Curious if anyone has experimented with AI tools for auditing work. I've been playing around with ChatGPT for analyzing FirstEnergy tariff language and it's surprisingly good at explaining complex rate structures. Obviously can't replace human judgment but wondering if others have found useful applications. What are your experiences?
Anyone using AI/ChatGPT for utility auditing yet?
I've been testing it for initial bill reviews and it's hit or miss. Good at catching obvious calculation errors but struggles with utility-specific nuances. Used it to analyze an MLGW industrial rate schedule and it missed some key demand ratchet provisions. Still useful as a starting point though - saves time on the basic stuff so I can focus on complex issues.
Randy's right about the nuances. I tried using it to analyze Duke Energy's time-of-use schedules and it completely misunderstood the on-peak definitions for Florida vs the Carolinas. However, it's excellent for writing client reports - give it the key findings and it creates professional summaries that just need minor editing. Saves hours of writing time.
Been using it extensively for Dominion Energy accounts here in Virginia. The key is being very specific with prompts and always double-checking the output. I created a template prompt that includes rate schedule details and it does a decent job flagging potential issues. Found a $12,000 billing error last month that the AI initially identified, though I had to verify the math myself.
Harriet that's encouraging. Could you share your prompt template approach? I'm struggling with how much context to provide without overwhelming it. The FirstEnergy tariffs are complex enough that I feel like I need to explain half the utility industry just to get useful responses.
I start with basic utility info - company name, service territory, rate class, then paste the relevant tariff sections. Key is to ask specific questions rather than "analyze this bill." Instead of general analysis, I ask things like "does the demand charge calculation match the tariff provisions" or "identify any discrepancies between billed kWh and rate blocks."
That's a good approach Harriet. I've also found it helpful for explaining complex utility concepts to clients. Instead of spending 30 minutes walking through demand ratchets and time-of-use rates, I can generate clear explanations tailored to their specific situation. The client gets better understanding and I save time for actual analysis.
One thing to watch out for - I caught ChatGPT making up tariff provisions that don't actually exist. It was confident about a Duke Energy rate rider that was discontinued years ago. Always verify against the actual published tariffs. The AI is great for processing information you give it, but don't trust it to know current utility rates without verification.
Absolutely Leah - hallucination is a real issue with these tools. I always treat AI output as a rough draft that needs fact-checking. But even with that limitation, it's already changed how I work. Faster report writing, better client communications, and it catches some errors I might miss in routine reviews. Just have to use it intelligently.