A local platform for tenants and rental agents uses artificial intelligence (AI) to help manage risks.
Averly, a locally created application for rental agents, property owners and tenants alike, uses artificial intelligence, technology and behavioural analysis to help identify excellent tenants faster.
“To the best of my knowledge, it’s the first time that artificial intelligence is used in this particular way,” says Zabeth Venter, CEO and co-founder of Averly.
The application uses a survey of 22 questions to extrapolate a prospective tenant’s convictions and opinions about renting; ultimately deducing how they are likely to behave once they live in the property, Venter explains.
Measuring a person’s emotions and convictions is not straight forward. The questions were carefully formulated by a variety of experts working in economics, psychology and behavioural science. They were deliberately designed to make bold statements about the rental industry to invoke an emotional response.
Underneath all this, machine learning technology is used to capture these responses, interpret it using neuroscience and ultimately convert the information into a score. The better the score, the more likely the user will make a good tenant.
How does the machine learning algorithm work?
Like humans, the machine learning algorithms must be taught what to look out for, explains Emli-Mari Nel-Fanner, head of innovation and machine learning at Averly.
A lot of care was taken to use real-life, objective data from people in the field as well as information from experts to “teach” the algorithm on how to interpret the survey’s results.
“It needed to learn what type of answer is predictive of bad or good behaviour using data and probability models,” says Nel-Fanner.
Expert knowledge was collated to create a model of an “ideal” tenant’s behaviour.
Based on these inputs, each type of behaviour is turned into a number. This is then used in conjunction with the person’s answers to calculate a final score.
The data not only includes the actual answers to the questions but also how quickly they were answered and if there was any hesitation in answering. Hesitation can indicate the person is unsure or not telling the whole truth.
Turning all these observations into one’s and zero’s makes it possible for the algorithm to interpret them and to learn how to make better predictions every time a new tenant completes the questionnaire, says Nel-Fanner.
“That’s why this algorithm is a bit different from the usual AI models. It uses a probabilistic method that includes expert knowledge, but then also uses the data to be more specific. You can basically trace back your answers to understand your score.”
Before going into production, Averly asked actual agents and tenants to complete the survey to gauge its accuracy. Because the agents and tenants already had a working relationship, the results could be compared to what was already known.
“At first there was a lot of manual intervention involved to make sure the results make sense,” says Nel-Fanner.
Soon enough the algorithm was able to interpret the results correctly without the need for any human intervention. And with time and continued training, it will become better and better at making accurate predictions.
Tenant’s scores are also not set in stone. The algorithm also takes future behaviour into account. Taking care of a property and paying rent on time can all help tenants to improve their score for the next rental application.
“This helps people to understand what accountability means and why they should be mindful about their behaviour,” says Venter.
Averly also goes beyond the algorithm by providing a platform for agents and tenants to simplify and digitise the rental application process. Properties can be rented out in as little as twenty minutes.
Tenants upload all their documents only once and use it for multiple applications.
“Ultimately it’s all about saving you time, saving you money and helping you manage risks.”