What AI Transformation Actually Looks Like

Our Starting Point (and Why It Still Wasn’t Enough)

Akur8 is a software company built on machine learning (ML). We pioneered the application of ML for insurance pricing. Our engineering team includes more than 100 specialists in machine learning and data science, and innovation sits in the company's founding premise. If any organization was positioned to adopt AI easily, it was ours.

‍

It was still hard.

‍

Even with those advantages, AI adoption did not come naturally. It took more preparation, more trial and error, and more organizational work than we expected. If you run an actuarial department and have found AI adoption harder than the brochures promised, that experience is normal. You are not alone.

‍

We started the way most companies do. We gave every employee enterprise access to a leading LLM, encouraged experimentation, and waited. Within months, usage was high and impact was shallow. While employees used it to draft emails and summarize documents, the deeper, higher value work stayed the same. It was clear that the problem was with us, not the technology.

The First Challenge: Our Own Processes

Even as a young company, we had well-established ways of building and shipping software, and several of them clashed directly with what AI made possible. One example made this concrete for us:

‍

A top-performing engineer had taught himself to use AI to remarkable effect. With the help of AI, he built a complete, valuable feature on his own in just a few days. The feature never shipped. The team’s code review process was designed for incremental work over weeks; his peers lacked the bandwidth to review the work as quickly as it had been built, so it was viewed with suspicion. The engineer was demoralized.

‍

This made it clear to us that when you plug AI into a process built for a slower pace, one of two things happens: either the AI-generated output gets rejected by a workflow it was never designed for, or the workflow bends in ways that introduce risk. In both cases, you do not get the benefit, and you may actively damage trust in the technology within your team.

The Second Challenge: Culture

Beyond processes, two patterns emerged repeatedly across the company:

‍

“I don’t trust AI”

This concern was legitimate, and our answer was an honest conversation about risk: which risks were real, which were overstated, who was accountable when something goes wrong, and what role humans should play in the process. Once we took the concern seriously and clearly defined the division of responsibility, trust improved.

‍

“I don’t have time”

In practice, this often meant something more specific: "I don't know where to start." Learning new tools while maintaining productivity in your current role is genuinely difficult. We addressed this issue through three mechanisms:

1. Small-team coaching instead of company-wide training days
2. Protected time set aside for exploration
3. A champion on each team to help late adopters integrate AI into their work

The lesson underneath it all is that the constraint is culture, not technology. And culture changes slowly, with explicit support and a clear signal from leadership that this is not optional.

Three Keys to Unlocking AI Impact

Roughly a year into the effort, we settled on three requirements for real impact:

1. Rebuild the process for AI. Don’t try to force AI into workflows designed for a different pace and way of working. The two are often incompatible, and the process has to change.
2. Redefine roles explicitly. Around 30 to 40 percent of roles at Akur8 evolved. Product designers gained the ability to go into code and make changes directly. Engineers spent less time on implementation and more on architecture and product decisions. Crossing these traditional boundaries requires deliberate role redefinition.
3. Approach adoption from both ends. Bottom-up experimentation needs to be complemented by a clear top-down directive that AI adoption is mandatory. One creates curiosity; the other creates expectation. Together, they help carry the entire organization forward.

The AI Transformation Team

To make the transformation real, experimentation and mandates were not enough. We needed a permanent, cross-functional team that brought together engineers, technical leads, business performance specialists, and sales operations to transform Akur8 into an AI-first company, one process at a time.

‍

We designed the team with two defining properties that helped it work across the organization.

1. A mandate from the CEO. Transformation initiatives are often deprioritized when quarterly targets loom or blocked by departments that perceive them as a threat. Backing from the CEO gave the AI Transformation Team the legitimacy to have the conversations necessary to move the company forward.
2. Departments stay in charge. The AI Transformation Team does not take over functions within departments. It co-develops processes and ideas alongside team members, then hands ownership to them once development is complete. The more teams feel like design partners, the faster they will invest in and adopt the transformation.

With these two principles, the AI Transformation Team can operate in both directions at once. From the top-down, it sets direction, uncovers possibilities, and builds the framework and tools to move forward. From the bottom up, it listens to teams, identifies strong ideas, and scales them to other areas of the business.

The Result

More than two years in, this work produced something concrete. In the second quarter of 2026, we launched Akur8 Agents: AI built specifically for actuarial workflows and embedded in our platform. In the next article, we will discuss the principles that went into their development.

‍

Felix d'Alançon is Chief Operating Officer at Akur8. Ludovico Capparelli is Head of AI Transformation at Akur8.

‍