Bus times driven by AI

Thursday 4th January 2024

My very first work experience as a fresh faced management trainee almost fifty years ago was a placement in the schedules office at Belle Isle bus depot in Wakefield where four members of staff, including the long experienced Harry and Ted, compiled vehicle workings and staff duties for West Riding and Yorkshire Woollen’s six depots across West Yorkshire as well as one in Selby, North Yorkshire.

I remember spending hours learning the skill of drawing lines on huge graph sheets, obviously in pencil so mistakes could be easily corrected, showing how timetables are carved up into vehicle workings and from those then trying my hand at compiling the most efficient duties which drivers and conductors would work having been acceptable by their union representatives as meeting agreed practices.

It was often said profits were made or lost in a bus company’s schedules office such were the implications of tracing out those lines in the most realistic but cost effective way fo the business.

Over the last fifty years that tried and tested process of resource utilisation has seen many developments, not least in recent decades the use of computer software removing the laborious and time consuming ‘first cut’ of bus workings and duties enabling skilled schedule staff to spend their limited time fine tuning the resultant product into what long experience tells them will make for a workable set of duties by drivers.

More recently some Bus Groups saw further potential to streamline the process by centralising schedules production for all their operations across the entire country, but this risks missing vital local knowledge of what is and isn’t achievable, whether it be specific peak hour traffic conditions, school requirements, meal break or spreadover duty arrangements for staff or a hundred and one other idiosyncrasies often only known to those close to the action.

Now, the latest development being championed by First Bus, and it seems also used by Stagecoach, is to embrace the world of ‘Artificial Intelligence’ to analyse the wealth of data now captured every hour about on the road punctuality performance to compile timetables, schedules and duties which reflect varying traffic conditions.

It was reported in the trade press just before Christmas First Bus is spending £4.5 million with a company called Prospective – “we enable fleet operators, infrastructure managers and asset owners to run their operations in sync with the city around them” or as First explains “the new system will mean that local teams can be more agile with frequent, subtle changes to ensure timetables remain accurate throughout the year”.

It’s long been a practice in the industry to allow additional running time across peak hours to counteract delays caused by traffic congestion, not least since the Traffic Commissioners began taking a keen interest with roadside monitoring of how services were performing and imposing significant penalties for failings.

It’s often the case an extra bus, or more, are required at peak times to deliver the same frequency timetable as applies during the daytime or conversely, the frequency may be reduced to save deploying costly resources. London is a classic example of this with significant differentiations in running times to be found in the Capital’s bus schedules and extra vehicle/s slotted in or, more likely over the last couple of years, judicious frequency reductions on many routes to save costs.

However, in London most bus routes don’t have an officially published timetable which passengers can consult and instead work on the provision of a ‘turn-up-and-go’ headway across the day. Passengers are generally oblivious to any tweaks in journey times and/or the inclusion of more buses in the schedule to reflect traffic conditions.

This is not how things are working out as First Bus deploys Prospective’s AI technology. The result of this “cutting edge technology” might be “more punctual, reliable and comfortable bus services” (as per the above quote) but the reality for passengers is a more complicated, difficult to remember and even off-putting timetable with none of the simplicity or attractiveness of a regular frequency.

Take for example route 4, the long established route between Acomb and York city centre once famous for hosting that ‘cutting edge technology’ the ‘ftr’ bus back in 2006.

Instead of a timetable providing a 15 minute frequency with easy to remember departures, whether they be 00, 15, 30 and 45 or 10, 25, 40, 55 etc etc, passengers are now faced with a multiplicity of different timings not only at peak times, but throughout the day.

The differences may only be a matter of two to three minutes but the result is a confusing presentation of varied times along the route. Here’s an extract from the online timetable on the First Bus website (which unhelpfully is currently only displaying times up to this Saturday 6th January), and just to add to the complicated format, displays 12 digit “stop codes” for each timing point with no explanation of what they’re there for.

As you can see, instead of an even 15 minute departure, the time interval in minutes between the ten journeys shown above from Acomb is 16-16-18-18-15-15-15-17-18 with the length of journey into York’s Clifford Street varying as 38-38-35-35-35-35-34-34-34-34 minutes.

The next extract below continues this trend as the afternoon beckons with gaps between journeys inexplicably varying inconsistently as 15-16-15-18-15-15-15-15-17 minutes with journey times reducing consistently from 34 to 33 to 32 minutes.

As the afternoon wears on the trends continue with more inconsistent headway intervals and the journey time reducing to 31 minutes.

Even more odd, in the evening peak period, the journey time is cut again to just 30 minutes for departures at 17:17 and 17:42 but with the longest headway gap between them – at 25 minutes, before improving to every 20 minutes after 18:00, and a journey time down to 28 minutes and, finally, 25 minutes after 19:00.

To finish things off here’s the rest of the evening with a half hour’s gap to 20:13 them inexplicably a 34 minute gap to 20:47 before back to 30 then 31, then 30 and finally 31 minutes between journeys while journey times reduce to 21 minutes by the last departure.

I’ll not bore you with the inbound direction but it follows a similar inconsistent pattern with irregular gaps in the headway and varying journey times. Looking at the overall timetable, if I sketched it out on old style vehicle graph paper, there seems to be no reason why a regular 15 minute frequency can’t apply throughout the day moving to a 20 minute interval across the evening peak after 17:00 and through the evening itself.

But, I’m guessing the AI process looking at thousands of actual journey times on the road across a historic period of time reckons its timetable ensures “our buses are as efficient and punctual as possible”, to quote Simon’s endorsement on the Prospective website.

That’s as maybe for the statisticians but for passengers it’s a complete turn off. And I can’t believe Simon reckons by First Bus using this new AI process “we’re not only looking to drive improvements for our existing customers, but to make bus an affordable, more reliable and attractive part of everyone’s everyday transport mix.”

I’m sorry to be a party pooper but this is cloud cuckoo land DRT Mark 2. I just can’t see new passengers flocking to travel on route 4 with such an unattractive sales proposition conveyed by the above timetable, nor all the other routes now being scheduled in this way – I just took route 4 as a random example from the First Bus website.

Far better would be to promote the simplicity and regularity of a standardised ‘every 15 minutes frequency’ coupled with relentless promotion of the availability of real time information on websites such as bustimes.org and even the First Bus app so passengers can see where the bus actually is compared to the scheduled time, which is likely to be within 2-3 minutes variation anyway, if the AI data is worth anything. A variation to schedule like that is something most passengers would find acceptable – and frankly, more acceptable than the abomination of a timetable this AI process produces.

Conditions on the road vary by two to three minutes from day to day depending on a whole host of factors (eg numbers of passengers travelling, queries they might raise, the need to deploy a wheelchair ramp, road and weather conditions etc etc etc) and to try and schedule what must be an average of all these minor variations is just a complete waste of time and counter productive to the objective of making bus a “more reliable and attractive part of everyone’s everyday transport mix.”

Simon enthuses “the results we’ve achieved in West Yorkshire and the West of England have been incredible in such a short space of time, but we’re only scratching the surface of its potential. We’re excited to roll out this software and see the benefits it brings to our customers.”

Harry and Ted will be turning in their graves.

You can read the full article which appeared in a recent edition of Coach & Bus Week magazine here, or reproduced below this blog.

Roger French

Blogging timetable: 06:00 TThS

Comments are welcome but please keep them relevant to the blog topic, avoid personal insults and add your name (or an identifier). Thank you.