Australian car share platform Car Next Door has tapped machine learning (ML) and artificial intelligence (AI) consultancy Max Kelsen to deploy an ML system geared at reducing inappropriate car borrower behaviour.
Car Next Door's business model operates on peer-to-peer car sharing, with people able to rent out their own cars, utes and vans to others for a price.
Utilising Amazon Web Services (AWS) tech like managed ML service Amazon SageMaker, Max Kelsen built a machine learning application to assist Car Next Door’s investigations team to flag ‘bad actors’.
According to Max Kelsen principal ML engineer Michael Tremeer, the application , which was built and fine tuned over a 10 week period starting from January this year, uses previous reviews, driving behaviour, late returns and damage fees to make its predictions, which are then referred to the Car Next Door reviews team.
The team undergoes manual analysis and makes a call on the best action — either to ban the user, or to give them another chance.
Prior to the ML system, Car Next Door went through a manual background checking process, flagging less than two per cent of trips with unacceptable behaviour out of 250,000 trips a year.
According to the consultancy, this includes not refilling the petrol tank, parking issues and not returning the vehicle in a clean state.
For the new system, Max Kelsen utilised a database of 150,000 trips to create an algorithm to identify signals and patterns of behaviour that were above the threshold of acceptable use.
The consultancy claims the algorithm is “highly” accurate, with 40 per cent of users flagged by the algorithm to be banned by the investigations team, as opposed to two per cent from random review.
The algorithm also attributes a daily score to users, which can flag people with the strongest bad behaviour signals.
Additionally, the car share platform can also identify patterns of behaviour over time with the new system, as opposed to one off signals, and can allegedly identify bad actors three trips earlier than the manual process on average.
“By monitoring and actioning members to minimise inappropriate behaviour to almost zero, we can now take corrective actions to educate and nudge users who aren’t inherently bad, to have better behaviour in-line with the expectations of our community, rather than simply banning them – a better outcome for everyone,” said Andrew Bieber, head of product at Car Next Door.
“Without machine learning, we couldn’t deploy the equivalent of 1,000 people to look at every trip and every data point, every day, and have the same confidence and ability to detect and realise that certain signals together are an indication of potential malfeasance.”
Nick Therkelsen-Terry, CEO and co-founder of Max Kelsen, added the new system will play a part in helping the car share platform achieve its overall goals.
“Australian based, disruptive innovators like Car Next Door rely on building trust to drive long term customer loyalty," he said.
"Machine learning can play a key role in helping them achieve not only customer loyalty and reduce costs — it can also help them reduce the small number of inappropriate borrower behaviours. At Max Kelsen, we are excited to be able to work closely with Car Next Door to create long term value and support new business models."