Artificial intelligence (AI) is a top priority for many organizations. Leadership teams expect it to improve efficiency, create measurable business value, and support informed decisions. Yet in many cases, AI initiatives stall after the pilot stage. Even when early use cases show promise, scaling can meet headwinds.
Often, the problem isn’t the AI model or tools. Organizations may already have capable tools and strong technical talent. The breakdown frequently occurs when data is siloed in systems, processes are inconsistent, and the path from experimentation to production is unclear.
That’s why a practical question is emerging for chief information officers and data leaders: Is the issue your AI strategy, or the platform underneath it?
Fragmentation Across the AI Lifecycle
In general, organizations aren’t building their AI environments from scratch. A mix of technologies may already be in place, such as:
- Data warehouses and data lakes
- Business intelligence (BI) tools and reporting environments
- Data science notebooks and machine learning tools
- Generative AI pilots or proofs of concept
The challenge is whether these platforms and tools can work together to support repeatable, production-ready AI.
This is where some teams encounter friction. Common issues include:
- Data spread across multiple systems with inconsistent governance
- Limited visibility into how models are built, tested, and approved
- Difficulty reproducing results from one environment to another
- Prolonged handoffs between development and production
- New generative AI use cases without corresponding controls
Each of these challenges on its own may seem manageable. Taken together, they can increase risk and erode trust.
In practice, this appears in familiar ways. A model performs well in development, but the same result cannot be reproduced later because the data has changed or assumptions weren’t documented. A generative AI pilot gains internal attention, but no clear approach exists for monitoring as it moves into broader use.
The AI model may not be the main problem. The bigger issue may be how the data, tools, and processes around it are set up.
Why This Matters More With Generative AI
Generative AI has increased the urgency; not because it changes everything, but because it exposes weaknesses that already exist.
Traditional analytics and machine learning need solid data quality and governance. Generative AI raises the stakes. It introduces greater reliance on current, well-governed data; faster expectations around iteration; and increased scrutiny of outputs that may influence decisions, communications, or customer experiences.
Generative AI puts more pressure on organizations to have clean data and clear processes. Further, if the underlying data platform is fragmented (or nonexistent), generative AI will magnify existing weaknesses. For example, we worked with a global insurer whose data was spread across spreadsheets, portals, and disconnected reporting systems. Conflicting definitions, manual data reconciliation, and weak governance made it difficult to establish a trusted source of truth. Before advanced analytics and AI initiatives could scale, the organization first needed to harmonize data, improve lineage visibility, and implement repeatable governance processes. Once those foundational issues were addressed, the organization was able to improve decision making, reduce reconciliation efforts, and create an environment better suited for AI-driven outcomes.
How Data Platforms Affect AI Outcomes
Data platform discussions usually surface at this stage. Teams may spend substantial time comparing technologies, weighing features, integrations, and ecosystem fit. These are valid considerations, but they may not be the main issue.
For many organizations, the issue is whether they have a consistent process for turning data into AI solutions that can be deployed, managed, and monitored over time.
That perspective shifts the conversation from a product comparison exercise to a strategic review of operations.
A modern data platform can help organizations address questions such as:
- Can teams access reliable data without creating duplicate pipelines or ad hoc workarounds?
- Can pilots or experiments be tracked and reproduced with confidence?
- Can models or tools move into production through an efficient process?
- Can performance, quality, and potential risks be monitored over time?
For some organizations, deeper integration within an existing ecosystem may reduce complexity and support faster progress. For others, flexibility across environments may align better with a broader technology strategy or operating model.
The key point is that platform selection should align with business needs, governance considerations, and long-term architecture decisions.
Common Gaps
Across industries, similar challenges appear repeatedly, including:
- Data is available, but not operationally usable. Many organizations have no shortage of data. The issue is that it’s distributed across (or siloed in) systems, governed inconsistently, or requires significant manual effort to prepare for analysis and model training. Teams spend time searching, cleaning, and reconciling.
- Experimentation and pilots lack structure. IT and data science teams may be able to build and test models, but inconsistent documentation and tracking can create issues later. When inputs, model versions, assumptions, or results aren’t captured in a disciplined way, validation becomes challenging and auditability may become more complex.
- The move to production is inconsistent. This is a common point of delay. A model may perform well in a development environment, but the process required to deploy and monitor it in production isn’t clearly defined. As a result, promising use cases may stall before delivering value more widely.
- Governance is introduced too late. Security, access control, lineage, and compliance requirements are sometimes treated as downstream review items rather than core design inputs. This can lead to rework, delayed approvals, or solutions that are difficult to support under scrutiny.
- Generative AI moves faster than the control environment. Teams may move quickly to test AI agents, search tools, or content generation use cases. However, when validation, monitoring, lineage, and usage boundaries aren’t addressed early, the governance gap can expand at an alarming rate.
These gaps are common and don’t necessarily indicate a failed AI program. More often, they reflect an environment that wasn’t designed for scalable AI (yet).
What a More Effective Approach Looks Like
Organizations making substantial progress with AI aren’t simply adding new tools. They’re building a data foundation that can help teams access reliable data, develop use cases, and move beta projects into production with more consistency.
An effective approach starts with making data easier to find, govern, and use across systems. Teams need shared access patterns, clear ownership, consistent quality checks, and visibility into where data comes from and how it changes. Without that foundation, IT and data teams may spend more time reconciling data than improving use cases.
The next step may be a more disciplined data platform strategy, supported by clearer processes, stronger governance, and a practical view of how AI moves from concept to production. That includes versioning data and models, documenting assumptions, testing outputs, managing approvals, and monitoring performance after deployment. These elements can help teams scale promising pilots without relying on one-off or manual processes. This also can support semantic mapping, or the aligning of the meaning of key entities, concepts, and ontologies across systems, to help ensure the consistent interpretation of data throughout the AI lifecycle.
In this context, the objective becomes less about choosing the most feature-rich platform and more about developing the environment needed to help the organization use AI in a safe and productive way.
Platforms such as Databricks and Snowflake can support this type of environment; however, progress depends on how well the platform, governance model, and delivery processes work together.
Platform Trade-Offs That Affect AI Progress
Data platform decisions often come down to trade-offs. Some organizations may benefit from deeper integration within an existing cloud environment because it can simplify identity management, data access, governance, and tool connectivity. Others may need more flexibility across environments to support multiple business units, reduce dependency on one provider, or preserve options as AI needs change.
The choice depends on what the organization needs the platform to do. Leaders should consider where critical data resides, how teams will govern access, what skills are available internally, how AI tools and pilots will move into production, and how the environment will be supported over time. The right platform approach should support reliable AI delivery at scale while balancing integration, flexibility, governance, and long-term architecture needs.
Actions to Consider Now
For leadership teams reviewing AI strategy and data platforms, these critical steps can uncover where challenges may arise.
- Explore fragmentation across the current environment. Identify where data, tooling, and handoffs are disconnected, including reliance on manual workarounds or duplicate processes.
- Review the path from experimentation to production. Determine whether models and tools move forward through a consistent, supportable process or stall between development and deployment.
- Consider governance early, not later. Review data access, lineage, validation, auditability, and compliance expectations before tooling and pilots expand.
- Address generative AI risks directly. Consider how outputs are tested, monitored, and controlled, particularly when they may influence decisions or communications.
- Align platform decisions with desired business outcomes. The objective should not be to select a feature-rich platform in isolation. Instead, it should be to develop an environment that can support reliable, scalable outcomes.
How Forvis Mazars Can Help
Closing data gaps to support scalable AI may require considerable effort. However, the potential for time savings and resource reallocation may be more than worthwhile.
If your organization is considering how to modernize its data platform strategy and approach to AI initiatives, professionals at Forvis Mazars can help support your efforts.
Our AI Strategy & Integration services include helping organizations:
- Review data and AI platform readiness
- Identify gaps across governance, architecture, and processes
- Develop practical roadmaps aligned with business goals
Connect with us today to discuss how your organization can better support scalable AI.