Nowadays, there are two main car-sharing business models:
  • The based station model requires parking vehicles at pick-up stations defined by the service providers or the local administration. In this case, the biggest constraint is starting and finishing the travel at these stations. Therefore, this model requires optimised distribution of the number of vehicles at each station to meet the demand.
  • The free-floating model allows travellers to start and finish at any city location or operational area and park the vehicle anywhere within this zone. In this case, there are multiple possible optimisation strategies, posing significant challenges to the fleet administrator or the service provider. This model also offers enhanced possibilities for developing innovative solutions for optimisation at the fleet and travel levels.
Focusing on the second free-floating model, we identify two main problems:
  1. How to meet the estimated demand? This challenge includes the size of the fleet and a geographical area demand prediction to optimise the fleet’s efficiency.
  2. How can interventions at each service level increase the chances that each trip terminates in high-demand zones? Considering that most of the operational costs in this business model come from the reallocation of the vehicles, such a solution reduces idle time and increases the profit per vehicle. 

Both challenges require an accurate predictive dynamic model of pricing and reward that adapts to both people’s behaviour patterns and the provider’s objectives.

Machine learning algorithms can be reasonably accurate in their predictions these days, using proper training of the model. Therefore, many companies have already artificial intelligence systems tuned according to their service, although some use real-time demand more than a prediction.

The other crucial free-floating fleet challenge and the innovation trend in the car-sharing service are to adaptively align the origin and destination locations of the users based on their behaviour. The aim is to match the route at a specific time with the next high-demand area. In this context, we can act in several ways:

  • Reward a change in the drop-off location. This first option is to offer a discount or a reward to incentivise the users to leave the vehicle at a different place in exchange for a prize. This option enables operators to set rewards for ending the travel in an area more suitable for the following user (high demand zone) or closer to it. As previously mentioned, the operator needs to know the more profitable place for the next service in advance.
  • Reward a change in the pick-up location. This second option offers a discount or reward to change the pick-up location and allows operators to extend specific demanding zones. In this case, a user agrees to get the vehicle from a farther pick-up point. Once again, this requires knowing in advance the location of the car.
  • Paid relocation. This third option offers a reward or a discount if a user agrees to reallocate the vehicle. In this case, the operator manages the fleet with a reallocation that helps to meet the nearby demand. This option requires considering the reallocation time and the demand implications for not completing it in time, which adds an extra dimension to the complexity of the model.
  • Vehicle delivery. The last option requires the user to pay an extra fee to the operator to get the vehicle to a required location by hiring specialised staff or paying other users (the option previously commented as paid relocation). The fee finally charged to the final user represents an expensive extra service and an extra challenge to this business model in both cases.

In addition, we have to consider the reward to change the route of the user and its implementation. The reward that can change users’ minds is difficult to foresee since it depends on the country, the city, and the economic status of the operational area. For the beginning, one can approximate this discount related to the operational cost of manually relocating a vehicle to a more appropriate place and then refine the strategy per zone. The rise of blockchain technology has enabled operators to integrate automatic rewards, which naturally fit with the payments and escrows needed in the business. Current blockchain implementations facilitate the integration of discounts or rewards through smart contracts and generic APIs, enabling the connection with the AI model.

The approach can be challenging to implement even after defining all parameters. Therefore, our recommendation is to start by focusing on a few of these dimensions and improving the model with new ones when it becomes accurate. The most critical aspect in training the machine learning model is to get a representative dataset of the users, which requires experience in the city. Alternatively, surveys in the town can collect feedback and translate it into a starting dataset for the model. Moreover, these models need training for each city, considering not only historical data of the service but also external parameters like weather or events that can have a high impact on the accuracy of the approach. Getting an accurate predictive module is sometimes a matter of trial and error. Still, the benefits for the car-sharing business model can be significant since the overall efficiency might impact the operational cost reduction and lead to higher profits.

This blog post was written by Agilia Center team in April 2021.

< Thanks for reading. We are curious to hear from you. Get in touch with us and let us know what you think. >