Prove it saved money: M&V for school energy projects

Every energy project ends with the same question from the board: did it work? You replaced the rooftop units, or retro-commissioned the controls, or finally enforced night setbacks — and then the next twelve months of bills came in higher than the year before. Rates went up. It was a hotter summer. A portable classroom village landed on the same meter. Did the project save money or not?

Comparing this year’s bill to last year’s can’t answer that, because the two years differ in ways that have nothing to do with your project. Answering it properly is called measurement and verification — M&V — and the core idea is simple even though the industry wraps it in acronyms.

Savings are a counterfactual

The savings from a project are the difference between what the building would have used without the project and what it actually used with it. The second number is on your bills. The first number doesn’t exist — the un-upgraded building is gone. You can’t measure a counterfactual; you have to model it.

That’s the entire trick of M&V:

  1. Take 12–24 months of data from before the project — interval data or monthly bills, plus weather and the school calendar.
  2. Fit a model of how the building behaved: how much energy it used as a function of outdoor temperature, day type, and occupancy.
  3. After the project, feed the model this year’s weather and calendar. It predicts what the old building would have done under the new conditions.
  4. Savings = model’s prediction − actual usage.

This is what “weather normalization” means in practice: the model, not the calendar year, provides the fair comparison. A hotter summer raises the prediction and the actual together, so the difference — your savings — survives.

Model predictionMetered use
Baseline yearProject completeYear after
Monthly use, MWh: baseline model vs metered — the gap after the project is the savings (illustrative)

If your district works with an ESCO, this whole-building, billing-data approach is what the contract language calls IPMVP Option C. It’s the standard for projects big enough to move the whole meter, which describes most school HVAC and controls work.

One report is not enough

The traditional deliverable is a savings report twelve months after the project closes. The number is real, but a one-time report has a blind spot: savings erode. Schedules drift back. Setpoints creep as staff make one-off adjustments. The economizer that was fixed sticks again. Studies of retro-commissioning consistently find a meaningful share of savings gone within a few years — usually not because the work was bad, but because nobody was watching afterward.

This is where continuous, model-based M&V beats the annual PDF. If the baseline model runs every month — predicted versus actual, plotted since project completion — erosion shows up as a narrowing gap within weeks, while it’s still a schedule fix rather than a lost year of savings. The same anomaly-detection machinery that catches after-hours waste is, mathematically, doing M&V; the only difference is that here the deviation you’re tracking is the one you paid to create.

Be honest about uncertainty

A savings estimate is a model output, and every model output has error bars. A trustworthy M&V result says something like “savings of 14% ± 4% at 90% confidence,” not “savings of exactly $61,432.”

The practical rule: your expected savings need to be large relative to the model’s noise. A baseline model that’s accurate to ±10% cannot credibly verify a measure that saves 5% of the bill — the signal drowns. That’s not a reason to skip small projects; it’s a reason to verify them at the equipment level, or bundle them, rather than pretending whole-building data can see them.

What to ask for

Whether the M&V comes from an ESCO, a consultant, or software, the questions that keep everyone honest are the same:

Districts spend real capital on energy projects, and the projects mostly do work. M&V is how you know — and continuous M&V is how you keep it that way after the ribbon cutting.