Previous article linked

Having clean data on past usage and current rate plans isn’t enough to optimize our customers’ utility bills - we need context. Therefore, we enriched our dataset with external inputs like weather, location features, electricity market structure and alternative provider data. After cleaning and aligning timestamps, we used models to:

• Fill gaps and detect anomalies.
• Match stores to similar buildings using KNN (K Nearest Neighbor) based on usage and location.
• Forecast rate changes and evaluate incentives over 12 months.

We started pulling in external feeds—weather data, location-specific features such as the area in square feet, and alternative energy providers. Combining these disparate data sources was a real pain: aligning timestamps, resolving conflicting records, and dealing with missing values became a headache.

After going through the cleaning and normalization process, we set up machine learning models to automatically fill gaps and flag anomalies. We used the features and their corresponding weights like weather data, location-specific features such as the area in square feet to identify similar buildings and their usage data generated by a simulation model to the location of our customer through a KNN model.

With a cleaner, enriched dataset, we moved on to evaluating rate plans through a recommendation engine that automatically finds the lowest-cost rate plans and energy suppliers, as well as state incentive and rate plan recommendations for our customers to save further.

This is where AI-driven optimization makes the leap from parsing to decision-making. TrueMeter’s optimizer takes a top-down approach: start with actual usage data, apply eligibility rules, simulate scenarios, and constrain switching to what regulators allow.

For example, in our project with Yoshinoya, each location’s current plan was mapped, then cross-checked against eligible CCAs and rate plans. From there, the optimizer ran constrained scenarios:

• Assume usage remains consistent with last year.
  
  
• Ignore mid-year tariff changes.
  
  
• Allow only a limited number of rate plan/utility provider switches per year.
  
  

Here, LLMs again play a role: they help encode utility-specific switching rules from legal documents into machine-readable logic. They can parse the “fine print” of tariffs — the clauses that say, for example, “this customer class may only switch once every 12 months” — and map those constraints into our optimization engine.

The output wasn’t just a theoretical “best plan.” It was a list of concrete, executable moves: which plans to switch into for the next two billing cycles, and which paths to follow for the next twelve months.

Implementing through Automated Processes:

‍

Optimization on paper is meaningless without execution. A CFO cannot take a spreadsheet of “ideal plans” to the bank. Real savings require switching accounts, verifying bills, and monitoring over time. That’s why EcoTrove’s system is designed as a closed loop, not a one-time recommendation engine.

When the optimizer identifies a better plan, the system generates precise instructions: which account, which plan, which effective date. Today, our Customer Success team carries out those changes, guided by clear playbooks. In the future, these actions will be automated end-to-end through secure workflows, with humans remaining in the approval loop.

But executing a switch is only the beginning. Each new bill is compared to projections, and variances are fed back into the model. If the savings don’t materialize, we know quickly and can adapt. Over time, this feedback loop makes the optimizer sharper and more reliable.

The cycle continues month after month, because eligibility and tariffs change. Optimization jobs are scheduled according to each utility’s switching cadence, and usage patterns are monitored continuously. If a site’s load changes significantly or if a new rate becomes available, a new recommendation is triggered.

The benefit to customers is straightforward. Without this loop, optimization would be static — a one-time consulting report. With the loop, savings compound year over year. Businesses not only realize immediate cost reductions, they also stay aligned with evolving tariffs, regulatory pilots, and incentive programs.

In practical terms, this loop means three things for customers:

• Recommendations are executed, not just delivered.
  
  
• Bills are verified against expectations, catching errors and refining models.
  
  
• Optimization is continuous, not one-and-done.
  
  

In today’s policy environment, where regulators are rolling out dynamic programs and utilities are shifting cost structures, continuity is essential. Businesses can’t afford to get locked into yesterday’s best option.

Next article linked