Reducing Manual Effort Through AI-Driven Auto-Coding

AI Auto-Coding and the Future of Franchised Bus Operations

Public transport networks are the backbone of daily mobility, but delivering reliable service is no easy task. Behind the scenes, operators face constant challenges, unpredictable traffic, accidents, sudden road closures, or issues within the network itself, such as unavailable drivers or vehicle breakdowns. Each of these events can quickly disrupt operations and requires swift action to keep services moving and passengers satisfied.

From the passenger’s perspective, reliability is what defines service quality. If buses are scheduled every five minutes, passengers expect to arrive at a stop and wait no longer than that interval, typically just two to three minutes. When frequencies slip and arrivals become less predictable, passengers notice immediately, and this directly shapes their perception of the service being delivered.

Franchising contract models are designed to monitor the quality of service delivered to passengers by measuring each trip against the planned schedule. Any disruption to the network directly affects this measurement, reducing the perceived quality of the franchised service. To protect performance, bus operators must take swift dispatching actions, adjusting services in real time to minimise the impact of disruptions and maintain delivery standards.

How CFMS helps manage disruptions

The LTA’s current CFMS solution provides several tools for managing these day of operation disruptions, enabling the service controller to divert the bus service to avoid the disruption. These actions are aimed at minimising the impact to the waiting passengers. However, this has an impact on service delivery and as the actual service delivered is monitored for every trip, this flows straight through to the performance management system, which measures the quality of the service delivered and the potential that there is an impact on the KPIs and therefore the revenue of the bus operator.

A contracting franchise solution needs to be fair, recognising that there are issues which are outside the control of the operator and other issues where the issue is under the control of the operator .

CFMS, the current bus fleet management system used in Singapore, recognises that there are a number of situations that can impact service delivery. It is programmed to understand that some of these scenarios are attributable to the operators and therefore no compensation should be allocated to them as the situation was deemed to be their responsibility and the networks core KPIs are impacted. Other situations, for example a road closure due to an accident or public event, is not deemed to be the operator responsibility and so the CFMS recognises that they should not be penalised, rather the situation, which is unavoidable to the operator, taken into account and compensation provided.

This coding process enables the operator to justify the actions taken during the day and to receive appropriate recognition and for this to be included in the contractual compensation. If there is a road closure and the bus has to divert, then the mileage is still being operated, fuel consumed and drivers need to be paid. However, if the trip or part of the trip is not run, then the operator should not be compensated, as the service was not provided.

In CFMS this process of coding the missing trips and trip segments is performed as an offline action. Currently when the dispatcher action is setup the performance management system runs overnight, processes all of the operated trips against the planned service and detects the non-delivery. This requires the missing trips to be coded with the appropriate reason code, either deductible or non-deductible. This identifies the cause as being something external to the operator (there is no deduction made) or it is the operators problem and therefore there is an impact on the operation and a deduction is made.

This information is then processed in the output reports and considered within the operational statistics for the service delivery and the quality of the service provided.

This processing currently requires a team of coding clerks to process the records. They use information recorded by the service controller and other information sources to prepare and code the trips. This process takes time and requires significant information to justify, particularly non deductible codes. There is a whole justification and authorisation process to follow.

AI-Powered Auto-Coding: Driving the Next Generation of BFMS

The LTA’s new BFMS solution, building on the current CFMS solution, introduces the ability to enter the reason code as part of the dispatching action at the time of entry. Setup a diversion in the BFMS workstation and the dispatcher is prompted to provide the appropriate code to define the reason for the diversion. This is then propagated to the performance management system where it is automatically entered into the corresponding trip or trips.

The system also incorporates an interface to the bus operator dispatch system, which includes the ability to adjust the start time of a trip as well as to cancel / reinstate a trip or set of trips as well as make path dispatch actions to start a trip from a mid-way point along the route. In BFMS this interface is extended to enable the operators, if desired, to encode the reason codes within this interface, removing the need for the coding clerks to follow up for any missing data and manually code.

This simplifies the coding clerks activities, particularly where there is a referenced incident number which correlates to an incident recorded in the LTA incident management system, removing the need, in many cases, to provide additional information and evidence. There will still be some missing trip information that is not able to be automatically coded.

The new BFMS system will go further, integrating real time information on stop arrival and departure times into the performance management data. These real time observations are used in CFMS and BFMS to track the vehicles along the route and update predicted arrival times at stops and interchanges. This information is updated on arrival and departure from stops as well as location updates between stops.

The new system will use this additional real time information to provide two levels of analysis. The first uses real-time collected information, based on reported stop arrival and departure times. The second uses detailed statistical files from vehicles, which include precise first door open and last door close timestamps.

The performance management system makes an initial analysis of the performance using the real time information, providing initial views into the operated trips as well as non-operated trips or trip segments.

This initial calculation is updated when the detailed statistical files have been processed when the additional data points are fed into the performance system. All operating buses monitored by the control centre are processed. An initial performance assessment is then made using door-event data wherever it is available and supplemented with real-time arrival/departure information when it is not. The performance management system records these real-time trips as operated, but without the detailed door-based data, which is used for precise bus stop arrival and departure information. When door-event information becomes available, it is used to update both the trip records and the associated performance reports.

This significantly reduces the coding needed where a bus operated but for some reason the collected statistical information was not uploaded on return to the depot. This can occur due to for example a mechanical fault in the bus requiring the battery to be isolated prior to the collected statistical information having been uploaded. It will also ensure that, every morning, all the visible operated trips are already coded, even if the vehicle is still operating or the files have not yet been processed.

Conclusion

In summary, this system seamlessly links franchise contract management with the controller’s day-to-day workflow, embedding it into the franchise performance management process. The result is a more efficient, accurate, and responsive approach to monitoring service quality.

Interested in learning more about what auto-coding is already delivering and the additional opportunities it could unlock?