How Many Months of Consumption Data Does a Credible Contract Decision Need? Why Annualising a Few Months Misleads

Every broker has done it at some point. A new client sends over four bills, the spreadsheet multiplies the total by three, and a supposedly annual figure goes out to suppliers on a tender. It feels reasonable. It is also one of the most reliable ways to put a business on the wrong contract, with the wrong volume assumption, at a price that will not survive contact with a real winter.

The problem is not arithmetic. UK energy consumption is not evenly spread across the year, and the pattern is driven by weather that varies from one year to the next. Scaling a short window up to twelve months assumes a flat profile that almost no site has.

What annualising actually does to the numbers

Take a small manufacturer with gas-fired space heating and a modest process load. Suppose it uses 34,000 kWh across January, February and March. Multiply by four and you get 136,000 kWh. Now look at the real shape: heating runs hard in those months and barely at all from May to September. The genuine annual figure might be nearer 85,000 kWh, so the tender has overstated consumption by around 60%.

Run the same exercise on June, July and August data and you get the mirror image. Summer gas sits close to the site's non-heating baseload, so annualising it can understate the real figure by half. Either error travels straight through to the contract. The unit rate a supplier quotes is built on the volume it expects to deliver, the standing charge band for gas is set by the Annual Quantity, and the supplier's hedge is bought against your forecast shape.

When the forecast is wrong, the consequences are not theoretical. Underestimating can trigger a mid-term price review or a security deposit request. Overestimating on a flexible or take-or-pay arrangement leaves the client paying for volume it never used.

Gas is the worst offender

For most UK non-domestic sites with wet heating systems, gas demand is dominated by outside air temperature. Space heating typically accounts for the large majority of annual gas at an office, school, care home or warehouse, so consumption in a cold January can be three or four times a July figure at the same premises.

The industry knows this, which is why the gas settlement regime never uses raw scaling. A site's Annual Quantity is calculated from consumption between two meter readings taken between nine and 36 months apart, adjusted for the weather actually experienced over that period, then converted to a twelve month estimate (Xoserve, Annual Quantity). The adjustment uses the Composite Weather Variable, a regional measure combining temperature with wind and seasonal effects, so that a mild reading period is not mistaken for a low-consuming site.

Two points follow for brokers. The nine month minimum in the AQ rules is a useful signal: the industry itself does not treat a shorter read period as adequate for a weather-corrected annual estimate. AQ also lags reality. Xoserve's demand estimation work has noted that it can take up to twelve months for a genuine change in a consumer's consumption to be fully reflected in AQ (Xoserve demand estimation update). If your client has changed its heating controls, extended its opening hours or closed a building, the AQ on the portal may still be describing the old business.

Electricity is milder, but it is not flat

Non-half-hourly electricity is settled using profile classes, with consumption allocated across the year by profile coefficients applied to an Estimated Annual Consumption or an Annualised Advance (Elexon, Profiling). Those coefficients exist precisely because a shop's December day and its June day look nothing alike.

Lighting and heating push winter electricity up for many SMEs; cooling and refrigeration push summer up for others. A convenience store with open-fronted chillers can peak in August. A rural workshop on electric heating peaks in January. Annualise the wrong quarter for either and you will misprice the contract. On a half-hourly site you will also misjudge the capacity position, which feeds distribution charges and the risk of excess capacity penalties from the DNO.

Half-hourly data helps, but only if you have enough of it. Twelve months of HH data tells you the load factor, the peak periods and the weekend pattern. Three months tells you what the last three months were like.

Separating baseload from weather-sensitive load

The useful analytical move is to split consumption in two: the load that runs regardless of weather, and the load that responds to temperature. Plot monthly kWh against heating degree days for the client's region and you get a straight line whose intercept is roughly the baseload and whose slope is the weather sensitivity. Degree day series are published regionally and rest on Met Office temperature records (Met Office UK and regional series).

With that split, six months of data becomes far more informative. You can extend the fitted line across a normal year rather than multiplying a season by a constant. It remains an estimate, and it still needs sanity checks against operating hours and building fabric, but it is defensible in a way that a blunt multiplier is not.

Be careful about which year you call normal. Recent years have run warm by historical standards, so estimates built on a mild winter will understate a cold one. Ofgem's review of Typical Domestic Consumption Values showed how far underlying usage can move, with the median gas value cut by around 17% and electricity by around 7% (Ofgem, Review of typical domestic consumption values). Those are domestic figures, but the direction of travel applies to small commercial sites too, and it is a reminder that historic data ages.

So how many months do you actually need

A working standard that stands up to scrutiny:

• Twelve months of actual meter readings or HH data is the minimum for a firm price on a term contract. It captures one full heating and cooling cycle and matches the period suppliers are pricing.
• Twenty-four months is materially better, because it lets you see whether last year was typical and separates weather from business change.
• Nine to eleven months is workable if weather corrected, the same threshold the gas AQ rules apply.
• Fewer than nine months should be treated as an estimate, labelled as such on the tender, and paired with a plan for what happens if the volume is wrong.

For genuinely new sites with no history, use published benchmarks by building type and floor area, then cross-check against equipment ratings and occupancy hours. Tell the supplier the figure is an estimate. Suppliers price uncertainty either way, and a disclosed estimate is cheaper than a discovered error.

One caution on relying on EAC or AQ alone: both are themselves estimates derived from past readings, and both can be stale or distorted by a long run of estimated reads. Check the date of the last actual reading before you trust either number.

What a broker can do this week

• Ask for 24 months of reads or HH data as standard on every enquiry, and record the date of the last actual read for each meter.
• Before annualising anything, plot monthly kWh against regional degree days and note the intercept. A steep line means the site is weather sensitive and short-window data is unsafe.
• For gas, pull the current AQ and compare it with your own weather-corrected estimate. A gap of more than about 10% is worth an AQ correction discussion with the supplier, since AQ drives transportation and standing charge bands.
• On any tender built from fewer than nine months of data, state the assumption in writing to both client and supplier, and check the contract's volume tolerance and mid-term review clauses before signature.
• Diarise a review three months into supply, once fresh actual reads exist, so a bad assumption is caught early rather than at renewal.
• For half-hourly sites, check agreed capacity against the twelve month peak at the same time, as capacity and consumption errors usually arrive together (DESNZ Energy Trends is a useful cross-check on sector level demand movements).

Data discipline costs a few emails. Getting the annual figure wrong costs a client real money for the length of the contract.