UnBuild AI - Questione di Leadership

Summary

Marianna Leoni describes the shift from early generative AI experimentation to using AI to redesign business processes. Her central point is that projects often disappoint not because models cannot perform, but because organizations lack good data, governance, context, and a clear deployment plan.

Adoption and leadership

• Adoption has moved through two broad stages: an initial wave of post-ChatGPT experiments, often small and isolated use cases such as document summarization, followed by a more concrete focus on scaling AI across entire processes.
• Treat AI as a business transformation, not just an IT initiative. Leoni says CEOs and senior leaders need to decide where AI can create value, where it should not be used, and how its outputs will be controlled.
• Leoni describes three leadership responsibilities:
  1. Embed AI in everyday work and change how employees perform tasks.
  2. Use AI strategically to reshape the business and determine where to invest.
  3. Control quality and outcomes, including helping less-experienced employees recognize when AI results are unreliable.
• Leoni cites studies suggesting that at least 70% of CEOs effectively act as their company’s “Chief AI Officer” by taking responsibility for AI strategy.

Frameworks and operational playbooks

• Move from isolated use cases to end-to-end process redesign. Rather than inserting AI into one step of an existing workflow, assess whether the whole process can be redesigned.
• Build governance before scaling. Track pilots and proofs of concept, set priorities, define controls, and establish the processes and assets needed to operate AI reliably.
• Improve inputs before expecting better outputs. Organize internal documentation and data so systems can retrieve and use the right information.
• Pair efficiency with growth. Automate repetitive work where useful, then redirect employees’ time toward higher-value work, new services, sales, and cross-selling.
• Assess whether to build or buy:
  • Consider internal development for strategically distinctive capabilities tied to the company’s core business and intellectual property.
  • Prefer market solutions for commoditized, specialized problems where existing products are effective and scalable.
  • Evaluate the actual process and value created—not just whether AI makes coding or development cheaper.
• Assess legacy-system modernization based on the starting point. One possible route for banks is to add AI layers over legacy systems first, then gradually modernize or dismantle underlying components.

Risks, skills, and organizational change

• Common causes of failure include poor-quality or disorganized input data, weak governance, uncontrolled pilots, and deploying systems without sufficient business context.
• Leoni recalls large North American banks losing control of numerous pilots and proofs of concept. They paused work to create committees and prioritize applications, and the resulting governance reset cost time and momentum.
• Contextualization and communication are critical skills. People need to explain the task, relevant information, and intended use clearly; technical and coding skills alone are not enough.
• A financial-services chatbot example illustrates this point: a tool intended to help a help desk answer branch questions performed differently depending on the user. Branch staff got better results when they provided more context about the process and situation.
• AI changes roles rather than necessarily eliminating all human involvement. For example, employees who previously entered data may instead check whether machine-processed data is correct. Human review remains important, particularly in regulated settings.

Value creation and examples

• Leoni argues that companies need both cost efficiency and revenue growth; efficiency alone may leave them standing still.
• In model development, she says about 60% of the time was spent organizing data and preparing documentation, rather than designing the model. Automating or accelerating this preparatory work could help teams deliver models faster.
• She cites pricing, churn, and risk models as examples where faster development can matter, especially when businesses need to respond quickly.
• In lending, speed can be a competitive advantage: a bank that responds first to a customer’s credit request may win the business. However, legacy core banking systems can complicate major internal rebuilds.
• Neobanks and challenger banks may be better positioned to deploy AI because their data and platforms can be more accessible for these applications.

Metrics and targets mentioned

• 70%: Share of CEOs Leoni says studies identify as effectively responsible for AI strategy.
• 60%: Estimated share of model-development time spent organizing data and preparing documentation.
• No specific revenue, margin, adoption, or implementation targets or deadlines are provided. Leoni emphasizes that meaningful results require adequate investment; small, overly cautious efforts may not deliver significant outcomes.

Presenters and sources

• Marianna Leoni is the guest. She is introduced as having worked at Accenture, as a managing director and partner at BCG in financial services, and as CEO at Ionsulting.
• Gabriele, CTO and co-founder of Buildo, and Federica, introduced as a strategic consultant, host the podcast.
• The hosts also refer to a prior episode with Luca Foresta.