One Bill Isn't the Whole Story. Here's How We Estimate Your Annual Usage

Energy comparison accuracy is something nobody talks about honestly. Comparison sites show you a saving figure and present it as fact. They don't explain what assumptions produced it, how variable that figure might be, or in what circumstances their methodology fails.

We're going to explain ours. Including where it's less accurate and why.

What Every Comparison Tool Assumes (And Doesn't Tell You)

Standard comparison tools, every major Australian energy comparison site, compare plans using one of two approaches.

Approach one: You enter your usage The tool asks for your average quarterly kWh consumption or annual usage. You either know this or you guess. Most people guess, because most people don't track their kWh usage. The comparison is only as accurate as the number you enter.

Approach two: Postcode with average assumption The tool takes your postcode and applies a national or regional average consumption figure. The most common figure used in Australian comparison tools is 3,900 kWh per year for a single-rate household. If your household uses 7,500 kWh per year (common with ducted reverse-cycle air conditioning), the comparison is wildly underestimating your usage. If you use 2,200 kWh (small apartment, minimal heating and cooling), it's overestimating.

Neither approach uses your actual bill. Why that gap changes the answer, not just the precision, is the subject of postcode vs bill.

What We Do Differently

We read your bill. With your permission, you upload the PDF. Our engine extracts the actual data from the document.

Here's what we extract and use:

This data set allows us to establish your actual current plan cost for the billing period. Not an estimate, but a calculation from real numbers.

The Hard Part: Annual Estimation From a Single Bill

Here's where we need to be honest about limitations.

A single bill covers one billing period, typically 60–92 days. That period may be your peak summer consumption quarter, your quietest shoulder quarter, or anything in between. Multiplying one quarter's consumption by four and calling it your annual usage is wrong for most households. It's most wrong for households with significant seasonal variation.

Our approach is seasonal extrapolation. We take your billing period consumption and apply seasonal adjustment factors to estimate what your usage looks like across a full year. Summer vs winter electricity bills shows how far apart the two ends of that curve sit.

How the seasonal adjustment works: We know, from real bill data across Australian climate zones, approximately what proportion of annual usage falls in each quarter for different household types. A Queensland household with ducted reverse-cycle in a January billing period is likely to be near their annual consumption peak. We apply an adjustment that scales their January usage down to reflect that it's a high-consumption outlier in the annual pattern, not a representative average.

Where this is accurate: The seasonal extrapolation is most accurate for:

• Households with significant space heating/cooling (the seasonal curve is well-defined)
• Bills that are clearly seasonal peak or trough (the adjustment is most important and most reliable at extremes)
• States with clear seasonal energy demand patterns (most of eastern Australia)

Where this is less accurate: The seasonal extrapolation is least accurate for:

• Households with very flat year-round usage (small apartments with no heating/cooling)
• Households in mild climates where seasonal variation is minimal
• Bills from shoulder seasons (spring/autumn) where the adjustment factor is smaller and the baseline is more representative anyway

What we show you instead of a percentage: Above the plan list, on every comparison screen, one panel names what the figure is made of. Not a grade out of ten and not a plus-or-minus band. It sits on one of four rungs:

• A rough guess. A postcode and nothing else. Useful for a feel of the market, and it is not your number.
• A better guess. A postcode plus your answers about the house, the appliances and the people in it. Still an estimate, because nothing here was read off a meter.
• Measured, short of a full year. Your own usage, so the shape of your year is yours rather than an assumption, but it stops short of twelve months.
• Measured, not estimated. A full year of your bills. Your year as it was billed, with nothing scaled up from a shorter period.

This used to be a table of percentages: plus or minus 5%, 10%, 20%, 25%, matched to bill types. We took it down, and the reason is worth stating because it is the same reason this guide exists.

We printed a band we had not measured. The plus-or-minus 25% figure was a reasonable-sounding estimate of our own uncertainty, not a back-tested one. When we did back-test it, against eight real gas bills, the actual error ran from 83% under to 149% over, and six of the eight fell outside the band we were printing. A confidence range that is itself a guess is worse than no range at all, because it borrows the authority of a measurement it never had.

So the rule now, in the code as well as on this page: a numeric band may be printed only for a path where a back-test has measured that path. None has yet been measured to a band we would publish. Until one is, the rungs above describe the evidence in words, and we would rather tell you your figure came off one bill and a typical seasonal curve than dress that up as plus or minus 12%.

The rung describes the baseline: what this household paid over the period. Every plan price beside it is a projection of what a different plan would have cost on that usage, and that stays a projection even at the top rung. "Measured, not estimated" means your year as it was billed. It does not mean the saving is exact.

How Our Accuracy Compares

A postcode-based comparison using a 3,900 kWh annual assumption for a household that actually uses 7,500 kWh produces a saving figure that can be off by hundreds of dollars, in either direction. The error isn't bounded or disclosed. The comparison just shows you a number and implies it's correct.

Our seasonal extrapolation, with the evidence behind each figure named, is more honest than an undisclosed assumption-based comparison, and it starts from your own meter reads rather than from a national average. The same scrutiny applied to the industry as a whole is in are energy comparison sites accurate.

The only more accurate comparison is one with your full 12-month bill history or smart meter data. If you have access to your smart meter data through your retailer's app or portal, uploading that data alongside your bill gives us a better annual consumption estimate. More bills = better accuracy.

What We Don't Do

We sort by the cheapest for your household, and we don't recommend. A plan that's technically cheapest based on usage assumptions but requires a direct debit you can't reliably maintain, or has a tariff structure that penalises how you actually live, may not be the right plan for you. The order is arithmetic. We show you the options, we show you the conditions attached to each one, and we let you choose.

We don't claim precision we don't have. Every comparison screen carries the rung its baseline sits on, above the plans rather than in a footnote, and every plan on it repeats the usage figure it was priced on and whether that figure was read off a bill or assumed from a postcode. We'd rather tell you the figure came off one bill and a typical seasonal curve than print a plus-or-minus we have not measured.

We don't show plans we can't verify. Every plan in our comparison comes from retailers' published plan data, the Product Reference Data they lodge under the Consumer Data Right. It is the same public data that powers Energy Made Easy in the jurisdictions Energy Made Easy covers; Victoria is not one of them, and runs Victorian Energy Compare instead, though Victorian retailers publish to the CDR the same way. We use the public plan data only and we are not an accredited data recipient. If a plan isn't published there, we don't show it, so what you see is limited to plans we can verify are real, current and generally available at your address. A retention offer made to you over the phone, a door-to-door deal or embedded-network pricing is not published anywhere we can read.

The Bottom Line on Accuracy

A comparison built from your actual bill, seasonally adjusted, with the evidence behind it named, and priced against retailers' published plan data.

It is not perfect. No single-bill comparison can be. But it is dramatically more useful than a postcode entry producing a plan list based on a usage figure that has nothing to do with your household.

Upload your bill. Read the rung we show you, because it tells you what your figure is made of. Understand that the number is an estimate. And then decide whether the saving is worth ten minutes of your time to switch.

For most households, it is.

Which Bills We Calibrated Against, and Why That Isn't Your Bill

There is an obvious question buried in the paragraphs above, and a reader who has also read our privacy policy will already have it. We say the engine was calibrated against real Australian bills. We also say we delete your bill the moment we have finished reading it and keep no copy. Both are true, and leaving you to reconcile them yourself would be exactly the kind of gap this guide is supposed to close. So here it is.

The calibration corpus is two things, and neither of them is a customer's uploaded bill.

Consented real bills, tested during development. Our own household bills first, then bills that family, friends and a handful of small businesses gave us specifically so we could test the engine on them, knowing that was what they were for. That consent was asked for and given for that purpose. It is a development corpus, held separately, and it is not fed by the live service.

Compiled sample bills. Retailers publish their own "how to read your bill" pages, and those pages carry annotated specimen bills showing exactly where each retailer puts the usage block, the rate table, the supply charge and the meter reads. Those are published documents about a fictional account. A good part of the work of teaching a parser a new retailer format is done from them.

What is not in it: the bill you upload. The document itself is deleted as soon as the read finishes, and it is never sent to an outside company. Part of reading it may be done by a model we run on our own machine in Australia, which our privacy policy describes in full. Nothing from your bill leaves that machine while it is being read, and it is not used to train anybody's model. The program that runs the model does check with its supplier for a newer version when the machine starts up, which is before any bill is read and carries no part of one. It does not join the development corpus and no page of it is kept for us to look at later.

One honest qualification, because the privacy policy makes it and this page should not be softer than the policy it points at. We do keep a de-identified record of each comparison: the postcode, the network, the tariff, the usage and rates, what you were paying and what the cheapest plan we could price would have cost. No name, no address number, no account or meter number. That record is what tells us whether the comparison is landing where it should, so it does inform accuracy work in aggregate. What it cannot do is put your bill in front of anyone, because your bill no longer exists on our side by the time the record is written. Section 10 of the privacy policy sets out exactly what that record holds and what it does not.

The seasonal extrapolation was built the same way: from real bill data across Australian climate zones, in the development corpus, and from published seasonal demand patterns.