What problem does it solve?
A public transport operator runs a large, always moving network and has to tell a diverse rider base what is happening on it right now: is my bus coming, why is my train delayed, is the elevator at this station working. The Chicago Transit Authority (CTA), an independent government agency, operates one of the largest transit systems in the US, more than a million rides on buses and trains on an average weekday, 24 hours a day, across the City of Chicago and 35 surrounding suburbs, serving a diverse community that includes many multilingual commuters.
Traditional channels can struggle to keep up: a call centre works through a queue, station signage and a schedule app show the plan more than the live reality, and a rider who wants to report a problem, a dirty train, a broken air conditioner, a safety concern, often has to call or wait to flag someone down. NJ Transit, for example, currently reaches riders through station and onboard digital signage, its DepartureVision and MyBus systems, mobile apps, websites, SMS and push notifications, email alerts, social media, real time and third party data feeds, and public address systems, and the agency says it wants to unify these into one authoritative source of information.
- The Chicago Transit Authority operates one of the nation's largest public transportation systems covering the City of Chicago and 35 surrounding suburbs, operating 24 hours a day with over a million rides on buses and trains on an average weekday.Chicago Transit Authority Connects with City: AI Chatbot Bridges Language Barriers and Empowers Riders (2025)
How does it work?
- Understand the request. The agent classifies whether the rider is asking a schedule or trip question, asking about a known disruption, or reporting a problem on a vehicle, at a station or with staff.
- Answer from live data. Trip and delay questions are answered from the operator's real time vehicle location, schedule adherence and service alert feeds, not a static timetable.
- Turn a report into a case. A free text report ("the AC is broken on the Red Line") is structured into a category, a location and an urgency, and logged with a reference number the rider can follow up on.
- Flag what is urgent. Safety concerns and other urgent situations are flagged for fast human follow up, on a stated time target, rather than sitting in the same queue as a routine cleanliness report.
- Hand over what needs a person. Safety incidents, complaints, unsupported languages and complex itinerary questions go to a human agent with the conversation already captured.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Emerging
- Channels
- Web chat, Mobile app, SMS
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Customer experience, Lower cost to serve, Inclusion and access, Speed and cycle time.
Indicative value
A transit agency carrying 150 million passenger trips a year
USD 33,750 to USD 551,250
Human handled contact cost avoided per year
How this is calculated
Formula: trips * contactsPerTrip * automationShare * costPerContact. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Passenger trips per year trips, trips per year | 150,000,000 | 150,000,000 | The reference agency. |
| Assisted contacts (calls, chats, social posts, in station reports) per trip contactsPerTrip, contacts per trip | 0.001 | 0.002 | Editorial assumption, replace with your own contact volume. |
| Share of contacts the agent resolves or logs without a person reaching them first automationShare, fraction of contacts | 0.15 | 0.35 | Editorial assumption, replace with your own. No evidence on this page reports a containment, automation or deflection share: the customer service reach and conversation completion figures Google Public Sector reports for CTA measure different things, and NJ Transit's Navvie is still a pilot with no outcome disclosed yet. For scale only, CTA staff review over 250 incidents a week across a system with over a million weekday rides (Google Public Sector), which does not by itself imply a share of contacts. |
| Cost of a human handled contact costPerContact, USD per contact | 3 | 7 | Editorial assumption for a blended phone, chat and social contact. Replace with your own fully loaded cost. |
What it leaves out: Gross avoided contact cost only. It leaves out the cost of running the AI and its integrations, the value of faster and more accurate disruption information to riders, and any change in the number or quality of maintenance and safety reports the agency actually receives.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
NJ Transit
United States · Logistics and transportation · 2026
NJ Transit launched Navvie, its first AI powered chatbot, alongside a redesigned website; Mass Transit reported this by early September 2026. Navvie is available around the clock to help riders plan trips and get schedules, alerts and transfer information. NJ Transit describes it as a pilot: results are being analysed before a decision on integrating it into the mobile app, and the agency separately issued a request for information for a larger, unified real time customer communications platform.
No outcome disclosed.
Chicago Transit Authority
United States · Logistics and transportation · 2025
The Chicago Transit Authority (CTA), the independent government agency that runs Chicago's buses and trains, worked with Google Public Sector and Quantiphi to build Chat with CTA, a multilingual virtual assistant on its website. Riders ask when their bus is coming and report issues; the chatbot answers in five languages (English, Spanish, Polish, Simplified Chinese and Filipino/Tagalog) and delivers detailed and timely reports to maintenance crews, including flagging urgent situations for fast follow up. CTA staff review over 250 incidents a week spanning buses, trains and train stations. Google Public Sector says the chatbot "has grown CTA's customer service reach by over 63%" and that "since launch, there's been a 16% improvement in conversation completion".
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Real time vehicle location, schedule adherence and service alert feeds
- A taxonomy of report types (cleanliness, mechanical, safety, lost property, staff conduct) mapped to the right team
- Current fare, accessibility and policy content
- Rider language data, to prioritise which languages to support first
Systems to integrate
- Real time transit operations feed (vehicle location, schedule adherence, service alerts)
- Incident or work order system for maintenance and operations teams
- Notification channels (SMS, app push, website and station alerts)
- Contact centre or social media platform for handover
Complexity: Medium
Answering from a static timetable is easy; the work is integrating live vehicle and service alert feeds so answers reflect reality during a disruption, and wiring reports into a system that maintenance and operations teams actually work from, in more than one language.
- 1
Start with schedule and disruption questions
Answer "when is my bus or train coming" and active service alerts first: the highest volume, lowest risk questions, and the ones a static timetable answers worst during a disruption.
- 2
Add structured issue reporting once answers are trusted
Turn free text reports into a category, a location and an urgency, and give every report a reference number the rider can check back on.
- 3
Define what counts as urgent, in writing
Agree with operations and safety teams which report categories and keywords trigger fast human follow up, and the time target for that follow up.
- 4
Support the languages your riders actually speak
Prioritise languages by rider population data, not by what is easiest to add first, the way CTA supports English, Spanish, Polish, Simplified Chinese and Filipino/Tagalog.
- 5
Test before riders do
Build a test set of real questions and reports per category and per supported language, including ambiguous and urgent ones, and run it on every change.
- 6
Widen channels once the first one is proven
Launch on the website or app first, measure containment and report routing accuracy, then add messaging channels and voice.
Guardrails
- Trip and disruption answers only from live operational data, with a refusal when the feed is stale or unavailable
- Every report gets a case reference and a routed owner, tracked to closure
- Urgent or safety related reports flagged for human follow up within a stated time target
- Input and output guardrails against abuse and off topic prompts, tested after every change
- Personal data in reports (names, contact details, photos) masked in logs and model prompts
KPIs to instrument
- Share of contacts resolved or logged without a person, per question and report type
- Time from an urgent report to a confirmed human follow up
- Reports confirmed as real issues by maintenance or operations teams, versus all reports logged
- Customer satisfaction on chatbot interactions versus human handled ones
- Language coverage of the assistant against the rider population it serves
Human in the loop
Operations and safety staff review every flagged urgent report and decide the follow up. A team monitors handover reasons and unmatched report categories weekly, and approves any new report category or supported language before it goes live.
Common failure modes
- Confident but stale disruption information
- The agent repeats a schedule that a live disruption has already overtaken. Ground trip and delay answers only in live feeds, and say clearly when live data is unavailable.
- Reports that go nowhere
- A report is logged but never reaches a team that acts on it. Wire every report category to an owning team and track reports to closure, not just to intake.
- An urgent situation misclassified as routine
- A safety relevant report is filed as a routine cleanliness complaint. Err toward escalation on ambiguous language and review missed urgent cases weekly.
- Coverage gaps that exclude riders
- The agent supports only the languages that were easiest to add, leaving other riders no better off than before. Prioritise languages by rider population, not convenience.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
Article 50(1): riders must be told they are dealing with an AI system, unless that is obvious from the context. Answering trip questions and logging reports is not a listed Annex III use; it would need a fresh assessment if the same agent decided eligibility for a reduced fare, a concession or paratransit access, which touches access to an essential public service.
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Riders must be informed that they are interacting with an AI system unless this is obvious from the context.
Controls to put in place
- AI disclosure at the start of every conversation
- Case reference and an owning team for every report, tracked to closure
- Escalation rules for urgent and safety related reports, with a monitored time target
- Guardrail and regression tests rerun after every model or prompt change
Frequently asked questions
- What results have transit agencies reported from this kind of chatbot?
- Google Public Sector reports that the Chicago Transit Authority's Chat with CTA chatbot, built with Google and Quantiphi, grew CTA's customer service reach by over 63% and lifted conversation completion by 16% since launch, and that it helps intercept urgent situations within five minutes of a rider's first message. NJ Transit's Navvie is a narrower assistant for trip information only, launched as a pilot by early September 2026 according to Mass Transit, and the agency says results are still being analysed.
- Does this replace real time apps like a trip planner or a map app?
- No. It answers from the same kind of live vehicle and service alert data those apps use, but inside a conversation, and it adds a way to report a problem and get a case reference, which a trip planner does not do.
- How does the agent decide a report is urgent?
- Operations and safety teams agree in advance which report categories and language (for example anything describing a safety threat) trigger fast human follow up, with a stated time target, rather than joining the same queue as a routine cleanliness report.
- What should stay with a person?
- Safety incidents, complaints, lost property claims above a set value, languages the assistant does not yet support, and any itinerary question complex enough that a scripted answer would mislead the rider.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for public transit passenger information and disruption reporting", last verified 28 September 2026, https://www.blits.ai/ai-use-cases/public-transit-passenger-information-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 28 September 2026: First published