---
title: "Turning meter data into savings: baselines, KPIs and alerts"
description: "Why total kWh is a useless measure, how to build an energy baseline against production, and the four analyses that actually find money in metering data."
date: "2026-08-04"
author: "Divakar B"
source: "https://energycalchq.com/blog/turning-meter-data-into-savings"
license: "© 2026 Divakar B. Quote with attribution to https://energycalchq.com/blog/turning-meter-data-into-savings"
---

Most energy monitoring projects end the same way. The meters are installed, the
dashboards are built, everyone admires them for two weeks, and then nobody opens
them again. The data keeps accumulating and nothing changes.

The problem is almost never the data. It is that the dashboard answers a
question nobody has — "how much are we using?" — instead of the question that
leads to action: **"is that more than it should be?"**

Here are the four analyses that turn meter readings into decisions.

## 1. Stop reporting total kWh

Total consumption is not a performance measure. It goes up when you produce
more, which is not a problem, and it goes down during a shutdown, which is not
an achievement. A month-on-month comparison of total kWh tells you almost
nothing about how well the plant is running.

The measure that works is **energy against output**:

![Scatter of weekly energy against weekly production with a fitted baseline](/blog/energy-vs-production-baseline.svg "Two numbers come out of this line, and both of them are actionable.")

Plot weekly energy against weekly production — tonnes, units, pieces, occupied
room-nights, whatever your plant makes — and fit a straight line. Two numbers
fall out:

**The intercept** is what the site consumes at zero production. Fixed load:
lighting, HVAC, compressed air leaks, idling equipment, standby power, office.
You pay for it whether or not anything is made.

**The slope** is the specific energy consumption — kWh per unit produced. This is
your real efficiency number, and it is the one to track month on month, because
it is unaffected by how busy you were.

A plant whose total consumption fell 8 % last month might have produced 15 %
less. The slope tells you it actually got worse.

## 2. Attack the base load first

The intercept is nearly always the cheapest saving available, and it is the one
that keeps paying at 3 a.m. on a Sunday.

Look at a full week of interval data and find the minimum. That is your true
base load. Then ask what is running:

| Usual suspect | How to confirm |
|---|---|
| Compressed air leaks | Base load falls when the compressor is isolated overnight |
| HVAC running out of hours | Correlate against the BMS schedule |
| Lighting in unoccupied areas | Walk the site at night, once |
| Idling machines on standby | Submeter one, watch a weekend |
| Oversized transformer no-load loss | Constant, unaffected by anything |

Compressed air is the classic. A plant with a compressor running through the
night to feed leaks is burning full electrical cost to produce nothing, and the
meter data proves it in a way that a walk-round argument never does.

The test costs nothing: shut the plant down properly one weekend and watch the
interval data. Whatever is still being consumed at 4 a.m. on Sunday is
consumption you are buying for no output at all.

## 3. Manage maximum demand deliberately

Demand charges are billed on the peak 15 or 30-minute average in the month, so a
single bad half-hour sets the charge for all of it. That makes it uniquely worth
managing — and uniquely easy to manage badly.

With interval data you can see:

- **When the peak occurs**, and what was running. Usually two or three large
  loads that coincided by accident rather than by necessity.
- **How close you run to contract demand.** Exceeding it attracts a penalty rate,
  typically one and a half to two times the normal demand charge.
- **Whether your contract demand is right at all.** Consistently peaking at
  70 % of contract means paying for capacity you never use — the reasoning is in
  [connected load, maximum demand and
  diversity](/blog/connected-load-maximum-demand-diversity).

The fix is usually scheduling rather than equipment: stagger the start of large
loads, avoid running the compressor and the furnace up together, move a batch
process off the peak. That costs nothing and it shows up on the very next bill.

## 4. Track power factor where it is billed

Power factor is measured at the point the utility bills it, so track it there,
at the incomer, on the same interval as the demand.

What the data tells you that a single monthly figure cannot:

- Whether the power factor is poor **all the time** — which suggests fixed
  compensation — or **only at part load**, which needs an
  [APFC panel with the right step size](/blog/apfc-panel-step-sizing).
- Whether an existing capacitor bank is actually switching. A step that has
  failed shows up as a power factor that stopped responding to load.
- Whether you are going **leading** at night, which several tariffs penalise as
  firmly as lagging.

Convert what you see into a target and a capacitor rating with the [power factor
calculator](/tools/power-conversion), and check the money against [what the
penalty costs](/blog/power-factor-penalty-what-it-costs).

## Alerts that people do not switch off

Fixed thresholds fail in both directions. "Alert if kW > 400" fires every day
during normal busy periods and stays silent through a quiet week when something
is badly wrong. Within a month everyone ignores the emails.

Alert on **deviation from expected** instead:

| Alert on | Not on |
|---|---|
| Consumption above the baseline for this production level | A fixed kW threshold |
| Base load higher than last week's minimum | Total daily kWh |
| Demand projected to exceed contract this interval | Demand exceeded (too late) |
| Power factor below target for 3 consecutive intervals | A single low reading |
| Phase currents diverging by more than 15 % | Any single current value |
| **No data received** for 30 minutes | Nothing — silence is invisible |

That last one matters more than the rest combined. A dead gateway produces a
flat line at zero, which looks like a plant that stopped. Alert on the absence of
data, not just on its content — the mechanism is in [building a
gateway](/blog/building-a-modbus-energy-gateway).

And set a budget: **if an alert fires more than about once a week without
producing an action, it is misconfigured.** Fix it or delete it. An alert nobody
acts on trains everyone to ignore the ones that matter.

## Reconcile, or trust nothing

Add up your submeters and compare against the main meter. The difference is
unaccounted energy, and it should be small and stable.

- **A large gap** means an unmetered load, or a CT ratio entered wrongly in the
  configuration. Both are worth finding.
- **A gap that grows** means a meter is drifting, or a
  [CT is saturating](/blog/ct-ratio-and-burden-for-meters) as load rises.
- **A negative gap** — submeters totalling more than the main — is almost always
  a wrong CT ratio or a double-counted feeder.

Do this monthly. It is the only check that tells you whether any of the rest of
your analysis is built on real numbers.

## Make it somebody's job

The analyses above take an hour a week and they are worth more than the
monitoring system cost. But they only happen if:

- **One named person** reviews the data on a fixed schedule.
- The review produces **actions with owners**, not observations.
- Savings are **verified against the baseline** afterwards, so the next
  investment is easier to justify.
- The baseline is **re-fitted** when the plant genuinely changes — new equipment,
  a new product mix — and the old one is kept for comparison.

A monitoring system without that hour is a data logger. The meters, the
[bus](/blog/rs485-wiring-for-modbus) and the
[gateway](/blog/building-a-modbus-energy-gateway) are the easy part; they are
also the part that gets all the attention, and the part that saves nothing on
its own.