Energy Budget Forecasting

Energy is one of the larger controllable line items in most operating budgets and one of the least rigorously forecast. The typical method is to take last year's total and apply a growth rate. This produces a number that is defensible in a budget meeting and useless afterwards, because when the actual comes in 14% high nobody can say which of the four independent things that drive the cost was responsible.

The fix is not sophistication. It is separation. Energy cost is the product of four drivers that move for unrelated reasons and on unrelated schedules, and a forecast that models them separately is both more accurate and — more importantly — diagnosable when it misses.

Why Energy Forecasts Miss

Three structural reasons, in order of how much damage they do:

• Blending the drivers. A single escalation percentage applied to total spend implicitly assumes volume, commodity price, delivery tariff and peak charges all move together. They do not. In a year when the commodity price falls and the delivery tariff rises, a blended forecast is wrong in both directions simultaneously and reports itself as roughly right.
• Un-normalized weather. Last year's consumption embeds last year's weather. A mild winter becomes next year's baseline, and the variance that follows gets attributed to price or to operations rather than to the fact that the baseline was never a normal year.
• Assuming the delivery side is flat. Buyers watch the commodity market because that is what gets quoted. Meanwhile the regulated delivery charges, riders, capacity obligations and transmission allocations move on a rate-case calendar nobody is tracking. In many markets these have been the fastest-growing part of the bill, and they are the part a supply contract does not protect you from.

The Four Components, Forecast Separately

1. Contracted Volume at Contracted Rates

The easy part, and the part to lock down first. For every account, you know the contract rate, the term, the expiration date and the approximate volume. This block should be forecast with near-certainty — the only uncertainties are consumption variance within the bandwidth and the expiration date falling mid-budget-year.

Build the schedule at the account level with expiration dates visible. That schedule is simultaneously your budget input and your procurement calendar, which is the point at which forecasting starts producing decisions rather than just numbers.

2. Uncontracted Volume Against the Forward Curve

Any volume not under contract during the budget period is exposure. Price it against the forward curve for the relevant delivery period and hub — not against today's spot price, and not against last year's average, both of which are common and both of which are wrong.

Carry this block with an explicit range rather than a point estimate. The forward curve is a market expectation, not a forecast, and presenting it as a single number invites a false precision that will be held against you. A range with a stated basis is more useful to a CFO than a point estimate with none.

3. Delivery, Tariff and Riders

Model these from the utility's filed rates rather than from your own history. Specifically: the current rate schedule, any rate case filed or pending with an expected effective date, and the rider schedule. Utilities publish this. Pending rate cases are public dockets with timelines, and a rate case decided in March affects nine months of your budget year.

This is also where a forecast exercise reliably finds money that has nothing to do with forecasting — because building the model requires confirming which rate schedule each account is on, and that is the moment mis-classification surfaces. See utility tariff optimization.

4. Capacity, Transmission and Peak-Driven Charges

The most forecastable component of all, and the most frequently ignored. In capacity markets your obligation for the coming delivery year is set by peak tags that have already been measured. You are not predicting these; you are looking them up. The same is true of transmission allocations in markets that use a coincident-peak method.

Two consequences. First, this block can be forecast with high confidence a year ahead. Second, it makes the value of peak management legible to finance: the tag set last August is a line in next year's budget, so a curtailment program has a quantified budget impact rather than a hypothetical one. See peak load management.

Weather Normalization, Briefly

You do not need a degree-day regression to do this usefully, though it helps. The minimum viable version: pull three to five years of monthly consumption, pull heating and cooling degree days for the same months from the nearest weather station, and establish roughly how many kWh or therms each degree day costs you. Then forecast against normal degree days rather than last year's.

The point is not precision. It is that the budget stops silently assuming last year's weather repeats, and that weather-driven variance can be separated from everything else at year end.

The Accrual, and Why It Needs Validating

Utility invoices lag consumption, and meter read dates do not align with the accounting calendar. So every period closes with an accrual — an estimate of energy consumed but not yet billed. Three things make accruals systematically wrong:

• Flat monthly estimates on seasonal load. An accrual built from an annual average overstates spring and understates summer, every year, predictably.
• Estimated meter reads. When the utility estimates a read and trues it up later, the correction lands in a subsequent period. Without tracking, this reads as a price variance that never happened.
• No feedback loop. The accrual is booked, the invoice arrives, the difference posts to the current period, and nobody records that the estimate was off or by how much.

Accrual validation is the discipline of reconciling each prior accrual to the invoice that eventually arrives, recording the error, and using the error history to improve the estimate. It is unglamorous and it is the difference between an energy line finance trusts and one they add a contingency to.

Variance Analysis That Tells You Something

At period end, decompose the variance into four buckets that map to the four drivers:

• Volume variance — we used more or less than planned.
• Weather variance — of that volume difference, how much is explained by degree days.
• Price variance — the rate differed from plan, on the uncontracted block.
• Rate and tariff variance — delivery charges, riders or peak-driven charges differed from plan.

Each bucket points at a different owner and a different action. Volume variance is an operations conversation. Weather variance is nobody's fault and should be excluded from performance assessment. Price variance is a procurement conversation about timing and hedging policy. Tariff variance is a regulatory-tracking conversation, and usually the one nobody owns.

Cadence, Ownership and Tools

A workable rhythm is a monthly accrual and reconciliation, a quarterly reforecast of the open position against the current curve, and an annual rebuild that refreshes the tariff assumptions and the contract schedule. One person owns it, with access to interval data, the contracts and the tariffs — all three, or the forecast is a guess in a spreadsheet.

On tools: a spreadsheet is genuinely adequate up to roughly twenty accounts if the structure separates the four drivers. Above that, the constraint is not modeling but data collection — gathering, validating and normalizing invoices across dozens of accounts and several utilities is where the hours go, and it is what platforms are actually worth paying for. Buy for the data pipeline, not for the charts.

What Good Looks Like

A defensible energy budget is one where you can answer, in a single meeting: which accounts are contracted and until when, what the open position is and what it is priced against, what the delivery-side assumptions are and where they came from, what the capacity and transmission tags are for the coming year, and what last year's variance decomposed into. That is a short list, and almost no organization can produce it on request. The ones that can spend measurably less, mostly because the exercise surfaces the decisions early enough to act on them.