Can PropTech solve the biggest hurdles in residential property valuation?
As discussed in our previous blog post, data has been embraced by virtually every industry, and the world of property should be no different.
Property data is a resource that is only growing in demand. Yet despite this, there is still no single definitive source to access it.
So, why is it being left behind?
The first hurdle: the sheer volume and cost of data
Property analysis is only as reliable as the quality of the data on which it’s based. Yet, the landscape of residential property records consists mainly of extensive collections of incomplete data in multiple locations and inconsistent formats, making searches tedious and inefficient.
Even if all data on a single property was available, in order to value that property accurately it would be essential to consider its neighbourhood. That consideration requires precise, up to date information on its street and surroundings, including property type, size, tenure and transaction history.
Worse, this data is often not available for free. Commonly used address finders, including the Post Office and Google Maps, permit only limited searches with substantial fees thereafter. If you were to collect data on the 30 million UK properties on Google Maps, it would cost over £40,000!
The second hurdle, extracting property characteristics from this data
The most basic data points needed to begin a property evaluation are property type (house, flat, bungalow), size (floor area in sq ft) and configuration (number of bedrooms, bathrooms, reception rooms).
Data on these can be found on H.M. Land Registry (property type and transfer history) and Energy Performance Certificates (EPCs) (type, floor area and habitable rooms - an amalgam of bedrooms and reception rooms.)
Out of the 26.9 million residential addresses in England and Wales, EPC records and H.M. Land Registry together cover only:
20.3 million property types (75.5%)
15.7 million floor areas (58.3%)
And there’s more, the third hurdle comes in the form of reconciling these records
In order to unify the data from these records, software is needed. However, different sources of data use different address formatting. Whilst minor differences in an address are very obvious to the human eye, attempting to match up these addresses is a significant challenge for computers. This is where more issues arise.
Instances where machines can successfully match data to the correct property without any manipulation decreases data coverage drastically to:
14.6 million property types - 54.2%
9.2 million floor areas - only 34.3%!
(Out of 26.9 million addresses)| Type | Floor Area | |
|---|---|---|
| Unmatched Address (EPC & H.M Land Reg) |
20.3 million (75.5%) |
15.7 million (58.3%) |
| Matched Address (EPC & H.M Land Reg) |
14.6 million (54.2%) |
9.2 million (34.3%) |
The solution
DomusView believes PropTech offers the solution.
We aim to be the first tech company to solve the issue of fragmented and disparate data. We are acquiring all pieces of the puzzle to ensure you have the full picture at your fingertips – we call this Property Intelligence.
How do we do it?
It all comes down to a combination of two effective solutions…
Step 1) Reconcile all available data
As mentioned previously, the integration of publicly available data may sound simple, but address matching poses a significant challenge for computers.
But, after months of development, our data experts have cracked the algorithms to solve this.
These algorithms perform intelligent data matching - linking data extracted from public records and assigning it to each property.
For example, an address picked at random from H.M. Land Registry, ‘Flat 9, 5-7, Belgrave Gardens, London, NW8 0QY’
displays as ‘Flat 9, 7 Belgrave Gardens NW8 0QY’ in the Postcode Address File (PAF) database.
Whilst obvious to the human eye, the additional presence of the house number ‘5-7’ and the city cause computers to struggle to recognise these two addresses as the same.
Our algorithms overcome this issue.
They successfully match millions of addresses like these and bringing together all records into a complete data table - a single source of truth - making it easy to deliver fast, actionable insights on any residential property.
Step 2) Identify the gaps and fill them in
Acquiring all the available data is vital to begin this process, but the data cannot offer a complete picture of a property if it is incomplete.
We have pioneered technological methods to generate this unknown intel, extracting the raw data and applying proprietary models.
Put simply, by employing deep learning methods, we are in the process of collecting geospatial data for all 26.9 million residences across the UK!
What does this mean?
We create digital representations of properties based on satellite imagery
We then use these digital representations to compare properties with no available data with properties with data
By finding the most similar shape and size of properties with data we can estimate the dimensional and configuration data for a property with no available data
This isn’t perfect but it is a very good estimation technique that we can validate
Better yet, with more data and the passage of time our estimation improves as we learn where our model works less well and make improvements
So, we’ve figured out the solution, but how far have we progressed?
Through developing these methods each day our coverage grows, and in turn, the accuracy of our data increases.
Stay tuned for the next post tracking our advancements so far…