Responsible AI in Japan: The Next Move for Insurance
責任あるAI活用先進国への転換 ― 保険業界における次の一手
Turning constraint into advantage.
Japanese insurers face two pressures at once. A shrinking workforce is making the current operating model harder to staff, and customer expectations are changing faster than legacy systems can follow. AI has meanwhile reached a level where it can take on genuinely complex, multi-step work rather than isolated tasks.
The paper argues that Japan's insurance sector is nonetheless stuck at an early stage, and that the reason is not primarily technical. Adoption stalls at proof-of-concept because of data readiness, unresolved questions about how roles change, uncertainty about how to judge return on investment, and—most often cited by the executives interviewed—the difficulty of changing mindset and organisational culture.
Its central argument is that the characteristics usually described as brakes on Japanese adoption may be the sector's most durable advantage. Caution, patience, a low tolerance for visible failure and a long-standing customer orientation are exactly the dispositions that responsible AI requires in a domain where decisions affect people's financial security. Treated as design principles rather than as delay, they position Japanese insurers to establish what good practice looks like in a high-trust industry.
For senior decision-makers, the paper's practical message is that the binding constraint is leadership rather than technology: the ability to commit under uncertainty, to set a direction before the business case is fully legible, and to redesign work rather than automate around it.
The question is not whether Japanese insurers will adopt AI, but whether they do it in a way that makes trust a competitive advantage.
Five findings from the research.
Culture, not technology, is the most cited barrier
The executives interviewed point to mindset and organisational culture—not technical capability—as the largest obstacle. Concern about the effect on employment is real, and in a market where long-term employment norms remain influential it slows decisions. The paper's response is to position AI as a tool that raises the level of human work, paired with explicit support for retraining and role change.
Proof-of-concept stalls on data, not ambition
Pilots fail to reach production largely because preparing raw data for AI use is harder than expected, and because the outcome depends on cross-functional cooperation established early. The paper recommends joint assessment of data quality by business, IT and data functionsbefore a pilot starts, and shared success measures agreed at the outset.
Legacy systems may be leapfrogged rather than replaced
Many Japanese insurers still run COBOL mainframes that cannot support modern APIs, with data divided by product line, business unit and region, and maintenance dependent on an ageing pool of specialists. Because current AI can operate across on-premise and cloud environments, the paper argues insurers may be able to move to an AI-oriented architecture without passing through the intermediate stages—drawing an analogy with Japan's early adoption of mobile payments on feature phones, ahead of smartphones.
Trust has to be designed in, not reviewed in afterwards
In a regulated, long-relationship industry the paper treats explainability, fairness and human involvement as design-stage requirements rather than post-hoc compliance. That means recording the basis for underwriting and claims decisions, keeping the decision process auditable, and deciding deliberately where a person stays in the loop—lighter for high-volume, well-defined claims; heavier where judgement varies widely.
Conventional ROI measures understate the value
Faster decisions, deeper insight and greater organisational adaptability are hard to quantify but material, and insurance product cycles are long enough that returns surface slowly. The paper proposes running short-term proof and long-term strategic investment as parallel tracks, and widening the measures to include quality, decision speed, customer loyalty and employee engagement.
These are the paper's own findings, summarised. The full argument, its examples and its interview material are in the Japanese edition below.
What the research implies for decisions.
The paper sets out an approach rather than a prescription. Four implications follow most directly from it.
- Lead before the business case is fully legible
AI's value is rarely clear at the outset, and waiting for certainty is itself a decision. The paper places the requirement on leadership: set a clear direction, connect technical and business functions, and involve employees in redesigning work around the customer rather than imposing it on them.
- Run three horizons at once
The paper recommends pursuing operational efficiency for early, demonstrable returns; customer-facing work that makes the value visible inside and outside the organisation; and a smaller number of exploratory bets on longer-term breakthroughs. Running all three spreads risk and keeps the organisation learning.
- Decide build, buy and partner deliberately
The paper's guide is to build where the work is differentiating or tightly regulated, buy where capability is commodity or moving fast, and partner to reach specialist or frontier expertise. It observes that many insurers adopt a hybrid: sourcing technology and talent externally while keeping governance and domain knowledge in-house.
- Start small, with people in the loop and stakeholders involved early
Scepticism left by earlier digital programmes is treated as a legitimate obstacle. The paper's answer is to begin with low-risk, high-volume standardised work using limited, trustworthy data, keep human oversight, and run alongside the existing process—bringing business, IT, external partners and regulators in from the start.
Where this page stops
This is a summary of published research, not advice on a specific organisation. The paper describes what the interviewed executives and the authors observe across the sector; it does not assess any individual insurer's readiness, and nothing here should be read as a regulatory or legal opinion.
Six chapters, four arguments.
The paper works through why AI matters now, where it accelerates insurance work, the practical obstacles to adoption, trust as competitive advantage, Japan's particular strategic opportunity, and what responsible adoption requires. Four arguments run across those chapters.
Where the value sits
How AI moves beyond efficiency into underwriting, claims and customer engagement, and what connecting long-siloed data actually makes possible.
Trust as competitive advantage
Why explainability, fairness and human oversight belong in the design of a system rather than in a compliance review after it.
The practical obstacles
Legacy platforms and fragmented data, the balance between automation and changing roles, security and privacy, and how return on investment is judged.
Japan’s particular opportunity
Where a cautious, consensus-driven adoption culture becomes an asset, and what leadership has to do to move work past proof-of-concept.
Written for the people carrying the decision.
Co-authored across two advisory firms.
Ichun Lai
Works with financial services organisations on accelerating responsible AI adoption. Has led more than forty transformation projects across Cigna, Tokio Marine, Goldman Sachs and SMBC, in finance, business management, strategy, internal audit and consulting roles.
Sakura Kawakami
Advises on business transformation, market entry and strategic partnerships in Japan and internationally, with more than 25 years in insurance, financial services and technology including senior roles at MetLife, Citigroup, Nomura Securities and Lehman Brothers.
The paper draws on interviews with insurance and technology executives and specialists, together with the authors’ own research.
Read the paper.
The paper is published in Japanese. This page is its English summary.
For a conversation about what the paper’s findings mean for a specific business, get in touch.
Structured work on a specific market question.
Published research sits alongside Eveil Intelligence, which is commissioned work on one organisation’s question rather than public analysis of a market.