The Indian Railways has successfully concluded the massive trial of an Artificial Intelligence program it built to fix a perennial, exasperating pain point — waiting lists.
Beginning a new era of how the Railways apportions its gigantic inventory of berths, the AI-driven module, ‘Ideal Train Profile’, was able to shrink the size of the waiting list by “five to six per cent”.
At the end of the trial, there were simply more confirmed tickets for travellers at the time of booking itself.
Developed by the Railways’ in-house software arm, Centre for Railway Information System (CRIS), the ‘Ideal Train Profile’ was fed with information on about 200 long-distance trains, including the Rajdhanis.
The AI module processed patterns that included how passengers booked tickets, which origin-destination pairs were a hit or a flop, and at which time of the year. It also looked at which seats remained vacant for how much portion of the journey period.
With data, to begin with, from the past three years stored in the AI module, engineers involved in the trial said the combinations of “training data” that ‘Ideal Train Profile’ could process were “virtually endless”.
They said the module learnt variable data sets or possible ticket combinations by dividing a single journey into the number of halts and also processing passenger behaviour during these halts.
“If there are 60 halts in a long-distance train, the AI has learned about 1,800 possible ticket combinations. If there are 10 halts, then there are typically about 45 ticket combinations and so on,” a senior Railway Board official who did not want to be named told The Indian Express.
According to officials involved in the testing, ‘Ideal Train Profile’ was made live at the start of the Advanced Reservation Period or a 120-day period before the departure of trains, in this case January end. The trial covered passenger reservation systems in seven zonal railways.
The officials said the Railways was keen to test out the kinks “well enough” before the May-June holiday period when the demand for confirmed tickets is the highest, as is the disappointment of passengers caught in endless waiting lists.
Rail Bhawan mandarins have generally admitted that a large number of users turn away from the Railways simply because the national transporter is unable to provide upfront what they come looking for: a confirmed ticket.
Sunil Kumar Garg, Additional Member (Commercial) of the Railway Board, conveyed this in a letter to General Managers of zonal railways at the start of the AI trial.
“The increased competition from airlines in the upper-class long-distance and buses in the short distance travel has been a cause of concern,” Garg wrote. “Further, introducing additional passenger trains to meet the growth in certain congested sections has been a challenge.”
“To improve the occupancy and to enhance the revenue of the existing reserved trains, the need for a robust Passenger Profile Management based seat-quota redistribution was being felt for quite some time,” the letter says.
But redefining the old rules — of distributing seats, freeing up seat quotas or deciding the number of confirmed tickets according to origin-destination combinations — was so far simply impossible, more so in real-time.
Now, officials said, with ‘Ideal Train Profile’ successfully completing its trial, there is hope.
Another Rail Bhawan official, who did not wish to be named, said the Indian Railways works with 1 billion ticket combinations for all its reserved trains.
“So said the training data this AI is going to absorb is phenomenal,” he said, adding that by informal estimates, the module can generate an additional revenue of Rs 1 crore per train a year.
“And as it is with AI, the more it learns the more accurate it gets.”