Been hearing a lot about AI tools for document processing. Anyone tried using ChatGPT, Claude, or similar for analyzing utility bills? NPPD's bills have gotten more complex with all the new rate riders and I'm wondering if AI could help identify anomalies or billing errors faster than manual review.
AI tools for utility bill analysis - anyone experimenting?
Hugh, I've been experimenting with Claude for Avista bill analysis. You can upload PDFs and ask it to extract specific data or look for inconsistencies. Found a fuel cost adjustment error that I might have missed - saved client $3,200. But you still need to verify everything, the AI can hallucinate numbers.
Dale that's interesting. National Grid's bills are pretty standardized so might work well. What's your typical workflow? Do you upload individual bills or batch process them? And how do you handle rate schedule verification - does the AI actually understand tariff structures?
Carl, I upload individual bills and give it specific prompts about what to look for. The AI is pretty good at basic math verification but struggles with complex rate structures. For tariff analysis I still do that manually. Think of it more as a smart calculator than a replacement auditor.
Makes sense. NPPD has some unusual coincident demand charges that would probably confuse an AI. But for basic sanity checks on kWh calculations and rate applications it might speed things up. Anyone tried it with interval data analysis?
Hugh, I tested it on some Chugach Electric interval data. The AI is actually pretty good at spotting usage patterns and anomalies in 15-minute data. Found some demand spikes that lined up with equipment issues. But again, you need domain knowledge to interpret what it finds.
This is fascinating stuff. Georgia Power's new TOU rates have so many time periods that AI might actually be helpful for validation. My concern is client confidentiality - are you guys comfortable uploading customer bills to these cloud services?
Ray, great point about confidentiality. I only use it for internal analysis and redact customer info first. Some clients specifically prohibit cloud processing in their contracts. There are local AI models you can run but they're not as capable as the big cloud services.
Been following this thread with interest. PPL has started using some kind of AI for their own billing validation, ironically. Caught them in an error recently where their system flagged something as unusual but they billed it anyway. $15,000 refund for my client.
Donna that's hilarious - using AI to catch utility AI errors! MGE is probably heading in the same direction. I think the key is treating these tools as assistants, not replacements. They can process data faster than we can but they don't understand the business context.
Exactly Linda. The AI can crunch numbers all day but it can't negotiate with the utility or explain to a client why their bill is wrong. Still need human expertise for the important stuff. But for initial screening and data validation, it's a game changer.
Great discussion everyone. I've been testing some AI tools myself here in Memphis with MLGW bills. The technology is impressive but Ray raises the right concerns about confidentiality. Make sure your client agreements cover any third-party processing before experimenting. Better safe than sorry from a liability standpoint.