At IBM Think 2026, I had a great conversation with Javier Cassini, Global Managing Partner for Hybrid Cloud and Data at IBM Consulting. We had just returned from Constellation’s Futures Forum conference. There I spent two days with 120 CEOs and board members. The theme was unmistakable. Metrics are now real, and the pressure to scale AI is no longer theoretical.
Javier confirmed what I was hearing on the ground. The questions have fundamentally changed. Customers no longer come to IBM for use cases. They want to earn dividends, and they want to do it on an industrial scale.
Three patterns that produce real results
Javier outlined three profiles that he sees consistently across IBM’s customer base.
- The first one is Productivity at scale. Instead of a 10-20% efficiency gain, you’re looking at a 40-60% reduction in operational costs, depending on the domain. The key insight here is that simply layering AI on top of existing steps won’t get you there. Entire workflows need to be redesigned from first principles. Finance transformation, order-to-cash, supply chain, and software development lifecycles. Everything will be reimagined around AI, not as an afterthought.
- The second one is speed of decision making. I love this framing. Because velocity is more than just speed. It has direction. That means collapsing decision trees, automating steps that previously required committee meetings every Friday, and doing it with better context and data than a human could manage alone. Very many processes are not expensive in terms of headcount, but are slowed down by friction at certain points. That friction is an opportunity.
- The third is perhaps the most strategic. net new revenue. Mr. Pearson was one such example who took to the Think stage. How do you build entirely new AI-native products and business models? Javier makes it clear that the second pattern is usually needed to fund the third. But boards are increasingly seeing this as existential. If you don’t learn and build your flywheel now, you could fall behind your competitors without realizing it.
Why sophisticated teams still struggle
This is an area that I think is too often left vague in conversations. Even teams with a strong data foundation and clear executive sponsorship will hit a wall. Javier’s perspective: It starts with security, trust, and governance. If you don’t invest in these early on, you’ll run into massive problems you weren’t expecting in a clean MVP environment.
He also explained a data preparation framework that I found really helpful. Most teams think about data in terms of how accurate it is and how structured it is. That’s layer 1. But you also need a semantic layer (what does the data actually mean?), a context layer (what does it mean in this particular situation, this market, these rules?), and a decision layer where feedback loops capture what actually happened and increase the model’s autonomy over time. The last part, the learning loop, is what separates organizations that earn compound returns from those that plateau.
Operating model is the real bottleneck
Javier said something near the end of our conversation that I’ve been thinking about ever since. The bottleneck has moved. It’s no longer a technology issue. It’s a question of operating model. The question is not whether the model is good enough; they are. The question is how quickly organizations can transform, how quickly they can lead teams to new ways of working, and whether leaders understand the art of the possible enough to provide true direction.
CEOs are acutely aware of this. At the Futures Forum conference, two things were top of mind: whether I had the right people and whether my organization could actually evolve beyond the standard profitability and productivity goals. Javier is seeing a proliferation of use cases that can never scale because there is no single top-down commitment to a single vision. Organizations cling to what they’ve already built and water it down, even if they haven’t reached scale yet.
What does success actually look like?
One of the metrics that Javier tracks is the level of agency. In other words, how much work can you delegate to an agency versus keeping humans in the loop? The model he described is a person on the edge. Set your direction, define what you want, delegate execution to agents, and expand its reach over time as reliability and observability increase. Autonomy is not a switch. It’s a dial. And rather than expecting the model to figure it out, build toward it through evaluation, feedback loops, and intentional governance.
Interestingly, the organizations that get there fastest tend to be in regulated industries. They already know exactly what humans can and cannot do. Clarity is a competitive advantage when designing AI systems with appropriate guardrails.
The conversation could have continued for another two hours. We’ve just scratched the surface when it comes to governance trade-offs and moving from automated decision-making to trusted execution. More on that in a moment.
