What problem does it solve?
A driver who calls a charge point operator is usually standing at a charger that does not work, often with a low battery and a schedule. Support demand rises with every new charger, spikes when a payment system, app or backend fails, and much of it arrives outside office hours. A human team sized for the worst peak is expensive, and one sized for the average leaves drivers on hold at the moment they need help.
Most calls are routine: the session will not start, the cable is locked, the card or app is not accepted, or the driver does not know which charger to use. Many can be fixed from the operator's backend, which is why a voice agent with access to the charge point management system can resolve a share of them while the driver is still on the line.
Retell's case study says support cost at swtch was rising 300 to 400 percent a year, and EVPassport had a single in house agent covering office hours while most questions came at the start of a charging session.
How does it work?
- Identify the driver and the charger. The agent takes the charger or connector ID from the driver or the QR code on the unit, and finds the driver's account and the live session.
- Read the live status. It queries the charge point management system for connector state, error codes, the last messages from the charger and the payment or authorization result.
- Walk the driver through the fix. It checks the cable, the plug and the app step by step, in the driver's language, and asks for the minimum the system cannot see itself.
- Act on the charger within limits. Where the operator allows it, the agent ends a session, releases a cable lock or sends a soft reset, then watches the charger to confirm the result.
- Refund or compensate by rule. Failed sessions and double charges go through the operator's refund rules, with a person approving anything above a set amount.
- Hand over with context. Unresolved cases, hardware faults, safety concerns and angry drivers go to a human agent or a field technician with the transcript, charger ID and diagnosis attached, and create a fault ticket for the network operations team.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Phone and voice, Web chat, Mobile app, Email
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Containment rate | Too few to pool | about 70% | 1 | 1 vendor |
| Cost reduction | Too few to pool | at least 50% | 1 | 1 vendor |
Value drivers: Lower cost to serve, Customer experience, Speed and cycle time.
Indicative value
A charge point operator receiving 6,000 driver support calls a month
USD 86,400 to USD 345,600
Support cost avoided per year
How this is calculated
Formula: calls * resolvedShare * costPerCall. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Driver support calls per year calls, calls per year | 72,000 | 72,000 | The reference operator. Retell reports 5,000 to 8,000 inbound calls a month handled by swtch's agent. |
| Share of calls resolved by the agent without a person resolvedShare, fraction of calls | 0.3 | 0.6 | Editorial assumption, set below the vendor reported figures on this page (near 70% at EVPassport). The vendor publishes no definition of resolved; replace with your own rate. |
| Cost of a call handled by a person costPerCall, USD per call | 4 | 8 | Editorial assumption for a fully loaded cost per handled call; replace with your own figure. |
What it leaves out: Gross handling cost only. It leaves out the cost of the platform, telephony and integrations, repeat contacts, field visits avoided and charging revenue recovered when a session is fixed.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
swtch
Canada · Energy and utilities · 2026
swtch, a Toronto based EV charging network operator active in the United States and Canada, built an inbound voice agent called Lucas on Retell to take routine driver calls about accessing its charging services. It absorbs call spikes during infrastructure or cloud incidents while human agents keep the complex cases. Retell reports that customer support costs fell by over 50 percent, calls are answered in about 5 seconds and Lucas handles 5,000 to 8,000 inbound calls a month.
- Cost reduction: at least 50%
"Reduced customer support costs by over 50% while continuing to scale"
Claimed by: vendor
Citywatt
Germany · Energy and utilities · 2025
Citywatt, a German charge point operator with more than 50 high power charging parks, has an AI voice assistant from Lemonflow handling first level support for more than 1,500 charging points. It answers in more than 30 languages, ends charging sessions, releases cable locks and resets chargers remotely, and passes complex cases to human agents, including at night and on weekends. No resolution figure is published; the operator calls the language quality better than expected.
No outcome disclosed.
EVPassport
United States · Energy and utilities · 2025
EVPassport, an EV charging network software and services company, had one in house agent covering office hours while most driver questions arrived at the start of a charging session. It moved to Crescendo voice and chat AI assistants backed by human experts, went live in 33 days and now has 24/7/365 coverage. Crescendo reports near 70% instant AI resolution and human time to answer under 30 seconds.
- Containment rate: about 70%
"AI now resolves the majority of inquiries, holding near 70% instant resolution while keeping humans easy to reach."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Charger inventory with IDs, models, locations and QR codes
- Live connector status, error codes and session history
- Driver accounts, cards, RFID tokens and roaming partner data
- Fault tickets and past support transcripts to choose the first intents
Systems to integrate
- Charge point management system for status and remote commands
- Payment, billing and roaming platforms
- Ticketing and field service dispatch
- Telephony or contact center platform for voice and handover
Complexity: Medium
It needs read and command access to the charge point management system, which differs by software vendor and charger model, plus the payment and roaming platforms for refunds. Charging hardware and protocols vary, so the first release usually covers the most common error classes on the most common models.
- 1
Rank the contact reasons
Classify a month of calls and tickets. Start with the three or four reasons that the backend can fix without a technician, such as a session that will not start or a locked cable.
- 2
Connect read access first
Give the agent live status and history before any command. Let it diagnose and hand over with the diagnosis attached for a few weeks while you compare its reading with what agents conclude.
- 3
Add remote actions one at a time
Enable session end, cable release and soft reset each behind an allow list, a retry limit and an automatic check that the charger changed state. Log every command with the driver and charger.
- 4
Staff the surge
Test the agent against a simulated backend or app outage. swtch built its voice agent to absorb spikes during infrastructure incidents while people keep the complex cases.
- 5
Close the loop with operations
Feed the diagnoses and failed resets into the fault management process, so repeated errors on one model or site become maintenance work instead of repeated calls.
Guardrails
- Remote commands only from an allow list, never on a charger with an active session of another driver
- Limits on refunds and compensation, with approval above the limit
- Safety issues such as smoke, damaged cables or a vehicle fire go to a person and the emergency number at once
- AI disclosure at the start of the call and a clear route to a person
KPIs to instrument
- Share of calls resolved without a person, with repeat contact on the same charger within 24 hours counted as not resolved
- Time to answer and time to resolution, for the agent and for people
- Remote action success rate per charger model
- Refund and compensation volume per resolved call
Human in the loop
Human agents take hardware faults, disputed charges, safety reports and any driver who asks for a person. Operations staff review a sample of resolved calls each week and every unresolved call that created an alert, and approve changes to the allow list of remote actions.
Common failure modes
- Command on the wrong charger
- A mistyped ID leads to a reset on another driver's session. Confirm the charger by location or connector and check for an active session first.
- Looping on a hardware fault
- The agent keeps retrying resets on a broken unit and the driver waits. Cap retries and hand over with the diagnosis.
- Backend lag
- Status in the system trails the real state, so the agent tells the driver the charger is fine. Compare against the last message time and say when data is stale.
- Undisclosed AI
- Callers may not notice they are talking to an AI; swtch's source says most drivers do not realize it. Disclose it at the start of every call, as Article 50 requires.
- Language and noise
- Calls from a roadside or a car park with accents and background noise lower recognition. Test with real recordings and offer a chat or SMS fallback.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A voice or chat agent that talks to drivers falls under the transparency duty of Article 50(1): people must be told they are speaking with an AI system. It does not assess creditworthiness or decide access to an essential service, and the charging hardware itself stays under its own safety rules.
Controls to put in place
- AI disclosure on every channel and a documented route to a person
- Allow list and audit log for every remote command and refund
- Call recording retention and consent handling under the privacy policy
- Change control and regression tests for each charger model and intent
Frequently asked questions
- What share of EV charging calls can an AI agent resolve?
Crescendo reports near 70% instant AI resolution for EVPassport, with its own human experts taking the rest. It is a vendor figure without a published definition of resolved, so measure repeat contact on the same charger before trusting a rate.
- Can the agent fix a charger during the call?
When it is connected to the charge point management system, yes for the common cases. Citywatt's Lemonflow assistant ends charging sessions, releases cable locks and resets chargers remotely.
- Does it replace the support team?
Not in the published cases. Lucas takes the routine inbound calls at swtch while human agents handle complex cases, and Retell reports that support cost at swtch fell by over 50 percent while the network kept growing. Citywatt's Head of Operations and IT calls the Lemonflow assistant "a genuine addition to our support team".
How to cite this page
Blits.ai AI Use Case Library, "AI agent for EV charging driver support and charger troubleshooting", last verified 11 November 2026, https://www.blits.ai/ai-use-cases/ev-charging-driver-support-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 11 November 2026: First published