How Accurate are AVMs?

TwentyCi
May 14, 2025

With the click of a button, you can discover the value of a property. Automated Valuation Models (AVMs) promise speed, scale and simplicity. However, it’s important to note that AVMs are only ever as good as the data they are fed. When an AVM is running on stale, sparse or skewed data, the outputs might look convincing, but accuracy can be questionable.

 

What is an AVM?

An AVM uses mathematical models and data analysis to estimate the value of properties. It draws on inputs such as recent sales data, property characteristics, location information and market trends to provide its valuations.

It allows estate agents, mortgage lenders, insurers and investors to obtain a quick, cost-effective estimate without the need for a physical inspection by a surveyor.

The AVM learns from real-world property data. The better the data, the more precise the output. If the AVM has access to up-to-date transactions, complete property records, granular location insights and accurate comparables, it can produce a reliable valuation.

In our previous blog, we discussed how AVMs and traditional valuations can co-exist and compared both valuation methods. In this blog, we are going to look into the importance of accurate data and why the reliability of an AVM hinges on the accuracy and quality of its input data.

 

Validating the accuracy of an AVM

How can we be sure AVMs are reflecting true market value? Here are the key factors to consider:

  • Accuracy – The information must be verified and free from errors and inconsistencies.

 

  • Recency – The AVM must use up-to-date information and be regularly refreshed.

 

  • Completeness – The AVM must use a comprehensive database to draw conclusions.

 

  • Representative – The AVM needs to cover a broad enough geographic range and property types.

 

  • Transparency - The AVM owner must be transparent about where the data comes from and how it is validated.

 

  • Testing – The AVM must be rigorously checked.

TwentyCi’s AVM

Let’s take a look at each of these factors with TwentyCi’s AVM in mind:

Accuracy

We are confident that our AVM is among the most accurate in the UK market – if not THE most accurate. To prove our exceptional accuracy, our AVM is within:

  • ±1% of actual prices paid for 20% of all properties. 
  • ±5% of actual prices paid for 74% of all UK properties. 
  • ±10% of actual prices paid for 90% of UK residential properties. 

Our overall median absolute percentage error is just 2.8%.

AVMs are powerful tools, but value estimates can vary which is why confidence scores matter. A confidence score is a number or rating that shows the user the reliability of the AVM's estimate and how much trust they can place in the valuation.

TwentyCi AVM uses an accuracy rating of:

  • Very High (VH)
  • High (H)
  • Medium (M)
  • Low (L)

With this in mind, in terms of TwentyCi’s AVM, 52.4% of all UK properties fall into the top two accuracy levels of Very High and High, with 47.1% in the next most accurate level of Medium. This means that just 0.5% of residential addresses fall within the Low accuracy rating, showcasing the precision of our AVM.

 

Recency

A strong AVM needs to be time-sensitive because market conditions and trends can change very quickly in the property world. To ensure our AVM is timely, we create a property value for every UK residential property, every week.

Our approach involves obtaining data on every sale price and every advertised price for UK properties. We then adjust all prices for inflation since the last time a completion or advertised price was set. We also make further adjustments for any hyper-local differences. For properties for which we do not have a sale or advertised price, we impute the AVM.

 

Completeness

What makes us so confident in the accuracy of our AVM is that it is measured across a vast volume of transactions, not just a selective slice of the market. We also have a data bank called DOMUS which holds over 400 attributes of properties including the property type, location, size and so on. All of these attributes are involved in the AVM calculation. We know this information for about 99.6% of all UK residential properties. With such highly granular data, adjusting prices based on comparable variables is easy.

We are also confident in the accuracy and completeness of our address database. We use AddressBase Premium for 100% accuracy using the Unique Property Reference Number (UPRN). The UPRN is a unique identifier assigned to every property in the UK helping to track property information across different systems.

We also hold an ‘as was’ history of each property’s AVM value. This historic valuations book dates back to January 2022. This means that historical values and their accuracies can be compared over time, allowing any risks involved in volatile AVM variances to be minimised. The historic valuations book is helpful for lenders who may wish to review past AVM results to identify any significant value changes, which could indicate the need for a full valuation survey. For insurers, they may wish to take a look at past valuations when analysing claims history. In terms of new house builders, the historic valuations book can be handy to see what happens to the new house build premium over time and where to set future pricing.

 

Representative

For an AVM to be successful, it needs to have enough data to compare. TwentyCi’s AVM has whole of market coverage which equates to 32 million UK properties. We’ve been producing AVM models since 2017 so not only do we have exceptional coverage but we’re experts in the AVM realm too.

 

Transparency

TwentyCi shares detailed insight into how our AVM operates and we are always happy to discuss it further with any client who would like to learn more. An AVM model must be transparent to allow its users to understand how the model works in practice and how it arrives at its valuation. TwentyCi can share in detail how the AVM works, including the information used to derive the model, any assumptions made along the way to valuation, and the methodology followed. This also includes the processing of risks.

 

Testing

We invest a lot of capital to continually develop and enhance our AVM family of models. AVMs can be benchmarked after sale to see if the predicted value compares well to the actual sale price. This is useful for identifying areas for improvement. TwentyCi's AVM benchmarking process is currently carried out monthly. AVMs also need to be calibrated, which means adjusting or fine-tuning the model’s algorithm and parameters to ensure the valuations are accurate. We will decide if any calibration is required, implementing the calibration within a week of identification so that it affects the following month's AVM values and accuracy ratings. Well-calibrated models tend to produce higher confidence scores.

 

Conclusion

You need clean, reliable data for accurate valuations. The accuracy and comprehensiveness of TwentyCi’s data ensure that our AVM provides a dependable valuation. Our transparency and regular validations to assess the stability of our results are what help make our AVM market-leading. Interested in finding out more? Get in touch.