6 AI Challenges We Addressed to Build an Agent That Actually Helps EV Charging Operators

As of late 2025, there are over 70,000 AI tools worldwide. And if you’ve tried a few of them, you probably agree that they often fall short when applied to real life problems. These problems might sound familiar:

• The AI output is almost right but it makes additional, basic mistakes
• Back-and-forth chat with the AI is frustrating and the AI usually misses the point
• The AI is too long-winded

Flipturn users don’t have time sit and craft the perfect AI prompts—they're in the field, on their phones, dealing with driver issues and site emergencies. When we built Flip AI, our AI charging operations agent, our goal was to create a simple, high quality product that helps EV charging operators with their real life problems, focusing on the following areas:

1. Data should be always accurate.
2. Reports should get right to the point.
3. Flip should prioritize actionable issues above “noisy” recurring problems.

Here are some of the challenges we addressed to take Flip AI from a prototype to a tool that EV operators use every single day.

1. Making the math correct

AI is only as good as the data it works with, and in our experience LLMs are good at working with natural language but bad at doing math. In early experiments we often saw reports that looked like:

District Yard: 2 chargers working, 0 chargers out of service, overall uptime 100%
‍- Charger 2, Port 1: Faulted status for 1,196 seconds - now operating normally

In this report, the math doesn’t add up between the 1,196 seconds and 100% uptime.

How we fixed it: In cases where we need the output to display multiple numbers, we have the underlying system feed those numbers to Flip AI directly. In this case we added a “site uptime percent” field that Flip can easily report.

2. Including vendor-specific explanations when reporting issues

Another early problem we saw was reports that contained vague problems without actionable follow up steps. For example:‍

Charger 35: Multiple error messages detected on both ports. While currently showing available, diagnostic logs show system instability.‍

What's the operator supposed to do with this? Which errors? Is it still happening?

How we fixed it: We built context-aware error lookup. Every manufacturer uses a different set of error codes, but they might map to a similar problem. For example, ABB's "ERR_GND_001" and Tritium's "T123" might both mean ground faults (electrical current leaks). Flip AI now looks up error codes on-demand to provide manufacturer-specific guidance and actionable recommendations.

Charger 35: Still faulted due to emergency stop button activation (2 occurrences). Faulted since yesterday.‍
- Recommended action: Inspect emergency stop button for physical damage or malfunction

3. De-noising vehicle-related errors

The AI would flag vehicle-side errors as charger problems and often hallucinate inaccurate recommendations for those errors. For example:

Charger #6: 2 "VehicleNotConnected" errors
- Continuing pattern from previous period (1 error yesterday, 2 today)
- Port functionality not impacted
- Recommended action: Monitor - may indicate loose connector or user behavior pattern

These errors are often caused by driver behavior (unplugging before the vehicle is ready) or vehicle-side issues, not charger malfunctions. Including them creates noise and distracts operators from real problems.

How we fixed it: We taught Flip AI to de-emphasize known vehicle protocol errors. When the data shows the charger is functioning correctly but the vehicle reported an error, Flip doesn’t bring these up as “issues” but puts them in a de-emphasized “notes” section. This lets operators focus on actual charger issues that need their attention. A new report looks like:

✅ No new or ongoing issues.

Additional notes: Charger #18 Port 1, Charger #4 Port 1, and Charger #10 Port 2 had vehicle-related errors but continued to have successful charging sessions after.

4. Highlighting when an issue has already resolved

Flip soon became really good at finding and summarizing charger faults, but if there were many faults in a row, it was hard to tell if the problem was still happening. For example:

Charger 30 Port 1: 17 CableCheckVoltageTooLow errors, blocking multiple charging attempts throughout the day ‍

This doesn’t tell you if the charger is still faulted, and if anyone has successfully used it since the error.

How we fixed it: We added two key pieces of information for Flip to use:

1. The most recent status of the charging port that errored
2. Whether or not there were any successful charging sessions after the error

After this fix, a similar charger issue would read:

Garage Stall 1: Both ports had insulation monitor failures at 3:48 PM CT Jan 11 that ended an ongoing charging session. No successful sessions since. Charger is currently available but may have a persistent cable insulation fault.

5. Offline chargers: urgent issue or being repaired?

Offline chargers were tricky to report well. If a charger goes offline unexpectedly, we want to call attention to it because it impacts the accuracy of energy reporting and energy management. However, if a charger has been offline for months, we don’t want to report that as an urgent problem because that usually means it's in the process of being de-commissioned or repaired. This how Flip used to report offline chargers:

Charger A remains offline

How we fixed it: We significantly improved offline charger health reporting by providing a new piece of data, offline_since , to Flip. Here is an example of how one our simulated offline chargers is reported now.

Charger has been offline since Dec 17, 2025 (over 1 month)

6. Concise metrics to share when "it’s all good"

When everything was fine, the AI would produce too much reassurance. For example, these additional notes were included for a site that had no operational concerns:

Additional notes:
- No errors reported on any charging sessions
- No failed starts or sessions ending early
- No alerts triggered during the 24-hour period
- All ports maintained 100% uptime with zero faulted or unavailable time
- Four ports (1, 2, 4, 5) achieved maximum utilization (100%)
- Charger firmware version 3.4.5 operating stably

Operators don't need to be told in six different ways that nothing is wrong.

How we fixed it: We pulled a charging summary to the top of the email to provide key metrics like uptime, session count, and connectivity at a glance. Then we reduced redundant status updates in the rest of the report. If a site has has been operating perfectly, Flip says so in one line and moves on.

Months of iteration paid off. Flip AI has seen strong adoption from top businesses and charging operators across North America. Here's what they're saying:

"This translates charger messages into English and puts it in one update I can review in 5 minutes. We're dedicating staff to solve the vehicle/charger interoperability issues Flip has highlighted" - Christian Bigger, Zero-Emission Fleet & Facilities Manager, Spokane Transit

"Flipturn's AI report is my go-to in the morning to see what I'm walking into. It's helped us identify and fix power module issues to keep chargers operational" - Roger Alvarez, Director of Depot Operations, Zeem Solutions

‍

Want to see Flip AI in action? Request a demo and we'll show you how it works with your charging network.

Already a Flipturn customer? Flip is included in every Flipturn plan and customers can turn it on starting today. Log in to get started.

About Flipturn

Flipturn is a leading EV charging management platform, helping organizations maximize charger uptime, process charging payments, and scale operations efficiently. Backed by leading investors including CRV and Accel, Flipturn serves Fortune 500 companies, commercial property owners, and major fleet operators across North America.