24/7 EV Charging Call Center: A Cost-Sustainable Hybrid AI Model
asistanim.ai Team
Editor
The shape of the problem
Turkey's 23 March 2026 amendment requires every charging network operator to offer a call center reachable around the clock. The numbers explain the operational pain: for a mid-sized CPO overnight volume between 22:00 and 06:00 is typically an eighth of daytime traffic, yet each overnight call tends to be urgent ("charging cut off mid-session", "connector won't release", "payment went through but no session started"). Manning that volume with the same daytime model means staffing 2-3 people who are actually busy for only a few minutes a night.
The result has been either an overpriced overnight shift or a line that drops drivers into voicemail. After the amendment, voicemail is out.
Three classic models and why they stall
| Model | Cost | EPDK compliance | Failure point |
|---|---|---|---|
| Full human 24/7 | High (3 shifts, 9-12 FTE minimum) | Full | Cost does not scale, night staffing is unstable |
| BPO outsource | Medium | Partial (ISO audit trails often scattered) | No OCPP backend knowledge, routes instead of resolving |
| WhatsApp and email only | Low | Non-compliant | No voice access, SLA timer not maintained |
None of these three lock in EPDK compliance without bending the P&L.
Hybrid model: AI as first ring, engineer as second
The backbone of a sustainable 24/7 setup is two-layered.
- First ring: voice AI agent. Answers every call. Authenticates the caller, looks up the station and session, picks the language, runs simple commands (remote stop, billing address change, pricing lookup).
- Second ring: on-call engineer. Steps in for technical cases the agent cannot resolve. Usually a one or two person weekly rota is enough, because the agent absorbs most overnight calls and the engineer's load stays well below daytime.
This compresses a 9-12 FTE shift load into a 2-3 engineer rota. Rotas replace shifts.
What the agent actually does on the first ring
Walk through it as a working model.
Call 1: "Charging cut off mid-session"
The agent matches the membership number to the inbound caller, queries the last session in the OCPP backend, reads the disconnect reason (e.g. CP-OUT-OF-RANGE). If the fault is transient, the agent restarts the session remotely. If it is persistent, the agent flags the station as a case, opens a refund and suggests a nearby alternative station to the driver. Duration: 90 seconds.
Call 2: "Connector is stuck"
The agent performs a station check, sends the UnlockConnector command. If the connector does not release, the agent escalates to the on-call engineer with station address, driver phone and error log attached. Duration: 60 seconds plus engineer handover.
Call 3: "German-speaking truck driver"
The agent detects the language in three seconds and continues in German. Wayfinding or card issues are answered in the driver's language with no translator in the loop.
Call 4: "Update my invoice address"
After identity verification the agent updates the record in the CRM and sends a confirmation SMS. Duration: 75 seconds.
These four cases cover roughly 85% of overnight volume. The remaining 15% goes to the engineer.
Without OCPP integration the model collapses
The agent's ability to resolve a call hinges on real-time access to the OCPP backend. The following surfaces have to be fully wired:
- Station state over OCPP 1.6J or 2.0.1.
- Remote commands (RemoteStartTransaction, RemoteStopTransaction, UnlockConnector, Reset).
- Session history, error codes, last meter values.
- Wallet and payment service (refund initiation, failed transaction lookup).
- CRM or ticketing system.
Without backend visibility the agent can only say "I have logged the case, the team will get back to you". That starts the 15 business day SLA clock but does not lift the human load.
Cost comparison
For a CPO running roughly 1,500 charge points the annual cost lines stack like this:
- Full human 24/7: 9-12 FTE plus management plus night premium.
- BPO call center: per-call pricing plus minimum capacity commitment, plus OCPP integration engineering on top.
- Hybrid AI plus on-call engineer: minute-based AI pricing plus 2-3 on-call engineers plus integration setup.
Payback for the hybrid model is typically under six months for a CPO of that scale because the most expensive shift hours (overnight and weekend) drop out entirely.
Rollout roadmap
- Map current call distribution by hour and language.
- Open access and remote command rights on the OCPP backend.
- Train the agent against the top five call scenarios.
- Write the escalation rules and the on-call rota.
- Switch the call recording archive on from day one for ISO 18295 audit.
- After two weeks of parallel running put the agent on first ring.
asistanim runs this stack as a pre-built model for charging network operations. OCPP integration, multilingual answering, the 15 business day SLA timer and the ISO 18295 audit archive ship together.
Frequently asked questions
- Does overnight call volume really justify a human shift?
- On most networks overnight volume is roughly a tenth of daytime, but every single overnight call matters because the driver is usually stranded. A voice AI agent absorbs the first ring and routes only genuinely technical cases to the on-call engineer.
- How do you handle foreign-language drivers?
- The agent answers in over 40 languages, English, German, Russian, Arabic and Farsi included. It detects the caller's language within the first three seconds and continues the conversation in it.
- Why an AI agent instead of a BPO call center?
- BPOs hit three walls: night shift cost, no OCPP backend knowledge and scattered ISO 18295 audit trails. The AI agent solves all three in one place.
