Utility Data Quality: The Hidden Blocker in Energy and GHG Programs

Most energy and greenhouse gas programs do not fail because the analysis is wrong. They fail because the data feeding the analysis is quietly broken. A meter reads estimated for three months, a gas bill in cubic metres sitting next to one in gigajoules, a building with two accounts where one was never mapped, a billing period that overlaps the one before it. None of these throw an error. They just produce numbers that are confidently incorrect.

This matters more as reporting gets stricter. The GHG Protocol treats the quantity of purchased energy as the core activity data for Scope 2, usually measured through a meter but sometimes estimated and taken from a supplier bill, per the Scope 2 Guidance. If that meter data is incomplete or in the wrong units, the emissions figure inherits the error. The protocol is explicit that inventories should be systematically neither over nor under actual emissions and that uncertainty should be reduced as far as practicable, a point developed in its uncertainty guidance.

The reason data quality stays hidden is that it rarely announces itself. A spreadsheet total still calculates. A dashboard still renders. The gap only surfaces when someone reconciles against the utility, an auditor asks how a number was derived, or a year-over-year comparison moves in a direction the building's operations do not explain.

The five failure modes that do the most damage

Utility data breaks in predictable ways. Naming them makes them testable.

1. Missing periods and gaps

A missing month is the most common defect and the easiest to miss, because a chart simply draws a line across the gap. Meter data management systems routinely detect and estimate missing daily and interval reads, which tells you gaps are normal enough that whole systems exist to handle them. The problem is when they go undetected and get treated as real zeros or averaged away.

2. Estimated reads treated as measured

Utilities issue estimated bills when they cannot read a meter. The GHG Protocol data quality hierarchy ranks measured periodic data above estimated engineering data, and estimates carry more uncertainty by design. If your dataset does not flag which reads were estimated, you cannot weight them, correct them when the true-up bill arrives, or explain them in an audit.

3. Unit mismatches

Natural gas arrives in cubic metres or gigajoules, electricity in kilowatt-hours or megawatt-hours, water in cubic metres or kilolitres, steam in pounds or kilograms. Mixing units inside one series is silent and severe. A single gigajoule value dropped into a cubic-metre column can distort a building's whole gas profile.

4. Broken meter mapping

A building is not a meter. Sites have multiple accounts, accounts change numbers, meters get swapped, and tenants come and go. When the mapping from meter to account to building is wrong, consumption lands on the wrong asset. Benchmarking and intensity metrics then compare buildings that are not what the label says they are.

5. Overlapping or misaligned periods

Billing periods do not line up with calendar months, and they sometimes overlap or leave gaps at the boundaries. Summing raw bills into a calendar year without aligning periods double counts some days and drops others. The error is usually small per bill and compounds across a portfolio.

What a validation layer actually checks

A validation layer is a set of rules that runs on every record as it enters the system, before it reaches a report. The goal is not to hide bad data but to flag it, so a human decides what to do with a known problem instead of discovering it later. Tools in this space already exist for a reason. ENERGY STAR Portfolio Manager, for example, ships a data quality checker that some benchmarking programs require you to run before submission.

• Completeness. Are all expected periods present for each meter, with gaps flagged rather than filled silently?
• Read quality. Is each read marked measured or estimated, and are estimates reconciled when the true-up arrives?
• Unit consistency. Does every value in a series use the same unit, with conversions applied and logged?
• Meter and account mapping. Does each meter map to the correct account and building, with changes tracked over time?
• Period integrity. Do billing periods align without overlaps or gaps when rolled to months or years?
• Outlier and continuity checks. Does a value swing in a way the building's operations do not explain?

How MartinAI helps

MartinAI reads utility data across every commodity and layout, from Green Button ESPI XML to PDF bills and spreadsheets, and runs validation on each record as part of the same pass. It checks for missing periods, distinguishes estimated reads from measured ones, normalizes units, resolves meter-to-account-to-building mapping, and aligns billing periods so monthly and annual rollups are consistent. Problems are surfaced with the record attached, so your team can act on a specific meter rather than hunting through a spreadsheet.

The output is a clean, standardized dataset that carries its own quality signals. That is what makes downstream work defensible: a Scope 2 figure you can trace to measured reads in known units, a benchmark that compares the right meters to the right building, and an audit trail that shows where an estimate was used and when it was corrected. It also lines up with disciplined energy management, the same discipline described in iso-50001-energy-management-without-spreadsheets.

Data quality is unglamorous, which is exactly why it stays unaddressed until a number gets challenged. Building the validation layer once, and running it on every record, moves the problem from something you discover late to something you catch early. That is the difference between a program that survives an audit and one that spends the audit reconstructing where its numbers came from.