Only recently has a shared understanding begun to take hold: deploying AI is not the same as creating value from it.
As more people use AI in their daily work, this distinction has started to resonate as a lived reality. AI performance depends on the quality, accessibility and governance of the data behind it.
In insurance, this is especially visible. Data has the potential to reshape the core of value creation itself, from underwriting and sales to claims assessment and payment. As Akio Yamaguchi, President and Representative Director of IBM Japan has noted: “Data is also a management resource, alongside people, assets, and capital.”
That framing points to a broader shift toward what I call data-driven management.
The idea is not simply that companies should “make decisions based on data.” That has been the aspiration for decades. Equally, a company does not become data-driven simply from a modernized stack, sophisticated dashboards or advanced AI tooling. It becomes data-driven when executives and frontline employees alike begin to actually make decisions differently because they trust the data enough to use it. This requires three things to happen: the first is ensuring that data is prepared in a state that can be safely used by AI. The second is an environment in which user-friendly tools and workflow processes are designed and implemented so that data and AI are naturally embedded into the flow of day-to-day work. The third is a corporate culture in which behavior change grounded in the objective facts revealed by data is actively encouraged and put into practice.
Where transformation really begins: People readiness
When I joined Sompo’s newly established Digital Transformation Promotion Department in 2021, one of my first steps was to set up a weekly one‑hour open consultation session — “Akiko’s Room” — where anyone could discuss digital challenges. What emerged from these conversations was striking: while people spoke about what they wanted digitalization to achieve, data was rarely mentioned. These conversations revealed that many departments were still not actively thinking about their own data as a strategic asset. Therefore shortly after launching these consultation sessions, I established a Data Center of Excellence (CoE) dedicated to data analytics. I asked people who had long worked with data in areas such as product design to join, and together we began advancing data democratization.
We started by making data on more than 20 million customers usable in employees’ daily work. All employees, including the CEO, now access a single shared database through the “SJ‑R Dashboard,” enabling department‑level and company‑wide comparisons. Previously, data was managed in isolated spreadsheets, offering little visibility into relative performance — even between neighboring teams. By visualizing data and enabling quantitative comparison, a foundation was created for evaluating performance in practice and discussing direction based on facts. Data that had previously been siloed within individual departments became visible company-wide through a shared dashboard, enabling decisions and discussions to be conducted in a common language.
Measuring decision quality, not tool usage
Dashboards, however, are not an end-point for transformation. What they do is show an employee what is happening. To facilitate a data-driven culture, that employee needs to also be able to ask why it is happening, what action should follow and what additional data is needed to improve the next decision.
Once people begin to formulate questions, they are able to identify what data is missing to make sound decisions. This awareness drives further improvements in data quality and creates a virtuous cycle of progress. Greater transparency, in turn, strengthens the organization’s capacity to ask better questions, establishing a culture of change rooted in the data itself.
In our own business, one example is in sales transformation. By combining internal data visible through dashboards with external data, we are reexamining how sales activities are conducted. In the past, the use of external vendor data tended to be an individual skill of high performing salespeople. But when that skill remains trapped in individual capability, it’s more or less unscalable. To make this process data-driven requires the opposite: figuring out what works, then codifying and making it broadly usable so that these practices can be replicated across the business. And it is here – when applied well – that AI can help turn isolated expertise into organizational capability because it can make previously invisible practices visible, reusable and improvable. And once that happens, data becomes both a way to improve existing workflows and imagine new ones. In insurance, this could mean rethinking how products are designed, how risks are assessed, how customers are supported or how claims are resolved. However, the idea of ‘applying AI well’ is an important one. Organizations are still not thinking through from the outset the most basic question of what exactly they want AI to achieve. The solution from my experience is to focus on the use case and set a clear objective so it is easier to envisage the expected return, judge the level of investment and the expected ROI. One example is our Building Smart Estimate service. When a damaged part of a building affected by an accident or natural disaster is photographed using a smartphone app, AI analyzes the image and automatically generates a repair estimate. Previously it had taken an average of two weeks to calculate insurance payments, but can now be completed in about one hour. In introducing the service, we of course took customer convenience into account. But we also looked at usage rates and set operational KPIs, such as reducing the burden on frontline staff by around 30 percent, enabling them to redirect capacity to other work or manage operations with fewer people. In other words, we considered from the outset how the service could improve efficiency in the face of labor shortages. We did not achieve everything, but I believe that by designing these objectives at the beginning, it becomes much easier to identify areas for improvement and make sound investment decisions.
This demonstrates clearly how AI-enabled data can lead companies to new business models.
Greater transparency, in turn, strengthens the organization’s capacity to ask better questions, establishing a culture of change rooted in the data itself.
Building a fit-for-purpose data foundation
Of course, data-driven operating models depend on data that the business can use with confidence. The requirements that matter the most here are quality, accessibility, AI readiness and governance. Much of the work of a Chief Data Officer like myself lies in holding those requirements in productive tension. After all, data that is highly controlled but unusable does not create value, nor does data that is widely accessible but poorly governed.
That balance becomes especially critical when AI is deployed at scale, because AI magnifies both the value and the flaws of the data environment around it.
In January 2026, SOMPO Holdings introduced an AI agent tool to approximately 30,000 employees across its domestic group companies.1
The goal was not only to maximize operational efficiency and productivity, but also to empower every employee to use AI as a “powerful daily partner” – a step toward transforming how work itself at the company is organized.
From my perspective as Chief Data Officer, the more important question was not whether employees would use the interface. It was whether the data behind the interface was accurate, permissible and actionable enough to support better decisions.
In a highly regulated industry such as insurance, AI is only as reliable as the data environment in which it operates. One principle I have enforced is that only data reviewed and cleared by humans should be made available to AI. This is what fit-for-purpose governance looks like within our particular context: not compliance as a separate function, but practical management of how data is created, classified, accessed and reused. In our case, that has meant strengthening controls around lineage, ownership and usage. For instance, individuals handling the data are required to attach data‑lineage labels — documenting who created the data, from which sources, for what purpose, for whom, and how — when converting sensitive contract or claims data into statistical datasets. These labels are centrally maintained and searchable via a shared data catalog.
This may sound procedural, but it has direct implications for business outcomes. For example, claims decisions require reference to policy wording. If AI cannot immediately reach the correct policy terms linked to a contract number, it may generate incorrect answers based on outdated terms from a decade ago. In this way, accurate results are only possible when data is properly prepared.
Success still lies with people
Once AI begins to play a role in decision-making, its judgments are shaped by the data it is given and the criteria on which it is trained. That is precisely why organizations need designs in which humans remain firmly in the driver’s seat. And to mobilize the organization around that effort, it is essential to share with employees a vision of the future that they can genuinely believe in. Discussions of AI and data inevitably become technical, and as a result, it is easy to end up in a situation where something is explained correctly but still not understood. Correctness that is not understood carries the risk of never being implemented. That is why, at the outset, it is sometimes better not to begin with the details, but instead to help people picture how their work will change, what the new landscape will look like, and what kind of value it will enable us to deliver to customers. In other words, what matters first is building a shared sense of how things will change. I myself am still on that journey, but I believe transformation begins not with “correctness,” but with a vision of the future that people can accept and make their own.
Have questions or want to learn more?
