---
title: "Choosing an energy meter for IoT: what actually matters"
description: "Accuracy class, register map, protocol and documentation — the four things that decide whether a meter becomes a data source or an expensive display."
date: "2026-06-12"
author: "Divakar B"
source: "https://energycalchq.com/blog/choosing-an-energy-meter-for-iot"
license: "© 2026 Divakar B. Quote with attribution to https://energycalchq.com/blog/choosing-an-energy-meter-for-iot"
---

Energy meters are sold on accuracy class and price. Neither predicts whether the
meter will be any use in a monitoring system. The meters that cause months of
trouble are usually accurate, cheap and completely correct — they just cannot
tell you anything, or they will not tell you in a way you can read.

Four things decide it, and only one of them appears prominently in the catalogue.

## 1. How the data gets out

![Comparison of pulse output, basic Modbus and a full register map](/blog/meter-output-options.svg "The difference is not accuracy. It is how many questions you can ask.")

| Output | Gives you | Good for |
|---|---|---|
| Pulse (SO) | Energy only, as counts | Cheapest retrofit, sub-billing where nothing else matters |
| Modbus RTU over RS485 | Everything in the register map | The default for industrial metering. Cheap, robust, long runs |
| Modbus TCP | The same map over Ethernet | Where structured cabling already exists |
| MQTT-native | Publishes on its own | Convenient, but you are tied to the vendor's schema and cloud |
| M-Bus | Common on European water and heat meters | Rare on Indian electrical meters |

**Pulse outputs are blind.** A pulse tells you a quantum of energy passed. It
cannot tell you the voltage sagged, the power factor collapsed, or one phase is
missing. If the only question you will ever ask is "how many units", pulse is
fine and cheap. Every diagnostic question needs a register map.

**Modbus RTU is still the right default** for most Indian installations. It runs
over two wires for a kilometre, costs nothing per node, and every gateway on
earth speaks it. Its limitations — no self-description, no discovery, polling
only — are real but manageable.

Be wary of meters that only talk to the manufacturer's own cloud. It works
beautifully in the demonstration and it means your data lives somewhere you do
not control, in a schema you cannot change, for as long as that vendor supports
that product.

## 2. The register map, and whether you can get it

**This is the single biggest practical differentiator, and it is never in the
datasheet.**

Before buying, ask the supplier for the Modbus register map as a document. What
you learn from the answer:

- **It arrives within a day, as a proper PDF with addresses, data types, scaling
  and byte order.** Good sign. This vendor has done this before.
- **It arrives as a photograph of a page from a manual.** Workable, but expect
  ambiguity.
- **"Our engineer will configure it at site."** Walk away. You are buying a
  dependency, not a meter.
- **No answer.** The map may not exist in written form at all.

Read the map before ordering and check for these specifically:

| Check | Why |
|---|---|
| Addressing convention stated | 40001-style or protocol address? An off-by-one here wastes a day — see [the address converter](/tools/modbus-address) |
| Data type per register | 32-bit float, 32-bit integer with a scale factor, or 16-bit? |
| **Byte and word order documented** | The single most common cause of nonsense readings |
| Function code | FC03 holding registers or FC04 input registers — meters differ |
| Which registers are *not* implemented | Gaps in a map return garbage rather than an error |
| Energy register width | A 16-bit kWh register wraps at 65,535 and you must handle rollover |

That last one catches people months in. If the map does not say, read the
register on a known load and watch what it does.

## 3. Accuracy class — and where it actually applies

| Class | Error | Use |
|---|---|---|
| 0.2S | ±0.2 % | Revenue metering at HT |
| 0.5S | ±0.5 % | Sub-billing, tenant metering, energy allocation |
| 1.0 | ±1.0 % | Monitoring, trending, fault-finding |
| 2.0 | ±2.0 % | Indication only |

For monitoring — finding waste, spotting a failing motor, comparing shifts —
class 1.0 is entirely adequate. For anything where money changes hands between
parties, use 0.5S and be prepared to defend it.

The class only describes the **meter**. The system's accuracy is the meter *and*
the CT, and the CT is usually the weaker link — particularly at part load, where
a plain class 0.5 CT is outside its specified range entirely. That argument is in
[CT ratio and burden](/blog/ct-ratio-and-burden-for-meters), and it is worth
reading before spending money on a better meter class.

Check the class is stated for the **class of connection you have**: some meters
are class 1.0 for active energy but class 2.0 for reactive, and reactive is what
your power factor calculation depends on.

## 4. The registers you will want later

Buying the minimum map is a decision you cannot revisit without replacing
hardware. Insist on:

- **Per-phase voltage, current, power and power factor** — not just totals.
  Diagnosis is almost always about the *difference* between phases.
- **Import and export energy separately.** Essential the day solar is added, and
  useful immediately for spotting a reversed CT.
- **kVA and kVAr**, so you can track power factor against the [tariff
  penalty](/blog/power-factor-penalty-what-it-costs).
- **Maximum demand**, with the integration period the DISCOM uses. This is the
  register that reconciles your system against the bill.
- **THD**, if there are any VFDs on site. It is the early warning for
  [harmonic problems](/blog/apfc-panel-step-sizing).
- **Frequency and voltage unbalance** — cheap to record, and the fastest way to
  prove a supply-side problem is not yours.

## Direct connected or CT operated?

Meters up to about 63 A connect directly. Above that, CTs.

Direct connection is simpler, cheaper and has one fewer error source, so use it
where the load allows. But a direct-connected meter must be able to carry the
full load current through it — which means the meter is in the power path and
replacing it means an outage. A CT-operated meter can be swapped live, with the
CT secondaries shorted first.

For retrofits, split-core CTs avoid disconnecting anything. They are slightly
less accurate and they must be properly closed and clean on the mating faces.

## Practical checks before you commit

1. **Ask for the register map.** Do not order without it.
2. **Buy one and bench it.** Wire a single meter, poll it, read voltage, and
   confirm the byte order matches the document. An afternoon here saves a
   fortnight of site debugging — [the float decoder](/tools/modbus-decoder)
   makes it a two-minute job.
3. **Check the baud rates and unit ID range.** A meter that only supports 9600
   and unit IDs 1–31 constrains your bus design.
4. **Confirm the response time.** Some meters take 200 ms to answer, which caps
   how many you can poll on one bus at a useful interval.
5. **Check what it does at power-up.** Does it restore its configured unit ID
   and baud, or default back? Meters that reset their address on power failure
   will find you eventually.
6. **Check the energy register survives power loss.** It should be in
   non-volatile memory. Verify by switching it off.

## The mistake worth avoiding

Buying twenty meters before commissioning one. Every issue in this post is
cheap to discover on a single unit on a bench and expensive to discover on
twenty units in an occupied building.

Buy one, poll it, read the register map against reality, then order the rest. If
the meter passes that afternoon, the rest of the project is
[bus wiring](/blog/rs485-wiring-for-modbus) and software — both of which are
problems you can solve.