There are those who drift into data and AI, and there are those who were born to it. Luca Piccolo (pictured), data and ML products delivery lead at Intact Insurance, belongs firmly in the second category.
Growing up in Italy, Piccolo discovered computing through his father’s machine, tinkering and teaching himself to program long before he ever set foot in a university lecture hall. That early curiosity, he said, shaped everything that followed. “I found it a really powerful sphere,” he recalled of his introduction to machine learning, a technology that, even then, struck him as something qualitatively different from the largely deterministic computing systems of the time.
Piccolo went on to read computer engineering at the Politecnico di Milano, Italy’s foremost technical university, which counts engineers, architects and designers among its graduates, before completing a master’s in Artificial Intelligence and Robotics. His reasoning for pursuing AI was characteristically precise. Machines, he observed, were powerful but rigid. Machine learning gave them something else entirely: the ability to learn from patterns, much as humans do, “without the need to be directed in each and every detail.”
That theoretical grounding translated into nearly a decade in management consulting, where Piccolo deliberately sought breadth. But as the years passed, the work began to feel insufficient. Consulting, at its senior levels, tilts heavily towards business development. Piccolo wanted something else, not to advise on transformation, but to drive it. “I really wanted to be the person that drives the transformation, makes things happen, sees that change end to end,” he said.
The decision to move in-house brought him eventually to Intact Insurance. The draw was not the scale of the brief, but something more intangible, a leadership team that understood that genuine transformation is measured in years, not quarters. “I liked the team and I liked that ecosystem,” he said.
Ask Piccolo why he thinks certain data and AI projects deserve investment whilst others do not, and he is unequivocal. Data, he insists, should not set the agenda; the business should. “Data as a function exists to sustain company priorities and imperatives, and not the other way around,” he said. Any initiative that cannot be mapped directly to a strategic objective, he suggests, should not survive scrutiny, and rightly so.
The published priorities for Intact’s UK operations, including enhancing broker experience and expanding broker distribution, optimising underwriting and claims for outperformance, targeting a low-90s combined operating ratio, and building more responsive and agile technology infrastructure.
That philosophy extends to how Piccolo approaches the politics of prioritisation. Identifying the true strategic imperatives of an organisation is itself a skill, he argues. In some businesses they are documented and disseminated, in others, they exist only in the heads of senior leaders. Either way, unearthing them is the data function’s first job. “Someone will know what really matters at any given point in time. So it can be a bit of detective work, but it’s our job to find that out and align to it.”
Credibility, he adds, cannot be claimed at the outset; it must be earned. Early in a transformation, in his view, the data function is best placed in what he describes as a “follower service mode,” responding directly to business needs rather than arriving with a pre-formed vision. Proactivity, once that trust is established, can follow, but it has to be incremental. Going in with guns blazing, he cautions, tends to backfire: “You come across as someone that loses sight of what the business actually needs.”
Nowhere is Piccolo more incisive than on the question of how to structure a data transformation. The prevailing orthodoxy is to build the foundations first, then layer analytics, then dashboards, then AI. In his view, it is a recipe for institutional patience running out before value materialises.
“I’ve seen many times in the industry strategies that work through those layers incrementally: year one, build the foundation; year two, build KPIs and management reporting on top; year three, build the dashboards; year four, build the AI. In my personal opinion, it just doesn’t work,” he said.
The data profession, he observes, operates under constant pressure to demonstrate its worth. Not every senior leader fully grasps what data teams do. In that environment, a transformation roadmap that promises returns in year four or five asks too much of organisational goodwill. “That type of roadmap eventually leads to the same end result, but it takes many years to show an actual return, and by then, you’ve eroded the trust you needed in the first place,” he said.
His alternative is what he calls the “vertical stripe” model. Rather than building each technological layer horizontally across the whole organisation, select a single domain or use case, build every layer within it, from foundations to AI, and demonstrate an end-to-end return before expanding. “The cycle to generate the first returns is much faster that way,” he said.
The distinction, he is careful to note, is not merely technical. The human dimension is transformed. The horizontal approach leaves “everyone holding their breath for five years, hoping the end state is what they were promised.” The vertical model, by contrast, produces a story that can be told internally, something tangible, visible, and repeatable. Change management, he argues, becomes dramatically easier when there is already something to point to.
If there is one thing Piccolo finds professionally frustrating about the current moment in artificial intelligence, it is the conflation of the specific with the general. The explosion of interest in large language models and generative AI, including ChatGPT and its siblings, has, he said, distorted the wider conversation. “That is a rather narrow part of AI, although it takes all the attention now,” he said.
Traditional AI, including fraud detection models, actuarial and pricing algorithms and operational automation, has been delivering measurable value in insurance for many years. The value case is well-established, the technology mature. For these applications, the principal task is scaling and embedding, not evangelising.
Generative AI is a different matter. Piccolo is neither dismissive nor credulous on the subject. He acknowledges its potential but is candid about its current limitations: “Sometimes there are strong value cases, sometimes it’s more of a solution looking for a problem,” he said. The industry, in his view, is still learning to ask the right questions of the technology, and the more important long-term task may be to interrogate underlying processes rather than simply accelerate them. “Over time we will see even bigger transformation as organisations start to question: have we automated the right things or should we reinvent how we operate?” he asked.
On the question of consumers, he is reflective. Looking across industries, automation has, in some cases, served the organisation rather than the customer, and the imbalance shows. He describes his own experience navigating an automated call centre for an hour before a human agent resolved the matter in minutes. “In some scenarios, the balance hasn’t quite been struck right for the end consumer,” he said.
Away from the headlines, Piccolo is watching a quieter revolution in a corner of artificial intelligence that many practitioners have yet to seriously engage with: Causal AI.
The vast majority of models in commercial use today, he explains, are built on correlation, the identification of patterns that tend to co-occur. Many interpret these correlations as causal relationships, but the models themselves make no such claim. “The model is not telling us that; it’s just telling us, when A happens, also B happens,” he said.
Causal AI addresses this gap directly. By embedding cause-and-effect logic within its architecture, it can answer a fundamentally different class of question, not merely what tends to follow what, but what actually produces what. For an industry in which decisions around pricing, underwriting and claims carry significant financial consequences, the implications are considerable. “A lot of the phenomena are actually cause-and-effect phenomena,” he said.
Piccolo believes the field is at an inflection point, still not widely applied in mainstream commercial settings, but poised to exert significant influence. “I think it’s going to have a huge impact on the industry, also through areas that we’re not even exploring now,” he said.
Piccolo has led data transformations long enough to have encountered resistance in every form. His perspective on it, however, departs from the standard playbook. We are often taught a framing, he observes, that divides stakeholders into supporters and detractors, a binary, that, in his experience, can harden into a self-fulfilling dynamic.
The observation that shifted his thinking came from a colleague, offered in passing: “Most people just want to come to the office and do a good job.” It is a disarmingly simple sentence, but Piccolo finds it clarifying. Resistance, in most cases, is not wilful obstruction. It reflects a divergence in objectives and incentives, pressures that the change manager may not have taken the trouble to understand.
“When I feel something is getting stuck, the key unlocker is taking the time and making the effort to understand what motivates them, what their goal is,” he said. That understanding, he argues, opens the possibility of a solution that genuinely serves both parties, rather than one that simply steamrollers the objection. “Maybe they’re under pressure from their leadership for something different from what we’re focused on,” he said. “Once you understand that, you can stop pushing in a direction that clearly works against them, and things tend to open up.”
Co-creation is his preferred term for the alternative to top-down implementation. It does not require business users to write code; it requires them to be present, consulted and genuinely influential. “It’s no longer something being done to them; it’s something being done with them.” The effect on adoption rates, and on the risk of a completed solution being quietly shelved, is, in his experience, substantial.
When pressed on what he would change if he could revisit his career with his current perspective, Piccolo does not deflect. His strength, an almost compulsive passion for the work, is also, he concedes, his principal professional liability.
“I tend to run into the fire a little,” he said. “I have ideas, I get excited, and I want to do all of them.” The problem is that real transformation requires the opposite disposition, a narrow, sustained focus applied over a sufficiently long period to create genuine institutional change. An organisation can only absorb change at a certain rate, and dispersing energy across multiple initiatives dilutes the very focus that makes transformation stick. “Real transformation often happens when you manage to have a narrower focus and sustain that focus over a meaningful period of time,” he said.
The lesson he draws is not to suppress the enthusiasm but to redirect it, to maintain a longer parking lot of ideas and to resist the pull of the adjacent opportunity. “If I look back, I might have picked up fewer side projects along the way.”
On the question of where the industry is headed over the next three to five years, Piccolo is uncharacteristically cautious, but deliberately so. “Anyone who thinks they know exactly where this is going is lying to themselves,” he said.
Amongst the currents he does identify: the continued scaling of traditional AI applications; the gradual emergence of genuine generative AI use cases as organisations learn to distinguish useful applications from fashionable ones; the rising influence of causal AI, and, perhaps most unpredictably, the shifting behaviour of end consumers who increasingly turn to AI chatbots rather than search engines to gather information before making decisions. The commercial implications for distribution, underwriting and customer engagement remain, for the moment, genuinely opaque. “I genuinely don’t think anyone knows yet.”
When the working week finally relents, Piccolo’s instinct is to ground himself firmly in the analogue world. Family comes first. He has a young family, and time with them, he said, “helps me to remind myself of the world outside of data and AI,” he said. At weekends, he turns to cooking, reading and music to fill the quieter moments, and recently returned to chess, a game he has picked up again after some time away.
The choice of chess as a pastime is, for a man of his background, not without a certain symmetry. Piccolo is well aware that AI did not merely learn to play chess; it arguably transformed the way it is understood. He points to IBM’s Deep Blue, historically one of the earliest high-profile demonstrations of AI capability, which famously defeated world champion Garry Kasparov, and to Google’s AlphaGo, which achieved the same feat in the far more complex Chinese game of Go some decades later. “It’s the perfect environment for a computer,” he noted of chess, “because there are clear non-ambiguous rules, there is no hidden information and a computer can look ahead indefinitely.” Expert players, he adds, have observed that AI has opened up entirely new avenues of exploration in the game, testing approaches that human intuition had long considered inadvisable. For Piccolo, it may be the perfect after-hours companion, a space where human creativity and machine intelligence work in harmony.
