AI Strategy Q&A With Juan Gorricho
The AI Table sat down with Juan Gorricho, data and AI strategy leader who’s held senior level positions at powerhouse organizations like TD Bank and Visa. Juan shared his insights on how to deliver AI value that sticks.
How do you balance short-term AI use cases that deliver immediate value with long-term data and platform investments that enable broader AI adoption across the organization?
Before I answer the tradeoff question, I push back on the framing. There's no such thing as an "AI use case." There are business problems worth solving, and AI is one of the capabilities you use to solve them. A fraud detection solution needs clean data, governed pipelines, defined business ownership, and often a process redesign before it needs a model. Those are the same prerequisites you'd face for a descriptive analytics dashboard or any other data product. AI is another link in the chain that can make the solution better once the rest of the chain is in place. When leaders frame initiatives as "AI projects" rather than "business problems we're solving with data, where AI plays a role," they end up investing in the wrong sequence.
That reframe changes the tradeoff question. It's not short-term versus long-term. It's a sequencing question: how do you structure work so that early wins fund and validate the foundation you're building?
At one organization, I inherited a team that was taking a technology-first approach, focused on loading data to the cloud without clear business alignment. The business was frustrated because they'd invested heavily but weren't seeing outcomes. I reoriented the strategy around three high-visibility business problems where we could deliver quick wins, while simultaneously building the underlying data platform those solutions would run on. It wasn't seamless. One senior business stakeholder told me flat out that he didn't believe the data team could deliver anything useful in 90 days, and he nearly pulled his support for the broader platform investment before the first solution landed. But when it did land, and his team saw the impact, he became one of our strongest advocates. That credibility bought us the runway to make the longer-term infrastructure investments.
The principle I follow: every short-term solution should leave behind a reusable asset. A fraud detection model should also produce a governed data pipeline that future solutions can use. A customer analytics dashboard should be built on a data product with defined quality standards that other teams can consume. If your quick wins are disposable, you're just accumulating technical debt.
The biggest mistake I see is organizations that separate these into two workstreams with two different teams. When the "innovation team" builds on shortcuts and the "platform team" builds in isolation, neither delivers lasting value. The teams that build business solutions need to be the same teams investing in the platform, because they feel the pain of bad foundations firsthand.
In your experience, what leadership behaviors and cultural shifts make the biggest difference between AI projects that stay in pilot and those that scale into core business operations?
I've never seen one of these initiatives fail because the model was wrong. It's always the data, the organizational readiness, or the lack of a business owner willing to change how they operate based on what the solution tells them. I deliberately don't call them "AI projects." They're data solutions that solve business problems, and AI is a capability inside them. That reframing matters because it forces leaders to pay attention to everything else that has to be true for the solution to work.
The cultural shift that changed everything for me was getting business leaders to co-own outcomes with the data team. At one organization, we moved from a model where the data team delivered analytics "to" the business, to one where business and data leaders shared accountability for every initiative. Business stakeholders had to define success in their own terms (revenue, cost, risk), commit resources on their side for adoption, and show up for regular reviews. That shift alone was the difference between pilots sitting in a drawer and solutions running in production, because now someone on the business side had skin in the game.
Governance plays a bigger role than most leaders expect, and the principle I apply is governance by design. It's the same idea as privacy by design: you embed governance, quality controls, and value measurement into the design of the solution from the start, with business, finance, and data co-owning the hypothesis. Solutions come out of the process with governance optimally pre-built rather than retrofitted. Organizations that bolt governance on after development consistently spend more time getting to production than the ones that build it in from day one. I've watched teams add compliance and data validation after the fact, and it's always painful and expensive. Governance by design feels slower at first, but it's consistently the accelerator because there's nothing to go back and fix.
The piece that ties it all together is investing in people before tools. Not turning everyone into data scientists, but building enough data literacy across the organization that leaders can have an informed conversation about what AI can and cannot do inside a given solution. When non-technical leaders understand model limitations and data quality tradeoffs, they make better decisions about where to deploy AI and, just as importantly, where not to.
How do you set expectations with executives and boards about what AI can realistically achieve in the near term, without over-promising or under-delivering?
I start every one of these conversations with a business problem, not a technology capability. When executives ask "what's our AI strategy?" the wrong answer is a list of tools and pilots. The right answer is: "here are the three business problems where the economics work, here's what we need to invest, and here's how we'll measure success." AI shows up in that answer as a capability that makes the underlying solution stronger, not as the strategy itself.
Equally important is who delivers the message. The data leader should not be the one presenting AI to the board in isolation. The conversation works when the business owner of the process stands up alongside the data leader, co-presenting a solution they jointly own. The business owner describes the problem and the dollars at stake. The data leader explains how AI makes the solution stronger. Together they show how the outcome ties back to a board-level OKR the organization is already tracking. That's a different conversation than "we deployed an AI capability." Now the business owner is accountable for the outcome in front of the same audience they already report to, not just consulted on it, and the work gets grounded in what the board already cares about rather than a separate AI narrative.
To make that concrete: instead of "AI will transform customer experience," the business owner puts it to the board as "we can reduce fraud detection response time by 40%, which based on current loss rates translates to X dollars in avoided losses, which moves the risk-adjusted return OKR by Y basis points." When you quantify the opportunity, the investment, and the connection to an objective the board has already committed to, executives can make real decisions instead of chasing narratives.
I'm also direct about what has to be true for AI to work. If the data quality in a particular area isn't ready, I say so. If the organization needs six months of foundational work before a solution can scale, I present that as the realistic timeline and explain what we'll deliver along the way. Executives respect honesty about prerequisites far more than optimistic timelines that slip.
I also always include what we're deliberately choosing not to pursue. We evaluated ten opportunities and picked three based on feasibility, impact, and organizational readiness. That kind of strategic discipline builds more confidence at the board level than a laundry list of everything AI could theoretically do.
