AVM stands for Automated Valuation Model and these have been used by financial institutions for over 15 years worldwide.

Simply put, they have been designed to meet the needs of lenders, brokers and consumers and can enable faster and cost effective mortgage decisioning.

The basic premise is that certain properties are straightforward to value and there is a sizable benefit in using AVMs to do this. The more sophisticated the AVM, the more capable it is of replicating the human valuation process.

Every day we use predictive models to make decisions, from Google Maps predicting journey times, to the Met Office predicting tomorrow’s weather. All are used frequently to help make decisions, but you would never expect these to be 100% accurate – and an AVM is the same.

For the majority of properties in the UK, an AVM achieves very high accuracy and this delivers significant benefits to lenders. To ensure that lenders use the model safely we help them understand its performance and test it rigorously in real-life and stress conditions.

For an AVM to be a success, stability and robustness are as equally important as accuracy is, and the ability to demonstrate proven performance through the cycle is a must.

Successful users are serious about reviewing performance and establishing decision rules to determine when to use an AVM or when to get some extra help, normally in the form of a professional valuer. Valuing property can be considered both an art and a science and while AVMs favour a scientific method, there are always cases which require human perspective.

For example, using an AVM to value a £2m home in London may be unsuitable if the available evidence is limited, whereas terraced homes in a street of similar properties transacting frequently are prime candidates for AVMs, and are safely used even for higher LTV applications. This is only possible if the lender can truly rely on AVM performance.

Myth: An AVM can’t be as accurate as a surveyor

Even the most accurate AVM will have outlier results, and often these are the ones that are reviewed most intently.

The key to ensuring safe use is to understand the likelihood of these events happening and to set controls to keep these occurrences at acceptable levels, as well as ensuring the AVM is constantly improved to identify them.

Myth: If an AVM is too successful, it will eat its own tail

A common argument is that if an AVM is hoovering up massive amounts of data that wouldn’t otherwise have been generated by a surveyor, then it’s denying itself vital data comparisons.

However, AVMs are not yet in a place where they can’t be tested. While there are easy ways to value properties in the UK, there always remains a significant volume of ‘harder to value’ properties – harder for AVMs and humans alike.

Myth: AVMs don’t react to sudden changes in the market

Consider the hypothetical; if an AVM is only using historic pricing information and the market has dropped 20% over a single day, how can the AVM know? In short, the AVM would be exposed to the 20% drop because there is a gap between when the model updated and the data it’s fed – but that’s the same condition for a human being.

In reality, these scenarios are also unlikely – even the Global Financial Crash and Covid-19 pandemic didn’t cause the market to shift at such a pace and an AVM is more objective and never speculates.

Ultimately, all surveyors and agents are using existing data as a benchmark – we’re all in a similar position.

It’s critical the AVM ingests data quickly so that it isn’t exposed to a long period of ignorance. As we are owned by Zoopla, we benefit from a lot of daily insight, which we use to inform customers and update our models and data.

Theo Brewer, is the director of analytics and consulting at Hometrack