Share with your CIO
Bradesco CTO Cintia Scovine Barcelos makes a pointed case about why most enterprise AI programs stall: companies treat AI as a software problem and skip the harder work of data governance, platform consolidation, and process redesign. The piece uses Bradesco’s decade-long BIA deployment as a reference arc, from isolated chatbot to institutional capability embedded across every business line. Kenyan banks KCB Group and Equity Bank surface as a secondary example of institutions facing the same scaling pressure. The core argument is that the enterprise execution gap is a governance and architecture problem, not a model problem.
What this means for your business
The failure mode described here is one of the most consistent patterns in enterprise AI: a portfolio of disconnected pilots, each technically functional, collectively delivering nothing because they share no data infrastructure, no reuse layer, and no unified accountability model. Whether this story is about you depends less on your industry and more on your org chart. If your AI initiatives report to business units rather than to a central platform owner, you’re almost certainly building the expensive, siloed mess Barcelos is diagnosing.
The argument’s strongest point is also its most underappreciated one. Governance isn’t a compliance checkbox sitting beside the AI program; it has to be encoded into the tooling that runs the program. A policy document saying “models must be explainable” does nothing when a credit scoring model flags a customer at 2 a.m. and no human is watching. The difference between a governance framework and a governance system is whether the controls fire automatically at runtime, not at the quarterly audit. Most enterprises are still running frameworks. That gap is where regulatory exposure lives.
The measurement section lands the sharpest practical claim: automated issue resolution rates, customer retention deltas, and cost-per-use-case are the metrics that actually tell you whether AI is working. If your current AI reporting consists of “number of use cases deployed” or “hours saved in pilot,” your measurement model is hiding failure rather than revealing it. The organizations that get this right will find AI stops being a line item to defend and starts looking like operating infrastructure, which is exactly when the CFO stops asking whether it’s worth it.
The falsification condition for this whole argument is Bradesco itself. If BIA’s decade of investment has produced measurable, auditable improvements in credit decision accuracy, fraud containment, and customer retention that outperform peer banks, the governance-first thesis holds. If those metrics are unavailable or unremarkable, the piece is a well-structured framework in search of proof. That’s the question worth putting to any vendor or consultant selling you a similar transformation roadmap: show me the before-and-after on a specific KPI, not the architectural diagram.
Based on reporting from From AI-First to AI-Powered: Bridging the Enterprise Execution Gap, originally published 2026-06-12 06:15:00.
