STEPS TO DATA + AI SUCCESS
Preparing data environments to effectively support AI implementations requires a strong organizational foundation.
Start with a top-down approach, advised Steven Karan, head of data, AI, and insights for Capgemini Canada. “Focus on ensuring the organization’s AI strategy can be directly connected to business priority outcomes. Second, implement a right-sized security and governance framework that protects the brand of the organization by minimizing unintended or unsanctioned use of AI. And third, conduct a holistic architecture review, ensuring the data platform, tooling, and required AI services are in place to sustain AI solutions.”
Develop an AI framework that “creates a standardized approach to be followed throughout the organization,” said Krishna Sai, SVP of engineering at SolarWinds. “Frameworks must ensure compliance, fairness, and transparency, including the implementation of observability tools that monitor data quality, lineage, and drift to maintain model performance.”
Data governance is essential, and this starts with understanding what data you have, the quality of the data, and whether it is trustworthy or not. “A robust data governance program ensures that the data used for AI projects is accurate, consistent, and reliable, which is crucial for building effective AI models,” Robinson said. “Without strong data governance, businesses may encounter data quality issues, leading to inaccurate insights and poor decision making. Furthermore, a comprehensive data governance framework aids in identifying what data the organization possesses, adequately preparing it for AI applications and ensuring compliance with regulatory requirements.”
Liddle urged the creation of a dedicated “AI strategy board” composed of “C-suite executives, business unit leaders, technology leaders, legal counsel, and finance representatives to shape the company’s AI vision and governance framework.” This board would provide oversight into the data-to-AI flow essential to AI applications. “IT and data leaders would collaborate with the board to reassess enterprise architecture, ensuring it’s fit for AI. This includes unifying data silos, implementing robust data classification frameworks, and automating data curation pipelines for handling unstructured data.”
Moving beyond “pilot purgatory” and deploying AI tools requires data leaders to identify and address gaps in their data environment, Baldenko explained. This includes following a “purpose-driven innovation playbook,” he advised. “Given the hype stemming from advances in generative and agentic AI, it can be challenging for technologists to avoid getting distracted by every new and exciting development. But it’s critical to align AI investments with advancing the company’s long- and short-term strategic objectives.”
A well-engineered “data estate,” centered around a modern data lakehouse architecture standard, is required to enable AI development at scale, Karan urged. “The lakehouse standard enables unified data storage across structured and unstructured data, scalable data lakes, seamless integration of BI and AI workloads, enhanced data governance with features such as ACID transactions, and robust meta data management.”
Start small with pilot projects, “leveraging automation and designing systems for scalability to help streamline workflows, reduce risks, and ensure AI’s long-term success,” Sai advised.
A scalable infrastructure is key to moving forward. “IT and data leaders must invest in scalable infrastructure, such as cloud-based systems; ensure high-speed data pipelines; and support large-scale operations,” said Maria Vaida, assistant professor of data science at Harrisburg University of Science and Technology.
“Instead of building models from scratch, leverage existing pretrained models, integrating them into workflows to extract the best features,” Vaida added. “Internal models can be developed for unique, proprietary data, maximizing value from specialized datasets. Integrating data from multidisciplinary domains enhances model robustness and broadens applicability. Upskilling teams in emerging techniques like graph neural networks and fostering transparency with explainable AI frameworks build stakeholder trust. Embedding privacy-preserving algorithms ensures responsible AI deployment.”
Of course, culture is the deciding factor in establishing a healthy data flow to AI applications and systems. “For too long, IT and business teams have been siloed, with business users making requests of the IT team without understanding the scope of the technology needed, and IT teams producing insights without knowing what business problem they’re being used to solve,” said Vartak.
To bridge this gap, “start with adopting a centralized data architecture to ensure cross-organization visibility and establishing data and AI regulation frameworks and education organization-wide.”
Having essential skills is also part of the data-to-AI realm. The top challenge is a pressing shortage of skilled data engineers, who are needed “to manage and process large volumes of data for AI and analytics,” said Robinson. “Data engineers are responsible for designing, building, and maintaining the infrastructure that allows data to be collected, stored, and analyzed efficiently, making it challenging for organizations to fill these positions.”
One workaround to this problem is “citizen data engineers—individuals who may not have formal training in data engineering but possess the necessary skills and knowledge to handle data tasks,” Robinson continued. “These citizen data engineers often come from various backgrounds and require tools and platforms that simplify data engineering processes.”
MEASURING DATA + AI SUCCESS
As with all key technologies, you can’t manage what you can’t measure: It’s critical to understand how the data-to-AI pipeline is delivering.
The true test comes after implementation, when it’s time to look at the impact of data management initiatives on the course of AI. “This involves identifying specific business outcomes and metrics that can be used to measure the ROI of AI projects,” said Hamilton. One way to “track is to monitor agent actions to assess performance, accuracy, and the application of safety and transparency guardrails. This helps in identifying areas for improvement and ensuring that AI systems are functioning as intended.”
Before all else, “It is critical that AI metrics are defined in alignment with your broader IT and business goals,” said Lieberman. “This ensures that the AI investment directly addresses specific business needs and that desired outcomes are clearly established at the outset of every project. While it’s tempting to explore the latest AI technologies from an engineering standpoint, in the end, it’s more effective to focus on simple, effective solutions that directly support your goals rather than overcomplicating implementations or chasing features.”
An oversight board or committee can also be instrumental in tracking AI data resilience, Liddle said. “Companies should track how quickly and effectively they can recover data in the event of disruptions. The alignment of AI initiatives with business objectives, as overseen by the board, provides a higher-level metric for gauging success. By ensuring that AI use cases deliver measurable business value, companies can assess whether their data environments are effectively enabling their broader AI strategy.”
There are also a variety of key performance indicators that can be applied, from performance metrics to data quality, said Karan. For piping data into AI solutions, he recommended cost optimization, operation efficiency, and business value. “Cost optimization KPIs measure and monitor the cost of compute, consumption, and storage. Operation efficiency metrics monitor the time taken for data processing tasks, and value metrics measure the number of business decisions supported by AI insights or business outcomes enhanced by AI.”
Vaida identifies productivity, data utilization, workflow optimization, and improved model performance as key measures of data-to-AI performance.
“Productivity should not only reduce the time required for data preparation, model training, and deployment, but also make domain experts more efficient in their work,” she said. “By providing them with reliable and transparent tools, organizations can foster trust and strengthen partnerships between technical teams and domain experts.”
Success starts with asking these questions: “Are we making better decisions faster? Are we avoiding risks like breaches or unintended bias in our AI models?” said Ithal. “If your data ecosystem is fueling smarter outcomes and staying within ethical and regulatory boundaries, you’re doing it right. Metrics like data access velocity, compliance adherence, and AI performance benchmarks can tell you if you’re on track.”