Gustavo Pisani —
Artificial Intelligence Executive

I occupy the space between the people who decide the investment and the people who write the code — and I translate in both directions. That's what makes it possible to defend an AI case to a board committee in the morning and discuss chunking strategy and context limits with the team in the afternoon.

Most AI projects don't die from lack of technology. They die in the gap between the demo that impressed everyone and the system that has to run every Monday at seven in the morning, with real data, an impatient user, and someone answering for the bill at the end of the month.

My job is to close that gap. In practice, that means picking the few use cases that sustain a roadmap, designing the architecture that survives scale, making sure the data foundation exists before the model does, and keeping the cost-per-interaction calculation visible from day one — because that's what decides whether it scales or gets switched off.

I started as a .NET developer in 2011 and reached a director role with full P&L and around 200 professionals. I've worked in consulting, retail, financial services, and in a company I founded myself — and every step has a number attached to it.

I work on AI products in production, strategy and portfolio, architecture and technical presales, data as a foundation, alliances and go-to-market, and the P&L and ROI calculation. These aren't separate services: a badly chosen use case brings down the best architecture, and the best architecture without a data foundation is wrong with conviction.

I always start with the process that already hurts and already has history in the data, I measure a baseline before go-live, I define guardrails and PII limits as a requirement — not as a backlog item — and I keep the cost-per-interaction calculation visible from day one.

That order isn't a methodological preference. It's what separates a system that keeps running from a demo that impressed everyone and got filed away.

I start with pain that has history. A process that already hurts and has already produced data is the only starting point where the result is verifiable. Without history, there's no way to prove it improved — only to claim it.

Baseline before go-live. If nobody measured what it was like before, any number afterwards is an opinion. That's the difference between a result and enthusiasm about the demo.

Guardrails are a requirement. Closed scope, answers with a traceable source, and sensitive-data limits defined before launch. Treating that as later refinement is the shortest path to having the project suspended by legal.

The calculation from day one. Cost per interaction times volume, plus the team that sustains it. That calculation is what decides whether to scale or switch off — and it's better to know it before the board asks.

I write about this in more depth in the questions I answer and on LinkedIn.

Over my career, across different companies and roles, I've worked on projects in Digital, E-commerce, Marketplaces, Integrations, Data, Experimentation and — increasingly — AI, inside accounts in retail, financial services, manufacturing, healthcare and consumer goods.

The list below is of accounts served, at different times and under different arrangements. It does not imply a current commercial relationship or endorsement by those organisations.

I'm studying Artificial Intelligence Management at FIAP (2026–2028) and hold 14 AI certifications issued by Anthropic, OpenAI, Microsoft, Google and IBM — covering everything from model fundamentals and limitations to agents, workflows and the Model Context Protocol. I speak Portuguese (native), English and Spanish at a professional level.