Case Studies / 10 September 2018

Pushing new boundaries in handling plans.

Establishing the correct legal extent and boundaries of a property is essential to delivering fast, accurate search results in conveyancing. Landmark has deployed AI to solve some of the challenges of handling manual plans, shortering handling times and reducing queries.

Example

Problem.

Manually uploaded plans can take longer for search providers to process due to manual interventions and administrative hand-offs.

Goal.

Digitised boundaries allow for faster processing (and often instant completion) of conveyancing searches. Develop an automated approach to handling map documents to trigger workflows with minimal hand-offs. Reduce delays, improve accuracy and enable faster processing.

Results.

Using a combination of image prediction and character recognition services, we successfully delivered a service that identifies official Land Registry Title Plans – of varying styles and quality – and accurately digitises sites, while adding unique capability to locate new build sites and plots.

Technologies.

  • Microsoft Cognitive Services, Custom Vision API, OCR
  • Image hashing and prediction services
  • Azure Storage, Cosmos DB

Background.

Landmark Information Group’s portfolio includes Landmark Legal and Environmental Reports, Argyll Environmental and SearchFlow – search providers who all deliver services to conveyancers.

Over a third of requests to our businesses attach PDF plans or scanned image files showing the outlined extent of the property. These boundaries are re-mapped by operations staff to our production systems, triggering workflows that ensure accurate search results are provided.

At peak times, requests can be queued for processing, especially those handed off to third party providers such as local authorities and water companies. This is less efficient than requests placed via our websites with a digitised boundary, which are instantly processed via integrated order APIs, allowing searches such as environmental reports to fulfil.

Objectives.

  • Reduce admin time/lag, decrease manual interventions
  • Create leaner processes for digitising sites
  • Make it easier to accurately locate new developments
  • Improve customer experience and enable self-service

AI Solution [#01].

We leveraged several new technologies, our existing property data service plus HMLR’s National Polygon Dataset [NPD] to create an automated workflow that handles multiple use cases.

In the first scenario, the service machine reads uploaded documents using Custom Vision API to recognise the likelihood of the content being a Land Registry Title Plan. It was trained to cater for the various formats and styles HMLR have adopted over the decades, to recognise which period the document is from, and to locate the position of the Title Number on the page. This identifier is read using OCR then used to call HMLR’s NPD to display the digitised boundary via their official API.

Clever error checks were implemented, such as screening to ensure a correctly formatted title number containing HMLR administrative area abbreviations had been identified. It was taught to improve results from poor quality photocopies or scans, and is also smart enough to recognise and ignore coversheets or erroneously included pages.

The solution works remarkably well for most title documents, and on those customer-facing websites where the service is deployed, magic-like results are instantly displayed!

AI Solution [#02].

During our build, we found a limitation in the HMLR NPD relating to Leasehold plans; their service often outlines the footprint of an entire building, not an individual flat or premise we are interested in. In this second scenario, we knew corrective intervention was necessary, but wanted to ensure legacy manual handling processes were not reverted to.

We deployed a map function that surfaces the footprint of the building as surfaced via the NPD, allowing customers to manually adjust the outline themselves where they are confident using mapping tools, or lets them digitally refer the case over to our operating staff to complete before digitally handing it back for approval prior to processing the order.

The AI part of the solution instantly sets up an internal environment in our proprietary Monitor Tool system, overlaying the plan uploaded on one screen and the digitised map for adjustment on the other, reducing time spent processing. It also checks against background data services to reduce potential for operator errors.

Working this way, we significantly reduce the likelihood of search queries or rejections downstream and to provide more accurate upfront quotes.

AI Solution [#03].

While training the algorithms to recognise the probability of a document being a Land Registry plan, an interesting by-product was discovered – the machine learning process got good at identifying when it had seen ‘something similar’ already. We harnessed this capability to solve our third scenario: improving how new build properties are found.

In most cases where a plan is not from HMLR, it will be a plot or site plan for a new build development. These require full manual intervention to locate and map, and our specialist operations teams add huge value by providing this service. There was an opportunity here to enhance the tools they use, especially when they have identified a site before.

The system uses perceptual hashing to create ‘digital fingerprints’ of images and documents. The algorithm then ‘best matches’ these fingerprints to new build developments we’ve previously processed – the machine equivalent of saying ‘I’ve seen a plan something similar to this before’ – by looking at the subtle differences between the images. So even when a property is outlined at the other end of a large site, it can direct the user to the approximate location ready for them to outline their plot.

This means the benefit of our experts having already located a new build site can be reaped by subsequent customers placing orders nearby.

Additionally, the service feeds and stores the input details of located sites back to our Property Location Service, allowing them to be found again in future when searching by developer or site name using our group Omnibox search solution.

The service is being trialled internally and via the new ordering platform at SearchFlow so it can learn from seeing more and more site plans, with the aim of becoming an accessible service to benefit all property professionals.

Outcome.

Thanks to some clever uses of AI, we have taken what could be a manually intensive multi-step analogue process and created a digital workflow that automates several tasks in locating and mapping properties.

Removing interventions has delivered slicker, more accurate ordering processes, and is reducing opportunities for queries or delays to creep in downstream in the conveyancing process. And by deploying these tools into our front-end user interfaces, customers can self-serve and accelerate work on their cases, augmented with invisible support from AI.

You can trial the features on SearchFlow’s new user interface now, and they will be rolled out to Landmark platforms in 2019.

Case Studies

No Posts available

Get in touch.

We love all things property and emerging technology. Start-up looking to consume data from our vast datasets in your app? Legal corporate or estate agency building proprietary tech aiming to leverage AI in your firm? Let’s talk.

We also champion thought leaders and innovators, so if you have an idea you’d like to explore via our blog, say hello.

hello@landmark.ai