The CIO betting against the current AI timeline
Steve Bronson joined Southern Glazer's Wine and Spirits as Chief Information Officer in September 2025, coming from McDonald's, where he led global technology infrastructure and operations. A year in, the headline from his public remarks is not a new AI rollout. It is a deliberate sequencing decision: continue maturing the company's OneTech platform and modernize the ERP backbone first, and treat AI capabilities as the layer that gets built once that foundation is trustworthy, not before.
That is a contrarian posture in the current environment, where boards routinely ask technology leaders what AI initiative shipped this quarter, and where the safer career move is often to point at a pilot regardless of whether the underlying data can support it reliably at scale. Bronson's approach is a bet that skipping the sequencing produces AI outputs that look impressive in a demo and fail in production, a failure mode that is harder to reverse than a delayed rollout.
Why beverage distribution makes this sequencing harder
Southern Glazer's operates the middle tier of the US beverage alcohol supply chain, the layer connecting suppliers to retailers and restaurants, inside one of the most heavily regulated distribution environments in consumer goods. State-by-state licensing rules, supplier contract terms and compliance reporting requirements vary constantly, which means the underlying data model has to reconcile far more structural complexity than a typical CPG distributor before any AI system built on top of it can be trusted with real decisions.
That complexity is exactly why Bronson's ERP-first approach reads as pragmatic rather than cautious for its own sake. An AI system trained on inconsistent regulatory or inventory data in this environment will not just underperform, it will generate recommendations that could create compliance exposure across the states where Southern Glazer's operates, a downside risk that a generic e-commerce AI pilot does not carry in the same way. Few consumer-facing retail AI launches carry that kind of regulatory tail risk, which is part of why this sequencing discipline is more visible in distribution than in retail storefronts.
The reporting line that signals how seriously this is treated
Bronson reports to Chief Growth Officer David Chaplin rather than to a CFO or COO, a structural choice that positions technology explicitly as a growth enabler rather than an operating cost to be managed down. That reporting line matters more than it might appear, because it determines what gets funded: a CIO reporting into growth has a different budget conversation than one reporting into finance, particularly when the pitch is spend more now on data foundations before AI spend pays off later.
The stated shift in framing, from asking what technology are we implementing to asking what business value are we enabling, is the kind of language every technology organization claims to use. What makes it credible here is that the sequencing decision, delaying visible AI wins in favor of infrastructure work, is the harder version of that philosophy to actually execute under pressure, not just state in an interview. Most CIOs say they are aligning technology to business outcomes; fewer are willing to let that alignment slow down the AI announcements a board is actively asking for.
What most retail and CPG technology leaders are doing instead
Across the retail and CPG stories we track weekly, the more common pattern looks the opposite of Southern Glazer's approach: ship a visible AI feature first, in a chatbot, a personalization engine or a content tool, and treat data governance as a parallel workstream to be cleaned up later. That order is understandable given how AI announcements currently move stock prices and board sentiment, but it defers the harder, less visible work exactly when the AI output volume it needs to support is scaling fastest.
The risk compounds quietly. Every additional AI use case layered onto ungoverned data increases the blast radius when the underlying inconsistencies eventually surface, whether as a bad customer-facing recommendation, an inventory error or, in a regulated category like alcohol distribution, a compliance finding. Bronson's sequencing is a bet that paying that cost upfront, in slower visible AI progress, is cheaper than paying it later at a scale that is much harder to unwind. The companies most exposed to this risk are the ones expanding AI use cases fastest while treating their data platform as a background maintenance item rather than a prerequisite, a gap that tends to stay invisible right up until it is not.
How to evaluate whether your own foundation is ready
For a retail or CPG CIO deciding whether to follow Bronson's sequencing or the more common ship-first pattern, the honest diagnostic question is not whether your data is perfect, no organization's is. It is whether you can trace an AI system's output back to the specific data sources it drew on, and whether someone in your organization is accountable for the accuracy of those sources on an ongoing basis rather than a one-time cleanup project ahead of a launch.
If that traceability does not exist yet, an ERP modernization or data governance initiative that looks unglamorous next to a competitor's AI headline may be the more defensible use of this year's budget, even if it produces fewer announcements. Boards asking for AI wins should be asked, in return, whether they would rather have a fast pilot built on data nobody can fully vouch for, or a slower rollout built on a foundation that will still be reliable two years and ten more AI use cases from now.
