Q&A with Marina Bukhanova on AI Driver Support: What Fin Resolves, and When a Human Takes Over
Most EV charging problems don't look like a broken charger. They look like a driver standing in a parking garage at 9 p.m. with a session that won't start and no one to ask. J.D. Power's 2026 U.S. EVX Public Charging Study found that 12% of public charging visits in the most recent quarter ended without a charge. ChargerHelp's 2025 reliability report found that more than a third of failed attempts happen on chargers that report as operational. When the hardware looks fine and the session still won't start, support is the product.
On July 3, 2026, Epic Charging launched Fin AI Agent, an AI support agent that answers driver questions inside the Epic Charging app and, since August, on the web for drivers who start a session by scanning a QR code at the charger. Between July 3 and September 1, 2026, Fin resolved 530 of 718 driver conversations without escalating to a person, about 74%, and brought first response time down from up to four hours to about 2.5 minutes.
We asked Marina Bukhanova, Epic's Director of Client Success and Utility Partnerships, what Fin handles, what it deliberately does not, and what changed for site hosts and for her team.
Fin by the numbers (July 3 to September 1, 2026)
Metric
Result
Driver conversations resolved without human escalation
530 of 718 (about 74%)
First response time
About 2.5 minutes
First response time before Fin
Up to 4 hours
Availability
24/7
Languages
English, Spanish, French
Channels
Epic Charging driver app and web (QR code flow)
Handles
Starting and stopping sessions, charger issues, payments, idle fees, site access and PIN questions
Escalates to a person
Anything that requires looking up a specific account, session or charger
Before Fin, what did driver support actually look like?
Before the chatbot, everything came through email or form in the app. Support has always been 24/7, but drivers were often waiting more than they should for an answer to something as simple as “why isn't my session starting?” Now most driver questions go straight to chat. It launched on the app home screen in July, and since August it's also on the web, so a driver can scan the QR code right at the charger and start chatting on the spot instead of opening the app first.
Today the bot resolves most charging-behavior questions on its own: starting or ending a session, location and charger details, payment setup, the fee structure at a site, and access issues.
How much faster is it, and what does that change for a site host or a fleet manager?
As of today the chatbot resolves about 74% of our driver conversations independently, and we still have human agents when escalation is needed. Our first response time is about 2.5 minutes. Before, when a customer submitted a ticket, it could take up to four hours to respond.
It makes a huge difference for site hosts, because they get fewer inquiries, and drivers get answers faster.
Editorial note: for questions that do need a person, Epic's data shows conversations that start in chat typically close in about a day, while tickets submitted by email take about three days.
What makes it accurate instead of generic? What is it actually working from?
It's accurate on anything covered by our knowledge base, meaning how our network and chargers actually behave, not generic FAQ answers. If someone asks why a session ended a certain way or what a specific error means, it can explain what's actually going on instead of giving the same boilerplate answer to every question.
It's also built to recognize what it doesn't know rather than guess, which is part of why we trust it to represent us.
Editorial note: Fin runs on Intercom's Fin AI Agent, configured on Epic's own support knowledge base, which is the same documentation Epic's human team works from.
What a resolved conversation looks like
Here is a real exchange from August, lightly anonymized. A driver opened the chat and typed: “I want to pay outstanding balance.”
Fin explained that an outstanding balance is cleared by updating the payment method in the app (Profile, then Payment Methods, then add or update a card, Apple Pay or Google Pay), noted that the balance is charged automatically once a valid method is on file, and asked whether the balance had cleared.
The driver answered: “No.”
Fin offered one more fix, removing the saved card and adding it again, since re-adding a card sometimes clears stale payment data. It then asked again, and added that it could connect the driver with a human agent if the issue persisted.
The driver answered: “It worked.”
The point is that within a couple of attempts, Fin gave the driver steps that actually resolved the problem, without repeating itself.
Where is the line between Fin and your team? What does it explicitly not handle?
The bot can cover general charging behavior, site-specific setup, and charger and vehicle details. What it can't do is look into our database. We intentionally do not provide Fin with access to our production database, based on our cybersecurity and compliance standards. So anything that needs a real account or session lookup, like “why did my specific charging session stop,” gets escalated to a human agent right away.
We use a CRM with escalation rules built in specifically so that the handoff happens automatically, and the driver isn't stuck repeating themselves to a person after already explaining it to the bot. Customer experience is the priority here.
Editorial note: in practice the boundary is data access. Fin answers from what Epic has documented about how the network behaves. It does not query Epic's backend for a specific account, session or charger, so it cannot investigate an individual session, change account details or resolve a hardware fault on its own. Those go to Epic's support team.
What has changed for your team and for customers?
The feedback since launch has been genuinely positive. Drivers notice when they get help instantly instead of waiting. For the team, it's freed up time to focus on site host support and on catching problems before they become tickets, instead of answering the same handful of questions over and over.
We're seeing ticket volume go down and CSAT go up at the same time, which tells us the bot is actually resolving things well.
What's next, and what are you deliberately choosing not to automate?
Our main goal is for the bot to stop sounding like an FAQ and start actually understanding the problem. When someone gets an automated reply, we want them to feel the issue is resolved, not that they've been run through a script.
The next step will be testing ways to connect it more deeply with our backend so it can eventually work with account- and session-specific details. But we're taking that slowly and testing it internally first, because we'd rather get it right than rush it out.
Why this matters if you operate chargers
A driver who can't start a session is not only a support ticket. Depending on the site, it is a failed session at a public location, a resident emailing the property manager, a fleet vehicle that leaves the depot without a full charge, or a one-star review that names your building. Support speed is part of charging infrastructure quality, and it belongs in the same evaluation as uptime, payments and load management. Federal NEVI rules already require that drivers have a way to report outages and malfunctions; the more useful question for any site is how fast that report turns into an answer.
Fin is one part of how Epic answers that question: an AI layer that clears repetitive questions in minutes, 24/7, in three languages, and a human team that picks up anything requiring account-level context, with the conversation already in hand. It is also the kind of integration Epic's CTO Michael Fridshtand described in April when he talked about putting AI “into your support stack to reply to customers faster.”
If you are comparing EV charging management software, ask every provider the same four questions: what share of driver questions gets answered without a person, how fast the first response is, what the AI is not allowed to do, and what happens to the context when a human takes over.
Quick answers
What can an AI support agent handle in EV charging? Repetitive driver questions that can be answered from documented network behavior: how to start or stop a session, how payments and outstanding balances work, what an idle fee is, why a site requires a PIN, and what a documented error means.
What should still go to a human? Anything that requires looking up a specific account, session or charger, changing account details, or judging a situation that the documentation doesn't cover.
How fast should EV charging support respond? Epic's Fin AI Agent responds immediately. When a human agent is needed, the average response time is about 2.5 minutes, and that number continues to improve. Before it launched, a ticket could wait up to four hours for a first response.
Does Epic Charging still have human support? Yes. Epic's support team handles every escalation, 24/7, and receives the chat context through CRM escalation rules so the driver doesn't start over. Epic also offers 24/7 phone support for select customers on higher subscription tiers.