Swiftly has been in the fixed route public transit game since, well, before Swiftly. One of our co-founders and my brilliant colleague, Mike Smith, spent decades in the transit software world fine tuning prediction models for hundreds of agencies around the world. He started Swiftly with a goal to deliver the absolute best predictive information for passengers — and we’ve done just that.

Swiftly’s fixed route transit predictions are the best in the industry, consistently 15 to 50% more accurate than other systems. Our prediction algorithms leverage a variety of historical data as well as truly real-time vehicle location information. We iterate continuously on our prediction engine to reduce error and ensure passengers catch their ride.  

And yet, there’s much more to great transit predictions than just accuracy.

You can hear best practices and new strategies for connecting with passengers in our recorded webinar, "Connect with passengers: Building passenger trust with 'real' real-time information," featuring speakers from Transit, IBI, and Swiftly.

Beyond prediction accuracy

Generating truly great real-time passenger information requires quite a bit more than predicting the predictable. It necessitates understanding the unpredictable, too.

We all know that transit doesn’t run in a vacuum. Life happens all around transit operations - operators don’t log into their onboard mobile devices, water main lines break and force detours, delivery trucks double park and slow corridors, and staff need unanticipated sick leave.

Similarly, riders don’t always search for information on an agency’s website or dedicated mobile app. They may have their own preferred third-party trip planning app, or they might rely on an SMS service or digital signage at stops.

So, while we pride ourselves on our extremely accurate real-time passenger predictions, we know that communicating service changes and getting information into people’s hands is just as important. Beyond the measurable accuracy of predictions, here are other key elements of great passenger info:

Data standardization

• Swiftly’s predictions flow out via GTFS-RT APIs, making it easy to share the same predictions wherever passengers are getting their information - on their phones, wayside signage, station kiosks, websites, or other apps.

High coverage

• Systems that rely on their own hardware for vehicle location data can only predict runs when operators log into the vehicle’s onboard mobile device. In some cities, we’ve seen operator log in rates hover at 50%! Swiftly employs algorithms to automatically assign vehicles to the right piece of work, so predictions can flow for up to 100% of service.

Fault tolerance

• Swiftly reads in location data from many types of devices on board, including from an agency's dispatch system (CAD/AVL) or independent routers, to create high fidelity vehicle location data. This means that if one feed goes down, predictions keep rolling based on other device integrations.

Dynamic operations

• Especially when operations diverge from the schedule, passengers need a way to depend on transit. Swiftly’s algorithms can detect when service isn’t running to avoid sharing predictions for pesky “ghost buses.” For even better passenger information and historical data, agencies can generate changes to service such as cancelations, in the Swiftly dashboard or via Swiftly’s API.

Dynamic operations, you say?

Predictions are ETAs for when a vehicle will arrive at your stop. But, what about all those vehicles that won’t? Service that isn’t running is just as important as the service that is.

Swiftly’s data engine can include cancelations, whether trips are canceled in the Swiftly dashboard or even in another system and fed in via GTFS-RT or Swiftly APIs.

And it’s a good thing it does: Cancelations account for the majority of Service Adjustments in Swiftly, and “reduced service” or “no service” make up more than 50% of Service Alerts reasons that flow through the system.

Knowing when service is explicitly canceled empowers riders to better plan their trips and builds trust with agencies; leaving riders to guess does the opposite.

Toward better Real-Time Passenger Information

Scrutinizing prediction accuracy is a worthy area of focus for any transit authority. As studies have shown, better ETAs reduce perceived wait time and can increase ridership. Yet prediction accuracy is really only a part of the story when it comes to communicating effectively with transit riders.

Real time passenger information must take into account the unpredictable: an AVL feed failing, vehicles that aren’t logged into an onboard system such as a CAD/AVL, and cancelations across different software systems. If we only focus on prediction accuracy, we can get caught up in looking at a fraction of the fleet instead of the holistic picture of service that passengers really care about.