The Intermediary – October 2026 - Flipbook - Page 78
T E C H N O L O GY
Opinion
Why insurance AI
gets stuck in pilots
I
nsurance has no shortage of
artificial intelligence (AI)
pilots. Across underwriting,
claims, fraud, customer
service and distribution, firms
have shown that generative AI
can summarise documents, surface
information and support decisions.
The more difficult question is what
happens aer the demonstration. A
model working well in a contained
environment is very different from
AI operating reliably inside a live
insurance process, connected to legacy
systems, sensitive customer data
and decisions that must be explained
and audited.
Research conducted by NTT DATA
among nearly 300 insurance leaders
found that the organisations most
advanced in rolling out AI-powered
services tend to enjoy stronger
financial outcomes.
Importantly, our measure of AI
maturity combined the presence
of sophisticated AI strategies with
enterprise adoption and measurable
return on investment: we were
looking for organisations making
progress on execution, rather than
simply access to the latest technology.
Make AI move faster
Governance is oen discussed as
though it inevitably slows innovation;
but in insurance, the opposite can
be true. Our research found that
67% of the most advanced insurance
AI organisations use a centralised
governance model, compared with
24% of those at the other end of the
maturity spectrum.
Clear ownership, common
standards and defined approval routes
can remove the uncertainty that
keeps promising technology trapped
at the pilot stage. This becomes more
important as insurers move towards
agentic AI. An AI assistant might
retrieve information or recommend
a next step, but an AI agent can
potentially complete several steps.
76
The Intermediary | October 2026
As AI takes on more of the process,
insurers need to be clear about
where it can act on its own and
where a person still needs to make
the decision. Those boundaries need
to be agreed before AI is rolled out
more widely, particularly in areas
where decisions affect customers,
such as claims handling, pricing,
underwriting or determining
eligibility for cover.
Architecture decides
The technology underneath the
model maers just as much. Among
the advanced organisations in
our research, 58% are rebuilding
applications with AI embedded
directly into core systems, compared
with 7% of the least mature group.
That difference is particularly
relevant for insurers operating
complex estates spanning policy
administration, underwriting, claims
and distribution.
Adding an AI interface to an
existing process may produce a useful
proof of concept, but it does not
automatically change how work gets
done. To create sustained value, AI
needs access to the right institutional
data and a route into the workflows
used by employees and intermediaries.
For brokers and distribution
teams, that could mean faster access
to information, more responsive
servicing or beer support around
customer and partner interactions,
rather than another disconnected tool.
Measure value
The strongest projects are usually
those where there is a clear business
problem, a repeatable workflow and
a measurable outcome. Submission
intake, document collection, claims
triage, fraud investigation and
renewal processes involve high
volumes of information and multiple
steps, but still allow firms to define
where human judgement and
accountability should sit.
ALICIA MAXIE
is senior director, global
insurance strategy and advisory
at NTT DATA
Where organisations match up
their AI and business goals, we can
see the results. Our research found
that 85% of insurers with fully aligned
AI and business strategies reported
a profit increase of at least 5% from
AI. By contrast, nearly half of those
without alignment failed to realise a
meaningful impact on profitability.
The lesson is not that every AI
project needs an immediate revenue
line aached to it, but that insurers
need to be clear about the business
outcome they are trying to improve
before deciding what to automate.
Experimentation to operation
The next phase of AI in insurance
will be less about how many pilots an
organisation can launch, and more
about how many useful capabilities it
can put into large-scale production. As
models become more widely available,
almost every insurance provider is
experimenting with the technology.
The new differentiator is between
those able to navigate the tricky task
of integrating it with the systems,
data, controls and people that make
an insurance business work, and
those stuck on an endless loop of
unsatisfactory pilots. That makes
governance and architecture just as
important to AI-powered growth as
mastery of the technology.
The firms that can establish
clear accountability, modernise the
workflows that maer, and connect
AI investment to measurable business
priorities, will be beer placed to move
beyond experimentation.
Agentic AI raises the stakes, but it
also makes the opportunity clearer:
the competitive advantage will come
from puing AI to work safely and
repeatedly, not simply proving that it
can work. ●
