The Intermediary – November 2026 - Flipbook - Page 94
T E C H N O L O GY
Opinion
Widen oversight,
don’t dilute
accountability
C
onversations in the
mortgage market
are increasingly
incomplete without
a nod to artificial
intelligence (AI).
For the role that we have, it’s less
about the front-end interface with
consumers or even brokers. The
question now is where emerging large
language models (LLMs) and agentic
AI can support due diligence, file
review, and oversight at the back end.
What we are starting to see is a more
nuanced answer. In short, the most
useful role for advanced technology
in lending may be to make oversight
more complete, rather than to make
lending decisions automatic.
Technology can help firms build a
clearer view of whether a loan book,
process, or customer journey operates
as intended. It can bring greater
discipline to verification, while
leaving consequential judgement with
the people responsible for it.
Mortgage lending has never been
a purely mechanical exercise. A loan
can meet a set of parameters and
still require careful interpretation.
A customer’s circumstances may
introduce new considerations. The
task is not simply to collect more
information. It is to use information
well and act appropriately when
something needs aention.
Beyond manual sampling
Historically, much assurance activity
has been limited by the practicalities of
manual review. Firms select samples,
work through documentation and
investigate what those samples reveal.
This remains a valuable discipline,
especially where experienced
reviewers test the quality of a decision.
However, sampling can only ever
offer a view of part of the population
92
The Intermediary | November 2026
at a point in time. As portfolios grow
and roduct ranges broaden, the gaps
between periodic reviews become
harder to ignore.
Technology offers the prospect of a
different kind of coverage. Properly
designed tools can draw together
information from approved systems,
identify inconsistencies and bring
exceptions to the aention of a
reviewer sooner.
The value lies in helping firms
look across a wider population more
consistently, rather than in treating
every exception as evidence of failure.
This should change the work
of assurance teams and credit
professionals, but it should not
diminish it. A flag is a starting point
for investigation, not a conclusion.
Someone still needs to establish
whether the underlying information
is reliable, whether a policy has been
applied in the right spirit and whether
a particular customer has been
treated fairly.
The most important questions in
lending are oen contextual. They
call for professional judgement and
the confidence to challenge a process
which appears acceptable on paper.
The limits of AI
The same is true of generative AI.
There are clear areas where these
tools may reduce the burden of
repetitive work.
They can help compare large
volumes of documents and support
the preparation of material for
human review. Used within
controlled environments, this
could make reviews faster and
more comprehensive.
But there are limits which firms
should acknowledge from the outset.
Sensitive customer and corporate
information cannot be treated as raw
JOHN BARBOUR
is senior director
at Broadstone Banking
and Credit Advisory Services
material for ungoverned external
platforms. Nor should a model’s
output be allowed to become an
unchallenged credit decision or a
diligence conclusion.
Firms need to understand the
provenance of inputs, the role the
tool is expected to play and the checks
which apply before an output affects a
customer or a funding decision.
Consumer Duty reinforces this
point. Outcomes monitoring is
not satisfied by producing more
management information.
Senior leaders must be able to
explain what their information
shows, identify emerging risks and
demonstrate that their response has
improved the customer experience.
Technology can make it easier
to detect paerns or prioritise
cases, but it cannot determine
whether an outcome is fair in
the full circumstances of an
individual borrower.
Balancing innovation
For mortgage businesses, the sensible
path is likely to be incremental. Firms
should start with defined tasks where
the purpose is clear. They should
test how well outputs perform, set
boundaries around data access and
establish escalation routes before tools
are deployed more widely. Reviewers
must retain the authority to challenge
both the data and the system’s
interpretation of it.
The sector does not need to choose
between innovation and control.
Its strongest use of technology will
combine broader evidence with
accountable decision-making. ●