With today’s advances in AI technology, what was complex and tedious can now be done quickly and efficiently. One of these examples can be found in historical lease interrogation. With a custom generative AI solution, a secure sandboxed environment can be created in which all historical real estate leases are uploaded.

This will enable any user within the firm to access a chatbot to ask it natural language questions about any aspects of the leases. This in essence becomes a dynamic lease management tool which delivers incredible instantaneous insights at the fingertips of all users.

What’s the problem

Today, most landlords have to juggle with vast amounts of historical leases. These are generally summarised in vast and complex spreadsheets or occasionally in dedicated lease management software. Existing solutions are static, rigid and inefficient which can create frustration. They are also not sufficiently flexible and only provide some insights with limited functionality. Any question outside of the prescribed KPIs requires manual reading of (often hundreds if not thousands of) leases. Significant time is spent by the insights team to generate presentations of portfolios and leases, depriving executives from time-sensitive insights which would aid decision making.

What’s the solution

Directly addressing the bugbears of real estate executives, Artefact has developed its proprietary Lease Explorer tool. This is a purpose-built and custom tool that allows for the rapid deployment of powerful Large Language Models (LLMs) within a sandboxed, proprietary environment containing historical property leases. Real estate executives are then able to interrogate their own historical leases directly with the use of a ChatGPT-like interface.

The Lease Explorer consists of five broad steps:

  • Lease ingestion: Uploading of all historical leases to a common data platform.

  • Key information extraction and preparation: Depending on the machine readibility of the leases, a variety of information extraction tools are deployed including Optical Character Recognition (OCR) and web crawling. Thereafter, information is processed and key elements are extracted into a tabulated format.

  • Q&A Chatbot: A simple Q&A chatbot interface is deployed where executives can ask questions in natural language.

  • Information retrieval: The original query and the retrieved relevant sources of information are passed to the LLM.

  • LLM response generation: The LLM generates a coherent response back to the user to answer their question, together with the retrieved sources.

Needless to say, adoption of such a tool as above would require a period of underlying change management as new business processes and ways of working would need to be adopted.

What kind of natural language questions could be answered

With the implementation of Lease Explorer, we posit some natural language prompts that a real estate executive might be able to ask the tool’s ChatBot. Of course, the extent to which the tool can accurately return a response to the below non-exhaustive question list is conditional upon the quality, structure and availability of input data, which may require access to information beyond that simply contained in a lease document.

  • How many leases of first time renters with our organisation do I have up for renewal in the next six months? Where are the units located and what is the average rent of these particular units?

  • What is the value of the leases coming up for renewal in the next year?

  • Can you graph my upcoming lease expirations broken down by month / year?

  • Of my current leases, can you tell me what percentage (by value) is offices? Mixed use?

  • How many dentist are renting from me? What is the average square footage of their leases? Can you plot this on a chart for me?

  • How many commercial tenants have exercised renewal options in the past two years?

  • What is the average lease duration for (commercial) tenants in a particular area (Zone 1)?

  • What unusual escalation clauses have become apparent in our portfolio in the last three years?

  • Can you show me a comparison of lease terms between retail properties in shopping malls versus standalone commercial locations?

  • What is the average security deposit amount requested in our leases over the last two years?

  • Can you provide a breakdown of our portfolio rental income by geographic area (city, county, etc)?

  • How many commercial leases include options for subleasing, and what percentage of tenants have exercisesd this option?

  • What percentage of our commercial leases include rent abatement clauses, and what is the average duration of rent abatement periods?

  • What proportion of the portfolio have leases with a break clause due in the next six months?

What benefits does this generate

The adoption of an AI tool like Enterprise ChatGPT will accrue a range of benefits to real estate professionals. On the first hand, the tool is likely to provide a sea change with regard to the speed at which insights are able to be obtained, thereby greatly empowering decision-makers with rapid access to critical information. Furthermore, the scalability and adaptability of such a tool to ingest an ever-growing volume of lease data would be invaluable to assist the future-proofing of a real estate business to the increasing demands of the industry.

Adoption of such a tool is likely to yield a step change internally within a real estate organisation too. The adoption of an AI tool will minimise any reliance on previously manual processes with respect to interrogating leases. Rather than relying on the incorruptibility of a spreadsheet and continuing service of key admin personnel, an AI tool would be able to take on board the questions in natural language of executives directly and provide accurate responses in a timely fashion. Essentially, overreliance on key dependencies for data search and retrieval is reduced and there is scope for executives to reduce resources required in the insight generation process. Further, in order to get the best value from such an AI tool, adoption is likely to require a business to undergo a transformative data hygiene process, thereby elevating the internal thresholds of data management as the insights generated can only ever be as good as the information collated.

GenAI additionally has certain benefits for such a tool over traditional AI / ML solutions. Traditionally, vast amounts of labelled data were required to train and evaluate a model for a specific task. This is no longer required thanks to generative AI, speeding up the development of such solutions. Further, generative AI allows solutions to evolve and change without massive additional re-work and development – solutions can be more flexible and less rigid.

How is accuracy established

Implementing a novel AI tool for the first time and enabling it to generate powerful executive insights with a mere prompt can no doubt be daunting, especially given GenAI’s well-cited drawback of “hallucinations”. The following steps when taken in concert should provide sufficient guardrailing to ensure the best value can be extracted from the tool.

  • Iterating definition of an AI with the actual lease documents such that the model has enough industry and company specific context to provide answers.

  • Validating processes continuously to refine model performance and enhance accuracy with a human in the loop (i.e. human validating the answers of the model).

  • Adopting of a ‘test and learn’ approach with feedback from relevant reviewers then being fed back into the solution.

  • Establishing detailed stepy-by-step instructions and guardrails to ensure that the intermediate and final outputs are accurate, specific to the company and without hallucinations.

  • Performing regular quality assurance checks to verify accuracy of information.

  • Employing a diverse set of test questions and answers to validate the model in order to identify the strengths and weaknesses of the system, facilitating the improvement of prompts and the need for additional building blocks and guardrails.

What limitations exist

For this solution to work as desired, there are a number of considerations to take into account. Firstly, the model would most likely be limited to UK leases, due to the property sector’s peculiarities around location and its legislations. The Lease Explorer would have no predictive capabilities, as only historical performance from existing documents would be interrogated.

Additionally, like any GenAI based tool, the Lease Explorer would struggle with unclear and vague user prompts, which might not yield the desired results. As such, it is always recommended to formulate questions to the chat interface properly, similarly akin to how one might ask a question to an analyst within their real estate team.

It is important to note that the reading and processing of PDF documents into machine-readable format requires OCR (optical character recognition) as a necessary step in the ingestion process. Documents that are not machine readable or that have a complex page format and structure might not be ingested correctly and downstream this is likely to affect tasks such as extracting information and providing reliable answers.

Furthermore, to answer analytical quantitative questions, further processes are required:

  • Extraction of key information from the leases, putting these extracted attributes into a structured and tabulated form in a database. This process can also be automated with generative AI, providing benefits over manual extraction in terms of development and deployment time. The data extracted can also be used as the source data for traditional business dashboards and reports, converting this task into a valuable business use case. Note that this step is only needed if the data to interrogate is not already available in a database in structured form.

  • Use generative AI to query the structured data extracted from the documents, converting the natural language questions asked by the users into SQL code that is then executed on the tabulated data. The unstructured documents can also be used in order to leverage both sources of data, providing a system to generate rich quantitative insights in an instant.

Costs and requirements

The deployment and acculturation of Artefact’s Lease Explorer tool would reasonably cost from c.£45 k and would require at least four weeks of working cooperatively depending on chosen functionalities. This would of course be under the assumption that uploaded leases are all machine readable, substantially similar and in English. Following deployment of Lease Explorer, an ongoing subscription would be required to any GenAI LLM that is used to build the custom tool.

Why Artefact?

Artefact is a leading global consulting company dedicated to accelerating the adoption of data and AI to positively impact people and organisations. We specialise in data transformation and data marketing to drive tangible business results across the entire enterprise value chain. Artefact offers the most comprehensive set of data-driven solutions, built on deep data science and cutting-edge AI technologies, delivering AI projects at scale across the property sector in the UK.

We are trusted partners to property businesses across residential, commercial, industrial and specialist asset classes. Our partners include FTSE 350 listed companies and similar-sized private organisations. With over 20 years of experience in real estate, our dedicated property team includes experts and chartered professionals in property valuations, urban planning, development, and financing.

Our previous work ranges from developing data-led dynamic strategies with our clients – informing them where to play and how to win in their chosen markets – to major operational changes, such as establishing new business arms and propositions. We have worked in every stage of the property lifecycle, from land acquisition to ongoing maintenance, and have worked with clients to scientifically improve these processes.