AI in Public Transit: Lessons from SEPTA’s CIO

Most transit agencies didn’t “lose riders” during the pandemic — they lost predictability.

When offices went hybrid, school calendars got messy, and service workers shifted schedules, the old playbook (fixed timetables, historic averages, seasonal patterns) stopped working. That’s why the most useful takeaway from SmartCitiesWorld’s conversation with Emily Yates — SEPTA’s chief innovation officer, formerly Philadelphia’s Smart City Director — isn’t a shiny new tool. It’s a mindset: systems thinking, shared data, and pragmatic innovation are the only way public transport can recover trust and run efficiently.

This post sits in our “Mākslīgais intelekts publiskajā sektorā un viedajās pilsētās” series for a reason. Public transport is one of the clearest places where AI in the public sector can produce measurable outcomes: better reliability, safer operations, smarter maintenance, and clearer communication with residents. The trick is doing it without creating a fragile, vendor-dependent black box.

What SEPTA’s story tells us about AI in the public sector

Answer first: SEPTA’s experience highlights that AI delivers value only when agencies treat mobility as an end-to-end system — not a set of separate projects.

Emily Yates’ career path matters here. Moving from a city smart city role into a major regional transit authority is a real-world example of how smart city technologies can (and should) carry over into core public services. Transit isn’t an “innovation lab.” It’s a daily, high-stakes operation.

That framing changes how you approach AI:

• You don’t start with “Where can we use AI?”
• You start with “Where are we bleeding time, money, and public confidence?”

For many agencies post-2020, the answers are consistent:

1. Demand is harder to forecast.
2. Staffing and maintenance constraints show up as reliability issues.
3. Riders expect real-time info that’s accurate — and they punish you when it isn’t.

AI can help, but only if your organization can connect the dots between operations, customer experience, asset management, and communications.

The pandemic didn’t create the problems — it exposed them

A practical stance: the pandemic didn’t “break transit.” It made weak assumptions visible.

Agencies that relied on:

• manual reporting,
• siloed departments,
• inconsistent data definitions,
• and slow procurement cycles

found themselves unable to adjust service fast enough.

In smart city work, there’s a familiar pattern: the city has data, but it’s fragmented. Transit agencies face the same issue — vehicle telemetry over here, fare data over there, maintenance logs somewhere else, and customer complaints living in yet another system.

If you want data-driven decision-making in government, the first milestone isn’t a model. It’s an agreement on reality.

Systems thinking: the unglamorous secret behind “smart mobility”

Answer first: Systems thinking is how you prevent AI projects from becoming isolated dashboards no one uses.

Emily Yates and the podcast discussion emphasize a key point for smart cities: transit performance is shaped by many agencies, not just the transit operator.

A late bus isn’t always a “bus problem.” It might be:

• a traffic signal timing issue,
• a roadworks coordination failure,
• a curbside management problem (double parking, delivery conflicts),
• a safety incident requiring police or emergency response,
• a communications gap that leaves riders uninformed.

This matters because AI projects fail when the scope is wrong. If you’re trying to improve on-time performance, but you only optimize the schedule, you’re ignoring the environment the schedule runs through.

What “systems thinking” looks like in an AI-enabled transit program

Here’s what works in practice — and what I’ve seen consistently separate progress from stalled pilots:

• Shared KPIs across agencies: If the city optimizes traffic flow for cars while transit tries to keep buses moving, you’ve created competing incentives.
• Integrated incident management: One source of truth for disruptions, with clear roles and timestamps.
• Data governance that’s boring but enforced: consistent IDs for stops/vehicles, standard definitions for “delay,” “missed trip,” “crowding.”
• Feedback loops to operations: analytics must translate into dispatch actions, maintenance plans, or service changes.

AI becomes valuable when it sits inside that loop.

Where AI actually helps transit agencies (and where it doesn’t)

Answer first: The highest-ROI AI use cases in public transport are prediction, prioritization, and anomaly detection — not flashy autonomous decisions.

Public sector leaders often feel pressure to “do AI” quickly. I’m opinionated here: start with operational clarity, not novelty. The best early wins are narrow, measurable, and tied to existing workflows.

1) Demand forecasting for hybrid cities

When travel patterns are volatile, schedule planning becomes guesswork.

AI-driven forecasting can combine:

• historical ridership,
• service levels,
• school terms,
• major events,
• weather,
• and even day-of-week changes in remote work.

The output shouldn’t be a fancy chart. It should be a decision support tool that answers:

• Which routes need higher frequency this month?
• Where should we adjust first/last-mile connections?
• Which time windows are under-served now?

That’s satiksmes plūsmas analīze applied to real service planning.

2) Predictive maintenance for reliability (and credibility)

Riders don’t care whether your failure was “unpredictable.” They care that the train didn’t show up.

Predictive maintenance focuses on probability of failure and maintenance prioritization, using signals like:

• vibration/temperature readings,
• door cycle counts,
• brake wear indicators,
• fault codes,
• time since last overhaul,
• and environmental conditions.

The point is not to eliminate failures (impossible). It’s to:

• reduce catastrophic breakdowns,
• increase mean time between failures,
• and plan maintenance windows that minimize service disruption.

For AI in public services, maintenance is a great proving ground because the outcome metrics are tangible.

3) Disruption prediction and better rider information

Real-time passenger information is one of the fastest ways to rebuild trust.

AI can flag anomalies such as:

• “bus is off-route,”
• “vehicle is stuck,”
• “headway gap is forming,”
• “station crowding is unusual.”

But the important part is what happens next:

• Does dispatch get an alert with recommended actions?
• Do riders get a message that’s specific and honest?
• Do customer service teams see the same status as operations?

A simple rule: If internal teams disagree about what’s happening, riders will notice.

4) Smarter staffing and resource allocation

Post-pandemic constraints aren’t only demand-side. They’re operational: staff availability, training pipelines, overtime costs.

AI can support:

• shift planning,
• overtime risk prediction,
• route-level staffing needs,
• identifying where absenteeism patterns correlate with particular depots or shift types.

This is sensitive territory in the public sector. It needs transparency and labor engagement. Still, it’s one of the most impactful areas because service reliability often fails due to staffing gaps, not technology gaps.

Where AI doesn’t help (unless the basics are fixed)

AI won’t rescue you if:

• your data is delayed by days,
• your asset inventory is incomplete,
• your teams don’t trust the numbers,
• or your procurement cycle turns “pilot” into a two-year ordeal.

AI amplifies what’s already true about your organization — good or bad.

A practical roadmap: from “innovation” to operational outcomes

Answer first: The fastest path to AI value in transit is a staged program: data foundation → targeted pilots → scale through governance.

Emily Yates’ emphasis on what it means to be innovative in public services points toward a healthier definition of innovation: less about novelty, more about results.

Here’s a roadmap that fits the public sector reality (budgets, procurement, accountability), while still moving quickly.

Step 1: Build a transit data foundation that teams actually use

Start by making existing data usable:

1. Create a unified operations dataset (vehicles, trips, stops, schedules, incidents, maintenance events).
2. Standardize identifiers (stop IDs, route IDs, vehicle IDs) across systems.
3. Set data quality thresholds (missing GPS pings, inconsistent timestamps, duplicates).
4. Choose 3–5 operational metrics that everyone agrees on.

This is “digital transformation in government” in its most valuable form: making everyday decisions less chaotic.

Step 2: Pick one use case with a clear “before/after” metric

Good first pilots are measurable and tied to a pain point:

• reduce bus bunching on a specific corridor,
• improve on-time departures at a terminal,
• cut repeat failures on a fleet subgroup,
• reduce crowding surprises on peak trips.

Define success in numbers before you build anything.

Step 3: Scale with governance, not heroics

Public sector AI programs die when they depend on one enthusiastic person.

Scaling requires:

• Model monitoring: drift, accuracy, false positives.
• Decision accountability: who acts on alerts and how fast.
• Security and privacy controls: especially with passenger data.
• Procurement design: avoid locking data into tools you can’t change.

If you want AI uzlabo e-pārvaldes pakalpojumus and public infrastructure management, governance isn’t bureaucracy — it’s how you keep the system dependable.

People also ask: common AI-in-transit questions (answered plainly)

Can smaller cities use AI in public transportation, or is it only for big agencies?

Smaller cities can absolutely use AI, but they should buy less and integrate more. Start with clean GTFS/AVL data, a basic analytics layer, and one pilot tied to reliability.

What data is needed first to use AI for transit reliability?

The minimum useful set is:

• real-time vehicle location (AVL/GPS),
• scheduled trips and stop times,
• incident logs,
• maintenance records (even basic),
• and consistent timestamps.

How do you handle transparency and public trust with AI?

Treat AI outputs as recommendations with traceability. Publish what you measure, explain what changes you’re making, and keep a human accountable for decisions.

What to take from Emily Yates’ approach (and apply locally)

The most transferable lesson from the SEPTA perspective is simple: innovation in public transport is operational discipline with better information.

If you’re working in a municipality, a transport authority, or a smart city program, focus your AI efforts on three outcomes:

• Reliability: fewer surprises, more consistent headways.
• Efficiency: better use of fleet, staff, and maintenance windows.
• Trust: accurate, timely communication that matches what’s happening on the street.

Those outcomes map directly to the goals of Mākslīgais intelekts publiskajā sektorā un viedajās pilsētās: improving city infrastructure management, mobility analytics, and data-driven decisions that residents actually feel.

If you’re planning your 2026 digital roadmap right now, ask one hard question: Which transit decision would become easier next month if your data was unified today?