By Elsie Rateiwa, Head: Financial Services Commercial Team at Lightstone
AI is transforming South Africa’s mortgage lending institutions, resulting in faster, more accurate, and more reliable decisions. As additional data sources are integrated, these benefits are expected to extend across the wider residential property ecosystem.
South African banks are quietly rewriting how they decide who gets a home loan. The old approach - a fixed set of rules a property either passes or fails - is giving way to something more intelligent: systems that understand risk, context and a bank’s own strategy well enough to help guide the decision itself. Banks still set the guardrails.
AI now reads the fuller picture: the property, the bank’s business strategy, its risk appetite. While these criteria are part of the rules engine, AI makes the process much quicker and is paving the way for intuitive decision-making where AI understands a client’s business, needs and aversion to risk.
The numbers to back it up
That efficiency comes down to two tools working in tandem: AiVM, Lightstone's AI Valuation Model, and EzVal. Lightstone’s AiVM is a globally accredited, AI-powered automated valuation model for South Africa’s residential property market, delivering fast, accurate, and transparent property valuations by processing extensive historical deeds data, comparable sales, and nuanced property features. EzVal is an end-to-end property valuation workflow platform primarily used by banks and valuers to optimise and streamline property valuations for the mortgage lending and bonding processes.
In practice, when a bank receives a finance request, it is funnelled to EzVal and the property address is validated within seconds. AiVM then values the property and generates a confidence score, also within seconds, and based on a bank’s rules, a decision can be made on the offer.
If the AiVM is not within required parameters, it goes to desktop and / or physical valuation. It all depends on the rules within the engine. All of this happens quickly, but once it reaches the desktop stage, that is where the human element comes in. If a physical valuation is needed, a valuer is allocated.
When Lightstone first presented the EzVal solution to the banks, valuation turnaround time was around seven to ten days. Lightstone set out to reduce this to three working days. Today, the average turnaround time for desktop valuations is just over one hour, and for physical valuations around 18 hours. With AiVM taking seconds, this has collectively made a significant difference to the banks’ turnaround time on mortgage application offers – and how many they can make.
When Lightstone first presented the EzVal solution to the banks, valuation turnaround time was around seven to ten days. Lightstone set out to reduce this to three working days. Today, the average turnaround time for desktop valuations is just over one hour, and for physical valuations around 18 hours. With AiVM taking seconds, this has collectively made a significant difference to the banks’ turnaround time on mortgage application offers – and how many they can make.>/p>
Collectively, banks are currently saving around R250 million a year on valuation fees . Since automated valuations still make up under 50% of the total volume, the potential for additional efficiencies and cost reductions remains substantial.
What comes next?
In the future, EzVal itself will decide on the need for more research or a physical inspection and book the appointment with the client and the valuer. This would usually be for high-risk or high-value properties, or where there is low confidence in a property because, for example, it has not been physically valued in over five years.
Lightstone believes AI will evolve from being a tool that simply processes inputs to becoming an intelligent advisor that understands each bank's unique lending, risk, and collections strategy. Today, AI primarily supports decision-making by analysing the information provided to it. The next generation of AI will go much further - interpreting context, identifying emerging risks and opportunities, and proactively recommending the most appropriate course of action based on a bank's policies, risk appetite, and strategic objectives.
Human oversight will remain fundamental. AI will not replace decision makers; rather, it will augment them by providing timely, evidence-based recommendations that enable faster, more consistent, and more informed decisions. Relationship managers, credit teams, and collections specialists will continue to make the final judgement, but they will do so with the benefit of an intelligent advisor capable of synthesising vast amounts of data in real time.
This shift is particularly significant in property risk, where decisions should not rely on isolated judgement or incomplete information.
The next layer of data
Climate risk data, for example, is becoming increasingly important. Lenders haven’t historically taken into account how much of their book is in flood-prone or high fire-risk areas. Bringing that data in doesn’t just protect the bank - it turns a once-off transaction into an ongoing relationship, because the bank now has a reason to stay engaged with the property, not just the sale.
Other data points that could find their way into intuitive decision-making include crime and municipal service delivery. Banks are increasingly asking about municipalities in terms of their grading, infrastructure, and return on investment, and this is consistent with banks wanting to help clients through the life cycle of their property and not just at the point of origination.
So, what does an AI future mean for buyers and sellers?
For a buyer, it means better understanding of the property value itself. Being informed about every aspect of a property – such as its exposure to flooding – at the time of purchase enables the lender to mitigate risk and empowers the buyer to make the property adjustments necessary to protect it from flooding.
The seller also is also empowered to fully understand the value of their property and how it compares to market expectations. Lightsone is also working towards helping sellers understand the renovation or development potential of their property, and how that can be factored into their sales strategy.
Looking ahead
Lending will be increasingly data-driven and tightly regulated, and banks will increasingly rely on products like AiVM and EzVal to provide speed and certainty in a world of change and disruption. The scale and depth of data underlying the use of AiVM matters because valuation models are only as strong as the information they learn from. By combining existing physical valuations, deeds records, cadastral data, municipal rolls, listing data, point-of-interest information, and imagery, Lightstone’s system builds a much richer view of property value than traditional methods alone.
For its part, EzVal connects valuation workflows with property data, analytics and risk insight, positioning valuation as an integrated component of credit strategy rather than a standalone step. What emerges is a more dynamic approach to valuation. Over the next two to three years, Lightstone’s roadmap focuses on extending EzVal into a central orchestration layer across the property transaction lifecycle – from purchase intent through valuation and credit approval to final registration.
If the past 20 years have been about modernising valuation processes, the next 20 will be about embedding valuation intelligence across the broader property ecosystem, where insight flows seamlessly through credit decisioning, risk modelling and portfolio management. Lightstone is aiming for is a single, coherent view of the property journey from intent to registration - less fragmentation, fewer manual handoffs, and faster, more transparent decisions for everyone involved.
