Feel free to reference the classic Gartner statistic (or whichever one you find) that 80% of AI projects fail to deliver. The reason that will hold true for a while? Because nobody is thinking from a true ecosystem perspective!
So it is fitting that for the final article of our six-part Machine Learning & AI series, we explore what data ecosystem considerations you need to make to succeed with ML and AI. Buckle up, this may be more learning and thinking than you’ve done in a while!
ML/AI Needs to Align with Your Data Ecosystem
The first thing to understand is that AI initiatives are rarely born out of true strategic planning. Instead, they come from the top or are proposed by vendors:
Top-down Executive Direction – Business leaders read about AI in Harvard Business Review, hear about it at conferences, or feel pressured by board members asking: "What's our AI strategy?" These leaders then push for AI adoption without understanding the foundations needed to make it successful.
Vendor-driven solutions – Technology vendors pitch "AI-enabled" platforms with promises of immediate ROI and competitive advantage. These sales pitches rarely mention the ecosystem requirements necessary for these tools to function effectively.
Either way, expectations for AI are sky-high while the path to successful implementation remains unclear. This disconnect leads to what I call the "AI sandbox trap"—isolated proof-of-concept projects that show promising results in controlled environments but fail miserably when moved to production.
According to Gartner, 60% of AI and machine learning projects will be abandoned in 2026 and 30% of GenAI projects will be shut down after proof of concept. This represents billions in wasted investment and countless hours of lost productivity.
Why such high failure rates? Because most organizations don't align their AI initiatives with their broader data ecosystem.
How ML/AI’s Four Core Components Connect with Your Data Ecosystem
In our first article on ML and AI basics, we outlined four core components necessary for ML/AI success:
Data
Computing Power/Infrastructure
Algorithms
Human Expertise
Thinking about this more holistically, data scientists or ML engineers do not build these areas, it’s a broader group effort. Starting from the bottom of the pyramid, we can see how each of these components intersects with multiple domains in the Data Ecosystem:
Data – No surprises here. Having quality data that is appropriately governed, secured, and accessible is a huge component of your data management function and data engineering capabilities. Without the right data flows, architecture, or oversight, the data will not be AI-ready in any way, shape or form!
Tech Infrastructure – The technical underpinnings get into the technical strategy and data platform. ML/AI solutions should not warrant an entirely new tech stack or approach, and it will require new thinking into how your existing technology stack evolves or integrates with legacy technology to enable these new capabilities.
Algorithms – While a lot of people think these are individual Python scripts, algorithms need to be developed, deployed, monitored, and maintained within your broader tech environment. Not to mention cross-collaboration across teams and feedback loops to make sure it hits the goals of the organisation or addresses business needs.
Human Expertise – With ML/AI changing the game, the rest of the organisation needs to adapt as well. My definition involves more than just hiring data scientists - it requires organizational change management and education. This intersects with:
Organizational structure
Operating model
Data and AI literacy
Stakeholder alignment
Change management processes
ML/AI isn't a standalone capability - it's the culmination of a well-functioning data ecosystem. Each initiative requires multiple ecosystem components working in harmony, and weakness in any area can undermine the entire effort.
The uncomfortable truth many vendors and executives don't want to acknowledge is that you can't simply leapfrog to advanced AI capabilities. You must build the ecosystem foundations first, or at least strengthen them concurrently with your AI projects.
Now let’s get into the detail and examine how you need to think about each domain in the data ecosystem when considering Machine Learning and AI.
Key Data Ecosystem Domains Enabling ML/AI
The perspective we are taking in the next section underscores the uniqueness of this newsletter.
99% of the AI and ML content available today is talking about building solutions, coding, poor data quality, use cases, etc.
While all this content is great in its individual silos, it doesn’t take a holistic approach. Therefore, it leaves out key details that people need to think about to make ML and AI a reality
I began this newsletter to take a more holistic perspective, to ensure people think of the big picture, instead of getting lost in the individual domains and details. And with the buzz around AI and Machine Learning, examining this overarching perspective here will be crucial for those wanting to make AI a reality in their organisations.
So let’s explore each area of the data ecosystem and the specific considerations needed to support effective ML/AI implementation. To do this we will go back to my Data Ecosystem one-pager and explore the relevance of each area/ domain. For each point we will outline a question to think about and how to consider it:
Business Drivers
The first consideration group is the most important because it is why your organisation even exists. Machine Learning and AI should never be implemented for their own sake but must directly support specific strategic priorities and solve real business problems.
We data people forget this sometimes, which creates the divide between an organisation’s strategic direction and data team-owned ML/AI initiatives. And while a lot of data people speak about use cases and the business strategy, it is more than that. This involves the business model, org structure, change management, etc.
Business Model: How will AI change your business model? How can it enhance your value proposition? While most companies are thinking about AI use cases, they aren’t considering that their business model may need to change. A lot of companies have turned into ‘data’ companies because of ML and AI. Should yours?
Business Strategy: How can data science solutions support your organizational strategy and roadmap? These capabilities need to specifically advance your strategic priorities and competitive positioning rather than be standalone solutions. What KPI or targets should it help advance?
Business Needs & Requirements: What are the most pressing challenges business stakeholders are facing? Where can ML and AI help? Don’t start with the solution; identify concrete pain points and use cases and then figure out how AI can deliver measurable impact.
Data Strategy: Where do ML/AI solutions fit into your data strategy? A data strategy should examine the maturity of your data foundations and capabilities, allowing you to better plan how the organisation can build value-adding ML/AI solutions with timelines and a positive ROI.
Data Investment Decisions: What budget do you have for ML/AI versus other data capabilities? It is essential to look at your data budget holistically, ensuring that your spending on AI does not take away from investment into your data infrastructure and other fundamentals.
Org Structure: How should your organisation be structured to embrace ML/AI? This is significantly ignored at every company I see. Part of success is ensuring that business users have the support they need to use AI tools and can embrace their positive impacts. Today, most of what you see is spent on AI to justify cost-cutting and little to no support for those employees to use those tools properly.
Operating Model: How do the operations of your organisation change with ML/AI? AI tools should make things easier, right? Without mapping out the impact on operations, processes and policies shift without employees being able to adapt. This directly aligns with the Org Structure shifts and is also one of the most underrated components of AI maturity.
Stakeholder Alignment/ Change Management: Ah, transformation and change management! How will you get buy-in from key stakeholders? Developing a comprehensive change management approach and working closely with involved stakeholders is crucial to building trust and driving adoption.
Data Lifecycle Process
Now we get into the data backbone that ML and AI solutions are built on top of. As everybody (apparently) knows, ML/AI can only be as good as the data that powers it. When considering the data lifecycle, there are a whole host of data capabilities that factor into how the data is generated, processed, stored, serviced and used.
Some companies tee it up to Data Engineering or MLOps, but in reality, it is far greater than that. In fact, just leaving it up to individuals in those roles is precisely where a lot of those problems start, as those employees can't consider everything in tandem while executing their day jobs:
Data Approach/Philosophy: The data lifecycle starts with the overall data approach. Basically, how does your organisation treat data and data management? How will that evolve to incorporate ML/ AI? Aspects like centralised, decentralised/ mesh, and hybrid factor into this, preceding decisions about how to set up your data workflows for future solutions.
Data Use Cases: What are the problems your ML/AI initiatives will solve? What are the technical specs and data feeds that fit into that? These ML/AI use cases build on your business needs and should be underpinned by details like data availability, technical feasibility, and measurable success criteria. A data/ AI product manager is crucial for this step!
Data Sourcing: Where is your data coming from? What data will be used for your ML/AI needs? This is where most data quality problems go wrong (upstream), and understanding exactly what data you need to support accurate model training is crucial before you put all the work into processing it.
Data Integration & Storage: How is your data feeding into the platform? There are a lot of considerations within this (e.g., data modelling, engineering, etc.), but really, it comes down to ensuring the data is structured to be used for ML and AI use cases down the line. Ideally, this would align with the business processes and data workflows that feed other analytical use cases.
Processing & Servicing: How will models access and use data in production? Design your platform to support the unique requirements of ML workflows, including version control for training data and how models serve existing analytical tools and business technologies.
Analytics & Data Science: The relevance of this capability goes without saying; do you have the right analytical capabilities and methodologies to develop ML and AI solutions? Reference my previous article for more details on how to productionise models, but overall this is about aligning your ML/AI solutions with your overall analytical suite and capabilities. In the end, these resources should be thought of holistically (not data analysts vs. scientists), and they should complement rather than compete with each other.
Underpinning Management
One of the trending elements of AI content is that lack of data quality = poor AI outputs.
They aren’t wrong. Here is where we consider the management layer, which provides the governance, quality controls, and operational support essential for sustainable ML/AI. Without these foundational elements, data feeding into ML/AI solutions quickly becomes unreliable and the models themselves turn into untrusted experiments rather than tools that enhance business decisions:
Data Governance: Should data be governed differently for ML/AI? And how do you oversee ML/AI development and deployment? Understanding the data needs for these solutions and then establishing clear policies for data governance, model development, validation, deployment, monitoring, etc. is crucial. And this should align with your broader data governance framework.
Data Quality Management: How do you implement data quality practices for ML/AI applications? My next article will dig deeper into this, but AI has changed the game for data quality. With less human intervention, quality monitoring, observability, and remediation processes need to be set from the outset with an overall recognition that ML models amplify data quality issues in ways traditional analytics may not.
Privacy & Security: How will you address AI-specific privacy and security concerns? A topic nobody really has an answer for… Nonetheless, consider policies for sensitive data handling in ML/AI contexts. Things like anonymisation, consent management, and access controls for both data and models have to be factored into development and deployment.
DataOps/DevOps/MLOps: Do you have operational processes to support the ML lifecycle? I’m lumping these all together because it really is a software, data engineering, and ML operational task to come up with these solutions. MLOps practices help address the unique challenges of machine learning development, but the wider software development lifecycle needs to be considered, as most of these solutions will integrate into other technical tools.
Data Consumption & Business Decisioning
While I said Business Drivers is the most crucial category, this might be the second. Why? Because the ultimate test of ML/AI success is whether it drives better decisions and actions. This is essentially where the value of these solutions is proven!
At a category level, this requires thinking about ML/AI not only delivering model outputs but also integrating those insights into business processes and building trust with end users. Without careful attention to how AI is consumed and applied, even technically sound models will fail to deliver business impact:
Dashboards/ UI Tools: How will you visualize ML/AI insights for non-technical users? Dashboards are not dead; UI has just evolved. Now, chatbots and other intuitive interfaces that may be embedded into technology should be used to help business users understand and trust ML/AI-generated outputs.
Reporting & Visualization: How do you communicate ML findings effectively to stakeholders? How do these tools integrate/ support existing reporting needs? Every company already has reporting & visualisation. While AI can create reporting efficiencies, organisations have to understand how it best fits into the processes without alienating stakeholders.
Data & AI Literacy: Do employees know how to use ML/AI and understand its capabilities and limitations? Invest in education and co-development to build appropriate levels of AI literacy across the organization. A considerable part of this is making sure business stakeholders are involved in and understand the creation of ML/AI tools.
Business Case Development: How will you measure ML/AI success? Before even launching a solution, establish clear ROI frameworks that account for both direct benefits (cost savings, revenue increases) and indirect benefits (improved decision quality, time savings).
Value Creation: How will ML/AI translate into tangible business outcomes? Outline the explicit path from model outputs to business actions and where embedded AI insights directly create value. This should be done in tandem with the business case!
Strategic Considerations
Beyond the whole process we’ve just gone through, more holistic thinking still needs to be done. Long-term ML/AI success requires considering your technology infrastructure, architecture, and organisational capabilities.
These strategic considerations I have labelled in the Data Ecosystem often get ignored because it doesn’t fit nicely into a single job description or role. But in the end, you need to factor these in, otherwise an organisation will never optimise AI’s role in their broader technology and organisational landscape:
Technology Strategy: How does ML/AI fit into your broader technology roadmap? We’ve reviewed some of the technology and infrastructural considerations for ML/AI. It is crucial to develop a coherent technology strategy that addresses the need for specialised AI-enabling tools and how you will invest in those.
Enterprise Architecture: Does your architecture account for ML/AI's unique requirements? Ensure your enterprise architecture framework explicitly addresses how ML/AI systems fit within your broader technology landscape, including data flows between systems, integration patterns for models, and clear boundaries between experimentation and production environments.
Tech Debt: How will legacy systems impact your ML/AI initiatives? Assess whether existing technical debt in data systems, infrastructure, and applications will constrain your ML/AI capabilities and develop explicit approaches to modernise critical components without disrupting ongoing operations.
Data Team Scalability: Do you have the data team necessary to support growing ML/AI demands? This builds on the organisational structure but from a future-focused lens. Design team structures, career paths, and knowledge management practices to incorporate ML/AI capabilities.
So there you have it, five pillars within the Data Ecosystem broken down for your ML/AI considerations. As I’ve said, understanding these ecosystem dependencies is the first step toward AI success, and without it, achieving success will be a lot harder!
Now, let’s briefly explore how to implement a holistic approach that balances ecosystem development with AI expectations.
Implementing a Holistic Ecosystem Approach
If you’ve gathered anything from this article, it is about thinking about things holistically. The question is, how do you implement holistically (because there is a lot to think about)?
Start with the Strategy – Align your ML/AI initiatives with the business strategy:
Define the “Why": Identify specific business problems ML/AI will solve with measurable outcomes directly linked to your strategy. Make it measurable and realistic.
Prioritise Use Cases: Focus on the highest business value use cases while considering data ecosystem readiness. That is where you start!
Secure Executive Commitment: Ensure leadership understands both the potential outcomes and the foundational work required. Get the investment into both.
Set Realistic Expectations: Be honest about timelines and investment needs rather than promising overnight transformation.
Assess Your Ecosystem Readiness – Before implementation, evaluate your current state:
Data Foundations & Technical Infrastructure: Spend time with engineers, analysts, and the tech team to assess data availability/quality and technological underpinnings for your priority ML/AI use cases.
Talent and Skills: Determine skills and capabilities required to deliver the use cases and evaluate them against what you have internally.
Process Maturity: Evaluate if your governance and operational practices can support ML/AI development and deployment.
Balance Foundation Building with Forward Progress – Ensure data fundamentals are being worked on while delivering ML/AI initiatives:
Target Foundational Improvements: Strengthen the specific ecosystem components that enable your priority use cases. Investment in AI should also go to these areas.
Implement in Phases: Like anything, ML/AI initiatives should be broken into smaller chunks that deliver incremental value. The same goes for the foundations that stand them up!
Establish Feedback Loops: Regularly evaluate both ML/AI outcomes and ecosystem health to adapt your approach. Steering groups should help with cross-collaboration and communication barriers.
Ensure Organizational Enablement – Don't overlook the human element:
Cross-functional Collaboration: People hate RACIs but build some framework for how different teams will work together to deliver AI solutions after they exist.
Continue Investing in Skills: Required capabilities will change across the organisation (not just within the data science team). Figure out where to invest in those skills.
Manage Change: Get buy-in and build confidence in ML/AI solutions. If people aren’t bought in, they won’t use it.
Evolve your Org Structure & Operating Model: Update the foundational pieces that underpin how the organisation works. ML/AI will change this, and you need to make sure this is reflected in these strategic documents!
The ML/AI Conclusion
Throughout this six-part series on Machine Learning and AI, we've covered everything from fundamentals to implementation approaches. This final instalment ties it all together with perhaps my most important insight:
Machine Learning and AI will never succeed until you consider the holistic Data Ecosystem approach!
Companies that view ML/AI as isolated technologies will remain in the failure category and wonder why they’ve wasted so much money.
In the end, success isn't about having the most advanced algorithms or the largest data science team. It's about creating a functioning data ecosystem where ML/AI can naturally deliver ongoing business value.
Organisations that embrace this holistic perspective will transform their businesses through AI, while others chase shiny objects without success.
Thanks for reading this series, it means a lot! Next week, we start to dig into a 3-4 part series on data quality, specifically how you can think about it with all the different tools and approaches. Subscribe, comment, and have a great Sunday!
Thanks for the read! Comment below and share the newsletter/issue if you think it is relevant! Feel free to also follow me on LinkedIn (very active) or Medium (increasingly active). And if you are interested in consulting, please do reach out. See you amazing folks next week!
This is such an insightful breakdown of why ML and AI initiatives fail when the broader data ecosystem isn’t considered. Aligning AI with organizational structure, data lifecycle, and stakeholder buy-in is critical for success. For companies looking to implement practical AI solutions, <a href="https://arcq.ai/ai-development/" target="_blank">ARCQ AI’s AI Development services</a> provide end-to-end solutions tailored to business needs. Their <a href="https://arcq.ai/ai-consulting/" target="_blank">AI Consulting</a> also helps organizations ensure that AI is effectively integrated across teams and systems. You can explore more about their <a href="https://arcq.ai/ai-development/" target="_blank">development offerings</a> to get started on building scalable and aligned ML solutions.