UK retail footfall fell 2.1% year on year in July. Weather is now among the largest swing factors in British footfall. A demand system fitted on seasonal norms cannot separate a hot month from a failing one, and it will read July as a market losing interest.

The formats disagree, and the disagreement is the finding. High street footfall fell 3.8% over the four weeks to 1 August, while retail parks grew 1.2%. London fell 5.3%. Five points separate two formats in identical weather. June had been worse on every format, footfall down 3.4%. Shoppers did not stop buying. They went where the car park was.

Helen Dickinson, chief executive of the British Retail Consortium, put London’s fall down to transport: soaring temperatures made tube and train travel less attractive, and some commuters stayed home. Scotland’s footfall rose 2.7% and Northern Ireland’s rose 2.6%, while England fell 3%. Nothing structural distinguishes a Glasgow high street from a Birmingham one. What separated them was the weather.

June supplies the control. Total retail sales rose 1.9% year on year across the five weeks to 4 July. In-store non-food sales fell 1.1%; online non-food rose 5.1%. Online took 39% of non-food spending, the highest share of the year, Dickinson said. The door lost June. The till did not.

In a heatwave, a door counter is a thermometer that files its readings as demand.

The month was not a marginal case. July’s UK mean temperature reached 17.4°C, 2.1°C above the 1991–2020 average and the second warmest July on record. Sunshine ran to 258.7 hours, 149% of average, the sunniest July in a series that begins in 1910. A buyer planning against the seasonal norm was planning against a month that did not arrive.

The comparison carries a second defect. A year-on-year figure measures something only if one of the two years was ordinary. July 2025 was the fifth warmest on record, sitting inside the warmest UK summer ever measured, a summer with four heatwaves. July 2026 is marked against a July already suppressed. The model is comparing one exception with another and calling the difference demand.

None of this is unknown to forecasters. The difficulty is where it lands in the calendar. Badorf and Hoberg, studying 673 brick-and-mortar stores for the Journal of Retailing and Consumer Services, put the weather effect on daily sales as high as 23.1% by store location and 40.7% by sales theme, and found that models without non-linear terms misestimate the extreme days. Weather forecasts improve sales forecasts up to seven days ahead; the improvement decays past that. An apparel season is committed months before that window opens.

There is also nothing in the record to fit against. By mid-July the UK had logged eight days above 34°C, beating a mark previously shared by 1976 and 2020, and 2026 became the first year on record to reach 35°C in May, June and July alike. A model learns the shape of a summer from summers it has already seen.

The error does not stay in July. A decline of this size enters the plan as a trend point and becomes the comparative that next July is judged against. Weather passes. The record it leaves in the plan does not.

The obvious rebuttal is that retailers have noticed. Dickinson says they are investing in air conditioning and more efficient refrigeration. That repairs the building. It does not reach the ledger where the month was recorded.

Any store manager on any English high street could have explained July in one sentence. The forecast has nowhere to put a sentence.