Missing the Express

Every morning I take the F into Manhattan to get to work near 42nd Street. The F is a local, so it stops everywhere. A few stops in, it pulls into Broadway-Lafayette, where the B and D stop on the platform right across from me. Both are expresses. If a D is pulling in as I arrive, I can step across, hop on, and save several minutes.

The problem is that I never know whether that D is coming. There's no cell signal in the tunnel, so I can't check an app, and nothing on the train tells me what's happening on the other line. So I guess. Some mornings I jump off and watch three F trains go by before a B or D shows up. Other mornings I stay on, and a D is sitting across the platform as my doors close.

Either way I lose time, and the decision is a coin flip, not a choice.

The information exists. It just never reaches the place where the decision gets made.

A transfer decision is only useful at one moment, the few seconds before the doors open, and in one place, the train itself. A phone app fails on its own terms, because the moment of decision happens exactly where phones don't work.

So deliver it on the hardware riders are already looking at: the digital screens inside newer cars and the countdown clocks on the platform. That hardware is connected to the MTA's train control and arrival systems, so it doesn't depend on the rider having signal at all.

How big is it? The subway carried about 3.7 million riders on an average weekday in 2024, per the MTA's own ridership tables, down from about 5.5 million in 2019. The MTA doesn't measure how many of those trips include a train-to-train transfer, since riders only tap on entry, so that share is an estimate. What matters for this feature is the subset of stations where a rider can switch quickly between a local and an express, sometimes across one platform. 59th St-Columbus Circle is one: the seventh busiest station complex in 2024, at about 16.9 million entries.

The time saved per rider is small. Multiplied across many transfers, it adds up to something worth validating against the MTA's ridership and travel-survey data before committing to the project.

The feature is a band on a screen that already exists.

Connecting Trains is a panel on the in-car screen and platform displays. For each upcoming station, it shows the next arrivals of any lines that connect there. As the F approaches Broadway-Lafayette, the screen above the doors reads: Next stop: Broadway Lafayette St. Connections: D (express) in 2 min, same platform; B (express) in 3 min. That's the sign at the top of this page.

Each connecting line gets its colored bullet, a one-word express or local tag, a countdown in minutes, and a small mark for same platform versus a walking transfer. A rider should be able to read it and act in under three seconds, about the time they have before deciding whether to stand up. A rider who wants the express now has a real choice instead of a guess.

Three details make it trustworthy rather than decorative.

The rider never needs a connection. Each train's onboard display gets arrival predictions for nearby stations through the same data backbone that already powers the platform countdown clocks. Between stations, the train knows its next stop and can ask for the predicted arrivals of the connecting lines there. For cars without screens, the same information shows on the platform displays at the shared station, and can be echoed in optional onboard audio.

Before anything gets built, I'd want to know what number would tell me it works.

These are proposed metrics, not measured results.

On the experience side, I'd track rider-reported confidence in transfer decisions through station exit surveys, and watch crowding and dwell time to make sure the feature doesn't create unsafe platform rushes.

A countdown that's wrong once is worse than no countdown.

Trust. If the countdown is wrong even occasionally, riders will jump off for a train that doesn't come and stop believing the feature entirely. It has to be conservative: show a connection only when confidence is high, and degrade to "check platform" when it isn't.

Hardware coverage. Older cars don't have screens. That argues for leaning on platform displays during a phased rollout and adding in-car screens as the data improves.

Safety. A two-minute connection could encourage riders to dash across platforms. The timing logic and the copy should never imply a connection is catchable if it realistically isn't.

Prediction reliability. The feature is only as good as the arrival predictions under it, so an early phase should cover the highest-confidence shared stations, not the whole system.

Which is why it wouldn't start on the trains at all.

Start on the platforms. A pilot at a handful of high-traffic stations where riders choose between a local and an express, beginning with true cross-platform transfers like 59th St-Columbus Circle, on platform displays first, because the hardware and data are most mature there. Broadway-Lafayette is in the pilot too, since the F-to-D-or-B decision it enables is exactly the behavior this targets.

Then the trains. After validating accuracy and measuring transfer conversion and catch rate, expand to in-car screens on the equipped fleet, then widen station coverage.

The gate. Each expansion waits for that station's arrival predictions to clear a trust threshold.

That’s the whole plan.

This feature doesn't require new trains, new track, or new schedules. It takes information the MTA already has and puts it in front of riders at the one moment it can change a decision.

For me, that means catching the express at Broadway-Lafayette on purpose instead of by accident. Across millions of daily riders, it means turning a coin flip into a choice.

Read the full case study on Medium →