1. TrueMeter’s AI: Parsing Bills 

Ever looked at your utility bill? It’s packed with arcane terms: franchise fee surcharge, conservation incentive, electric public purpose, power charge indifference adjustment, on-peak, off-peak, super-off-peak… It might as well be ancient Sumerian.

And those are just the line items. Each utility, each state, and each CCA (Community Choice Aggregator) has its own tariffs, schedules, riders, and incentives. Add in the fact that the data comes in inconsistent, non-tabular formats, and parsing it is a nightmare.

At TrueMeter, we built an AI energy agent that continuously lowers and pays the power bills for brick-and-mortar businesses—delivering a single fixed monthly invoice regardless of how many locations they operate or where. Our core engineering challenge was converting messy utility data into structured inputs for systems that run optimizations, cost-benefit analyses, and forecasting.

‍

The AI Pipeline: From Chaos to Optimization

We developed an AI pipeline that tackles the full lifecycle of utility data:

• Automated Data Capture: Instead of relying on customers to upload files, we can log into utility portals, download bills, and pull smart meter data automatically.
  
  
• Parsing Utility Reports: AI models process bills, tariff PDFs, and screenshots to identify layouts, detect text, and group relevant entities such as billing dates, charges, and usage data. LLMs enrich this raw extraction — normalizing inconsistent field names (“gen chg” → “generation charge”), correcting errors, and mapping items into a standardized schema.
  
  
• Tariff Normalization: LLMs are especially valuable here. Utility tariffs are often hundreds of pages long, written in legalistic text with complex tables. We use LLM-driven extraction to pull rate components and translate them into JSON with utility-agnostic attributes (e.g., seasonality, time-of-use windows, demand tiers). This makes it possible to compare across thousands of plans.
  
  
• Optimization Engine: With clean data in place, our models simulate multiple rate plan and incentive scenarios. LLMs help generate cost estimation logic for new tariffs quickly, applying utility rules to forecast future bills. By taking last year’s usage as a baseline, adding other considerations and applying switching rules, the optimizer identifies the lowest-cost plan available for the next billing cycles and beyond.
  
  
• Automated Billing & Monitoring: Finally, automation workflows execute plan switches and payments, while anomaly detection models flag outlier bills or suspicious line items. The periodic re-optimizations to lock in ~10%+ savings.
  
  

The Problem: Why This Was Previously Impossible

Before LLMs, this level of parsing and reasoning was intractable. You might build OCR scripts for one utility, but the moment you expanded to another with a different bill format, everything broke. Tariff PDFs — with their nested tables, footnotes, and conditional clauses — were too irregular for traditional parsers.

Imagine Maria, CFO of Pizza Queen, a franchise with 2,000 stores nationwide. Power bills consume 5–7% of revenue—double or triple net profit margins.

Here’s what Maria would face trying to optimize one store:

1. Connect: To begin optimizing, she first connects to the data source by setting up online accounts for the 2+ gas and electric utilities in just one of Pizza Queen stores.
2. Optimize: Since utility providers do not provide standardized usage data all in one place, she would have to painstakingly parse each monthly bill, compare them to other energy companies’ offers, and explore federal, state and local incentive programs for rebates or reduced rates. She needs to spend days if not weeks to parse and analyze the cost and benefit of each option and compare them to each other.
3. Implement: Undeterred, she pushes forward and negotiates with five different energy suppliers, each offering various rate structures. Recognizing the effort involved, she assigns a dedicated employee to handle ongoing vendor relationships and track usage trends.
4. Pay: Finally, she sets up payments for the store’s monthly utility bills and evaluates whether outsourcing the pay process—at about $1–2K per store per year—would be worth the time saved.

She certainly does not have time to do this for every single Pizza Queen out there. If Pizza Queen has stores in multiple states, this complexity is magnified given the different energy companies, regulations and incentives available in each. 

Faced with this impossible task, most CFOs and franchise owners simply throw up their hands and do nothing—accepting higher bills that quietly erode their margins. Those who do try to act often go down one of two costly paths: they hire consultants who charge hefty fees to deliver a one-time audit that quickly becomes outdated, or they rely on bill-pay agents that do little more than process payments. In the latter case, late fees still pile up, and despite having access to all the data, these services offer no real insight into how to actually lower costs.

Can we solve this problem with software?

When we started TrueMeter, the question we asked ourselves as developers, was whether we could solve this problem in an efficient and automated way from the connection phase to the payment:

Connecting with AI:

‍

‍

Our first hurdle was accessing and standardizing billing data. Utility APIs provide structured smart meter data (15-min intervals, bill periods, etc.), but miss critical details from itemized bills like taxes, charges, and addresses. Without a utility industry-standard format for the data or its presentation, developers like us face significant challenges when processing this information for energy reporting and analysis.

We let users securely authorize bill access and then we use AI to scrape data from PDFs and images.

Here’s a sample account summary from a local Bay Area restaurant:

‍

‍

We built a utility-agnostic system that turns screenshots into structured, analyzable data:

• Image Preprocessing: Normalize and extract visual features.
• OCR: Detect and recognize text regions.‍
• Layout Parsing: Identify structural elements (tables, headings).
• Entity Extraction: Tag and group relevant tokens (e.g., billing date, usage).
• Schema Mapping: Convert entities to a predefined JSON format.‍
• Prompt-Based Fine-Tuning: Use LLMs to normalize, correct, and enrich values.

‍

After constructing the initial pipeline, we validated with a sample eval set of bills to minimize the hallucination rate. The billing and usage data extracted through the utility API and bill PDFs need to be combined with the tariff data that is scattered in unstructured PDFs. We created a parsing system to extract rates from thousands of utilities and convert them into a unified JSON format with utility-agnostic rate components—enabling scale and adaptability.

‍

‍

The outcome of this process is a decision-ready dataset. For each location, we know the current plan, the eligible CCAs, the possible alternatives, and the constraints on switching. In practice, that means when a national chain like Yoshinoya asks which plan is best for its restaurants, we can answer across all its locations consistently.

To appreciate why this matters, consider the environment businesses face today:

• CCAs now serve more than 30 percent of California customers, complicating eligibility.
  
  
• Tariffs change quarterly as utilities integrate renewables and shift costs.
  
  
• Regulators are pushing time-of-use and dynamic pricing, multiplying the complexity of bill components.
  
  

In this shifting landscape, businesses need more than digitization; they need clarity. You can’t optimize what you can’t measure, and you can’t measure until you’ve structured the chaos. That’s the foundation of everything else we do.

Next article linked