About This Episode
In this episode of Data & AI Heads, host Ian Allison speaks with Eric Gonzalez, VP of Business Intelligence and Data Architecture at Eastern Bank and founder of the advisory firm Omnificity. Eric makes the case that a solid data foundation is the prerequisite for any successful AI initiative. Drawing on a decade of experience across healthcare, financial services, and startups, he explains why organizations that skip foundational data work end up with expensive proof-of-concepts that deliver no measurable ROI.
Eric walks through practical frameworks for identifying high-value use cases, including applying the Pareto principle to prioritize where data work will drive 80% of business impact, and mapping data initiatives directly to a company's top executive-approved initiatives alongside its P&L. He shares concrete examples from healthcare case management and reseller tax permit recovery that illustrate how data teams can shift from order-takers to strategic investigators.
The conversation also covers data centralization versus federation, the importance of shared data definitions across business units, and how to sequence pioneering analytics work with disciplined data architecture. This episode is essential listening for data leaders who want to earn organizational trust, align with business strategy, and build the infrastructure that makes AI actually work.
Key Takeaways
- A weak data foundation makes AI initiatives fail: deploying AI or machine learning on top of poor-quality, siloed data does not create value, it exposes and amplifies existing data problems.
- Use the Pareto principle to prioritize data work by identifying the 20% of use cases that will drive 80% of revenue impact or cost reduction before committing resources.
- To earn strategic influence quickly, ask for the company P&L and the executive team's top 10 approved initiatives, then build a matrix that maps each initiative to its revenue and expense impact across business lines.
- Data teams should stop taking orders and start asking why: before building any report or dashboard, ask what the business will do with the output and how it maps to an operational workflow.
- Business hypotheses should be tested with data, not confirmed by it. Eric's healthcare example showed that the CMO's assumption about adverse maternal outcomes was not supported by the data, but the investigation uncovered higher-value cohorts such as diabetic patients with low medication adherence.
- Pioneering and then formalizing is an effective sequencing strategy: allow data analysts and scientists to build and gain business adoption quickly, then have the data architecture team backfill those solutions into a centralized, governed platform.
- Preventing data silos requires a single shared definition of core entities like customer or member across all business units, with clear data lineage back to the source regardless of how each department tells its own story.
- Operational workflow design should precede analytics design: understanding what actions the business will take after seeing data determines what data actually needs to be built, collected, and modeled.
What We Cover
Frequently Asked Questions
Why do AI proof-of-concepts fail to deliver ROI even after significant investment?
According to Eric Gonzalez, AI POCs fail because organizations build them on top of a weak data foundation. When data is siloed, poorly integrated, or not well understood, an AI layer does not create value, it highlights how bad the underlying data quality is. He compares it to building walls and fixtures on a house with no foundation: the first real test causes it to collapse. Without operational teams who know how to use the AI tool in their workflow, even a technically functional POC becomes an expensive digital paperweight.
How should a data team identify which use cases to prioritize when building a data foundation?
Eric Gonzalez recommends applying the Pareto principle to find the 20% of use cases that drive 80% of cost, revenue, or operational impact. He suggests starting by requesting the company P&L and the executive team's top 10 approved initiatives for the year. By mapping those initiatives against revenue and expense across business lines, data teams can identify where their work will have the greatest measurable impact and apply their limited resources accordingly.
How can a new data leader quickly gain organizational buy-in?
Eric Gonzalez recommends two immediate actions. First, ask for the company P&L to understand the top-line revenue and expense structure. Second, identify the top five to ten initiatives that senior leadership or the board has approved for the coming year. Combining these two inputs into a matrix allows a data leader to show exactly how data work connects to the organization's most important priorities, which builds credibility and strategic relevance from the start.
What is the right way to handle a business stakeholder who just wants a report built?
Eric Gonzalez advises data professionals to stop taking orders without context and instead ask three questions: why is this being built, is there a better way to accomplish the goal, and what will the stakeholder do with the output once they have it? He argues that if a stakeholder cannot describe the three to five operational steps they will take after receiving a report, they have not thought through the analytics in a meaningful way. The real deliverable is usually not the report itself but the specific data and workflow needed to drive an operational action.
Should data teams be centralized or decentralized?
Eric Gonzalez positions himself as a centralization advocate in most situations, but acknowledges that organizations separated through mergers and acquisitions with completely distinct systems may require some degree of decentralization. His preferred model is what he calls pioneering followed by foundation establishment: data analysts and scientists build solutions quickly to gain business adoption, and then the data architecture team formalizes those solutions into a centralized, governed platform. Regardless of structure, he emphasizes that naming conventions, code deployment processes, and core data definitions must be uniform across the organization.
How do you prevent different business units from developing conflicting definitions of core data concepts?
Eric Gonzalez warns against shadow data teams and fragmented tooling where, for example, one team uses Tableau, another uses Power BI, and a third maintains 15 Excel workbooks in SharePoint. He argues that while different departments may have distinct narratives and analytics needs, everyone must pull from the same centralized, quality-assured data foundation. The key requirement is maintaining clear data lineage so that any department's report can be traced back to the same authoritative source, preventing conflicting definitions of entities like customer or member.
How should data teams balance following business-led hypotheses versus letting the data reveal its own findings?
Eric Gonzalez uses a healthcare example to illustrate this balance. A chief medical officer believed adverse maternal outcomes were a major cost driver and asked the team to investigate. The data showed only a few isolated cases that were not actionable at scale. However, the investigation revealed that diabetic patients with low medication adherence represented a far larger and more addressable population. He argues that the business provides the initial hypothesis and direction, but data teams must have the agency to reject that hypothesis and redirect attention when the evidence points elsewhere.
About the Guest
Eric Gonzalez
Eric Gonzalez is VP of Business Intelligence and Data Architecture at Eastern Bank and the founder of Omnificity, an advisory firm focused on data and analytics strategy. With over ten years of experience spanning healthcare, financial services, and technology sectors, Eric has worked with organizations ranging from early-stage startups to Fortune 500 companies. His expertise covers BI architecture, data foundation design, use case prioritization, and aligning data strategy with executive business objectives. He is a practitioner advocate for building centralized, well-governed data platforms as the prerequisite for successful AI and analytics initiatives.
Connect on LinkedIn →More Episodes