Back to case studies

Industry

Real Estate

Project Type

Workflow Tool

Size

Midmarket

Delivering £150M in insights into the UK real estate market

£150M

Opportunities created

3.4x

Increase in client win rates

5+

New UK districts added for business

How we partnered with a midmarket firm in the real estate industry to build out a differentiating AI based workflow tool - greenlighting their way to success with an underserved but high value target audience.

Background

The client were looking for a way to differentiate their offerings and capacity in an increasingly commoditised market for residential real estate - where the shift to self-research has led to near 80% of searches originating at aggregators (e.g. Rightmove or Zoopla). When assessing buyer intent in this space - one of the most under-served markets but high value markets is those buying for investment purposes (e.g. rental yield or capital appreciation).

As most aggregators are designed for those looking to purchase or rent a home for personal habitation - most buyers in this space resort to a wide array of third party blogs and websites in order to conduct the research required in order assess opportunities - often with inaccurate (e.g. out of date or factually incorrect) data.

We decided along with their team, that focussing on building a workflow tool in this space enabling a simple input of investment criteria and an output of properties that fit the criteria would allow them to offer something completely differentiated from other agencies - enabling them to capture more leads and win more deals - as well as rapidly expand to regions of the UK that they didn’t previously serve through partnerships.

Forecasting property investment opportunities

Given the vast array of public data sources, we worked with their team of expert real estate agents in order to build a model that could forecast investment yields across a variety of different strategies commonly sought by the target audience including:

  • Buy-to-let: Where the property is purchased as is and placed directly onto the rental market - in both managed and unmanaged settings.
  • Renovation: Where the property is purchased, renovations are made and the property is then either re-sold or let.
  • Short-term let: Where the property is purchased in order to facilitate short term letting opportunities - usually in high tunrnover markets (e.g. tourist hotspots).
  • Multiple occupancy: Where the property is purchased as a single household and converted into a HMO.

In order to successfully guide this project and ensure value would be delivered, we focused first on a single regional sub-market - allowing a shorter time to prototype and ensure that our model could accurately be expanded, at a much lower cost than modelling all target areas up-front.

For this, we initially aggregated data from the Land Registry and property listing websites along with historical data that the client had - splitting homes by various data points including bedroom count, bathroom count, size, binary features (e.g. garden/no garden) as well as sliding scale measures (e.g. renovated in last x years, renovated to y standard).

We then blended and averaged this data at the post code level (in order to proxy for attractors and detractors that could influence yield such as local infrastructure, proximity to amenities and transport links and detractors such as crime rates, noise pollution etc).

Next, we bucketed each of the properties into a sliding scale (0-10) for each strategy in terms of how fit for purpose the property was given the strategy criteria - based on data provided by the client - and back-tested the model against historical data from 2011-2021, with a predictive accuracy rate of 94.5%.

This accuracy rate was calculated that when a given set of investment input criteria were given, suggested properties achieved or would be on track to achieve the target investment yield in at least 94.5% of cases - excluding external out-of-range factors e.g. COVID-19 pandemic.

From machine model to human interface

Once the prototype model was delivered, building an interface to make the model human friendly - allowing for both simple and advanced usage depending on the accuracy and specificity of the investment criteria.

This was rapidly built, with the timeline from prototype model to a prototype human ready interface taking less than 4 weeks - and the client’s team were delighted by the simplicity and the cross-platform capabilities delivered.

Automating repeat outreach

To get the most out of this tool - the client wanted to be able to represent/match prospective buyers with sellers who hadn’t yet instructed them to place their property on the market. Having a warm buyer lead up-front would make all the difference when approaching potential sellers.

For this, we built an automation capability combining together generative AI along with third party APIs which enable direct mail automation - in order to personalise outreach to existing homeowners at scale - taking a task that would typically take days down to just minutes, allowing agents to focus on high value tasks and human touch points.

Into production and beyond

We conducted a phase based roll out and testing strategy - using cloud-infrastructure located in the UK to meet data privacy regulations along with continuously working with the client to measure the impact.

As a result, within 6 months of launch - the client saw £150M+ of investment opportunities unlocked for leads, with a blended 3.4x win rate increase blended between target audience buyers and sellers and was able to expand to 5 new UK districts they previously didn’t cover.