The Weather Demand Index
Weather is the largest demand driver that almost no analytics platform measures. It moves roughly a third of economic activity and influences 62% of what consumers buy (US and industrialised-economy estimates), yet most operators still treat a cold snap or a heatwave as noise in the numbers. This study quantifies how much demand moves, by event, by vertical and by Australian region, and shows the baseline math to forecast it.
Peak-to-trough swing in search demand for "heating system repair" between its summer floor and its winter peak. The same household problem, the same service, an almost sixfold change in how many people go looking for it, driven entirely by temperature. In Australia that peak lands in winter (June to August), not January. (WebFX, Seasonal Search Shifts in Home Services Demand, 2025; US data. AU seasonal timing inverted: BOM heating degree-days peak Jun-Aug.)
Why weather is a demand signal, not noise
Most demand models start with internal history: last year's sales, last month's leads, a trend line nudged by intuition. They treat the weather as an unmodelled shock, the thing that explains the variance you could not predict. That is backwards. Weather is one of the most measurable, most forecastable exogenous drivers a business has, and for whole categories it is the dominant one.
The macro number is striking. Weather conditions are estimated to affect as much as 35% of GDP in industrialised economies, and at the retail level influence roughly 2.3% of total output while materially moving store traffic and basket composition (ToolsGroup, Using Weather and Climate Data to Improve Demand Forecasting, US/global data; retail-weather literature, Springer, Applied Spatial Analysis, 2021, DOI 10.1007/s12061-021-09397-0). At the level of the individual decision, an estimated 62% of consumer purchasing decisions are weather-influenced, shaping what, where, when and in what quantity people buy (attributed to The Weather Company / IBM and the National Retail Federation, US data). Australia's climate makes the effect, if anything, sharper: the Bureau of Meteorology records far higher cooling degree-days across the north and a pronounced winter heating load in the southern and alpine states, so the same business sees two demand mountains half a year apart.
The point of this study is not the macro headline. It is the operating reality underneath it: the lift is wildly uneven. A heatwave does almost nothing to fashion and nearly doubles HVAC. A hailstorm is a same-week emergency for roofers and irrelevant to a heating contractor until the next quarter. If you manage demand with one seasonal curve, you are averaging across signals that move in opposite directions.
Two things matter in that chart. First, the magnitude: these are not 10% or 20% seasonal wiggles, they are multiples. Second, the timing, read on the Australian calendar: the frozen-pipe and heating peaks land in winter (June to August), the emergency-cooling peak lands in summer (December to February), and the roofing storm peak clusters in the spring-summer storm season (September to February). A single business that sells heating and cooling has two demand mountains half a year apart, and a budget split that ignores this is structurally mistimed.
Demand lift by event, vertical and region
The core of the index is a matrix: for a given weather event, how much does demand move in a given vertical? The figures below combine published peak-lift ranges with the structural patterns L.E.K. Consulting documents in its Weather Index for Service Essentials (WISE) framework. WISE is a US method; we apply its structure to Australian regions (NSW, VIC, QLD, WA, SA) rather than treating its measured US numbers as Australian. They are directional, not a promise for any single business; the methodology block below states exactly how a real per-account figure is computed.
Region toggles are shown for structure. The same event produces a different lift by geography: in Australia a cold snap is a routine winter event in the southern temperate and alpine markets (Melbourne, Canberra, Hobart) and a rare, high-elasticity shock in the warm north (Brisbane, Darwin, Perth), where heating equipment is sparse and a single hard cold front can drive heating and plumbing demand far above the national blend. Cooling, by contrast, dominates the north year-round and spikes nationwide in the summer heatwaves of December to February.
The single most useful operating insight in the whole dataset is also the least intuitive: the lag structure differs by vertical, and a single lag value is wrong. L.E.K.'s WISE work shows that degree-days in a given quarter drive HVAC shipments in the following quarter (a one-quarter lag), while extreme weather drives roofing demand in the same quarter, then a larger decline two quarters later as the storm-damage pipeline is pulled forward and exhausted (L.E.K. Consulting, A WISE Approach to Weather-Related Services Demand, US framework applied to AU).
Which verticals are weather-sensitive, and by how much
Ranked by peak demand lift in extreme conditions, the order is stable across sources: home services and HVAC sit at the top, fashion at the bottom. This ranking is what should govern how aggressively a weather signal is allowed to move a budget or an inventory plan.
Sensitivity tells you how big the prize is. The next question is whether modelling it actually improves a forecast, and the answer is a measured yes, with honest limits.
The uplift is real but bounded, and the gap between grocery (+20.2%) and home improvement (+3.3%) is itself instructive. Categories with frequent, low-ticket, weather-immediate purchases gain the most; high-consideration, lower-frequency categories gain less because the weather signal competes with a longer, noisier buying process. This is why the index reports sensitivity and forecastability separately. A vertical can be highly sensitive (big swings) yet only modestly forecastable (the swing is hard to call far enough ahead to act on).
The baseline math: degree-days and lagged correlation
The credibility of any weather-demand claim rests on two textbook constructs: the degree-day, which turns temperature into a demand-relevant quantity, and lagged correlation, which finds the delay between a weather event and the demand it produces. Both are public, both are decades old, and both are what a defensible model is built on. Nothing here exposes a proprietary method; it is the standard approach, which is exactly the point.
1. Degree-days from an 18°C base
A degree-day measures how far, and for how long, temperature sat above or below a comfort baseline of 18°C (the metric degree-day base). It is the standard unit linking weather to heating and cooling demand, and the Bureau of Meteorology publishes Australian heating and cooling degree-day maps on exactly this 18°C base (Bureau of Meteorology, Heating and cooling degree days, base 18°C, ~350 stations, climatology 1961-1990).
HDD = max( 18 − Tmean, 0 )
where Tmean = ( Thigh + Tlow) / 2 (daily mean, °C)
Fifty-two cooling degree-days in five days is an extreme-heat window. Against a typical southern-city January baseline of roughly 6 to 7 CDD per day, this run is about 50% above normal load, the kind of signal that, in a high-sensitivity vertical, justifies pulling cooling budget and crews forward rather than waiting for the phone to ring.
2. Lagged correlation: finding the delay
Knowing it was hot is not enough; you need to know when the demand shows up. The defensible tool is the Pearson correlation coefficient computed between a weather feature and demand at a series of time lags, with autoregressive terms included so you are measuring the weather's contribution and not just demand's own momentum.
In peer-reviewed HVAC load modelling, Pearson coefficients confirm the structure: autoregressive load terms dominate (load in the prior hour correlates at 0.94, prior 24 hours at 0.74), while temperature enters as a strong exogenous driver, and indoor set-point correlates at −0.94 (Nature Scientific Reports, GRU/IASO AC load model, 2025, DOI 10.1038/s41598-025-87776-0). Retail studies use the same family of tools, multiple linear regression with autoregressive terms, and find snowfall in particular shows pronounced lagged effects on apparel demand (empirical brick-and-mortar retail study, ResearchGate 335173325).
3. The guardrail: control for climate, or you will over-claim
This is the single most important methodological caution, and the one most weather-demand pitches quietly skip. If you do not net out climate (the predictable baseline seasonality of a location) before measuring weather (the deviation from that baseline), categories get falsely flagged as weather-sensitive when they are merely seasonal. A study explicitly on this point shows that failing to control for climate inflates apparent weather sensitivity (Accounting for Climate When Determining the Impact of Weather on Retail Sales, ResearchGate 373853767). The honest model measures the residual: demand above or below what the season alone predicts.
How far ahead you can actually act
A demand signal is only useful if you can act on it before it lands. Forecast reliability decays sharply with horizon, and the index is honest about it: temperature forecasts are reliable about 7 to 10 days out, precipitation 3 to 5 days, and seasonal demand patterns 3 to 6 months ahead (WebFX; WeatherTrigger, 2026, US data; the 7-10 day horizon is consistent with BOM forecast skill). The practical rule that falls out of this is to launch demand capture 30 to 45 days before peak, which on the Australian calendar means cooling campaigns and inventory by October to November and heating by April to May.
From lift to dollars: a worked demand-to-budget step
Sensitivity and forecastability combine into a budget decision. Suppose an Australian air-conditioning business knows from its own history that cooling demand lifts roughly 80% at the peak of an extreme-heat window, and a credible 30-day seasonal forecast points to a hotter-than-normal summer stretch. The reallocation logic is mechanical.
The mistake this prevents is the most common one in seasonal trades: cutting spend when in-season CPA rises during a demand trough. The correct move is to recognise the trough as a demand shift between service types, not inefficiency, and to follow demand to where it actually is rather than scaling down. Weather tells you where it is going to be next.
Why this is a profit signal, not just a demand signal
Weather-timed demand is the highest-intent, lowest-cost demand a service business will see all year. A burst-pipe search during a July cold snap is a customer who needs the work now, compares less, and converts faster than the same customer in December. Capturing demand at its weather-driven peak compresses the cost and time to convert, which is the denominator of what we call Profit Velocity: the rate at which a business turns marketing and sales effort into durable contribution margin. (Profit Velocity is an owned Blufire metric, defined in full in the Contribution Margin Playbook.)
Get the timing wrong and the same demand costs more to win and is worth less when you win it. This is the case for treating weather as a managed input to the budget calendar rather than a post-hoc explanation for a variance you did not see coming.
How this index is built, and what it does not claim
This study blends published external benchmarks (cited inline throughout) with the structure Blufire uses when it computes a per-account weather-demand figure. Where a number is a Blufire-style aggregate rather than a published external figure, it carries a Demonstrative data chip. No external statistic is presented as a measured Blufire aggregate, and no measured client figure is presented as published research.
Methodology block
The baseline method, stated plainly
A per-account figure is computed by (1) converting local temperature to degree-days from an 18°C base (the BOM standard), (2) counting extreme-weather events by type and region in the L.E.K. WISE manner, (3) regressing demand on these features with autoregressive controls and testing lags k = 0, 1, 2 to find each vertical's response delay, and (4) netting out climate baseline so the reported sensitivity is true weather response, not seasonality. This is standard, published practice. The work is in doing it correctly and per-vertical, not in any single secret step.
What to take from the index
The lift is uneven and directional
HVAC and plumbing demand swings 130% to 610% across the year; roofing and fashion swing under 70%. One seasonal curve averages opposing signals into mush.
Lag differs by vertical
HVAC lags a quarter behind degree-days; roofing spikes the same quarter a storm hits, then declines two quarters later. A single lag value is wrong.
Act 30 to 45 days before peak
Seasonal patterns are reliable 3 to 6 months out, near-term weather 7 to 10 days. The actionable window to commit budget is roughly a month before the peak.
Control for climate, or over-claim
Measure the deviation from seasonal baseline, not raw seasonality. Skip this and you will flag merely-seasonal categories as weather-sensitive.
Get the full Weather Demand Index dataset
The published study above covers the method and the headline figures. The full dataset adds the complete event × vertical × region matrix, per-vertical lag fits, the seasonal calendar by service keyword, and the worked degree-day-to-budget model as a working template.
Want this computed on your own accounts, weather-normalised and tied to margin? See Blufire weather modelling or book a demo.
Most ecommerce benchmarks stop at revenue and gross margin. The number that decides whether a brand survives is contribution margin. This study segments contribution margin, CAC payback and returns by category and brand size, shows the math behind each metric, and explains why the median DTC brand earns far less than its top line suggests.
A lead is not a cost. It is a priced option on a future customer. This study sets the 2026 reference points for what that option costs, how often it pays off, and how fast it turns into revenue, segmented by vertical and deal size.
Across a primary dataset of 156,110 direct-to-consumer customers, the repeat-purchase rate is 18.8%. This study takes that one number apart, then shows the owned-channel lifecycle math that moves it, and what each point of repeat-rate is worth in durable profit.