The Intermediary – December 2025 - Flipbook - Page 64
T E C H N O L O GY
Opinion
Data quality is
central to effective
property risk
W
hen we speak
about new
technologies,
there is a
tendency to
use catch-all
terms. Artificial intelligence (AI) is a
good example of this
Within the world of AI there is a
whole range of models and learning
methods – supervised, unsupervised,
reinforcement learning, deep
learning, generative models and
natural language processing (NLP).
The same is true when we talk about
automated valuation models (AVMs)
– not all are equal. All AVMs rely on
data inputs and use an algorithm to
calculate the value of a property. But
depending on the provider, not all will
use the same data sources or number
of data points. Not all providers
triage AVM valuations either, yet
understanding and siing outliers
out into a more appropriate valuation
journey is critical.
Data used by most AVMs include
sources registering property
transactions, Land Registry, and
comparables. Some feed in additional
information relating to flood exposure
or other environmental factors
affecting a property’s value.
Understanding the various
models is important for lenders,
particularly because margins are thin
on individual loans that are relatively
straightforward on the credit risk side.
Bringing the cost of writing a loan
down therefore has a tangible impact
on profitability. The easiest way to
achieve this is to use cheaper valuation
models that keep underwriting costs
down. However, too heavy a reliance
on an AVM that doesn’t consider all
the property risk factors properly is
likely to prove a false economy should
things go wrong for the borrower.
62
The Intermediary | December 2025
The key to managing risk factors is
to understand fully how AVMs differ
across the marketplace, and to ensure
that checks and balances are in place.
Breaking AVM risk down into its
component parts is helpful when
considering how best to use them.
Data is absolutely critical. There are
several factors affecting the quality
of data put into AVM models, as
identified by the Royal Institution
of Chartered Surveyors (RICS). At
a high level, these include recency,
availability, security, privacy,
ownership and ethics, provenance
and lineage, assurance, consistency,
collection methodology and scale
and range.
Transaction volumes are also
critical, along with meaningful
comparables. As RICS notes, in many
jurisdictions this data is unavailable,
of poor quality or substituted with
asking price data as a proxy.
As planning and regulations evolve,
there is also an increasing variety of
data being used as part of the valuation
process. Data points around Energy
Performance Certificates (EPCs) and
the Building Research Establishment
Environmental Assessment Method
(BREEAM) are also increasingly
incorporated. Then, there are more
traditional data sources that have long
been used by surveyors to gauge value,
such as using crime rates to assess
the aractiveness or otherwise of the
location and region, online consumer
ratings, air quality and broadband
availability, among others.
It is also worth considering the
volume of data being fed into models
– and how reflective it is of the whole
market with all its nuances.
The most effective way to manage
these data risks is to put in place triage
measures that protect against skewed
information.
STEVE GOODALL
is managing director at e.surv
There are two ways to do this
effectively. The first is to use realtime data gathered on the ground by
experienced valuers to feed into AVM
models on a daily basis, ensuring that
the most accurate assessment of value
is achieved. The second is to review
automated valuations based on other
risk factors – again, identified by
surveyors on the ground.
By constantly upgrading the data
inputs with real, verified information,
lenders can have true reassurance that
property risks that sit underneath
the bonnet of open data sources are
identified and caught before they
become an issue.
Where properties are flagged as
potentially higher risk or anomalous,
desktop and physical valuations can be
used to provide additional cover.
It’s generally accepted that AVMs
are more appropriate to housing
that can be considered reasonably
homogenous – in style, construction
type, location and age.
Where a blended loan-to-value
(LTV) assessment is needed for the
valuation of a whole loan book, using
AVMs can be a good way to keep costs
down for lenders.
RICS has previously flagged
concerns around the fact that many
models are designed to work at the
portfolio level, and by not considering
comparable evidence at an individual
asset level, produce a different
statistical distribution. There is a
place for this in the market, but
what lenders must be mindful of is
the appropriateness of the valuation
method and model to the type of work
being carried out. ●
