The situation
Nobody spends months researching an air conditioner. They buy one at 9:40 on the morning it hits 31 degrees, after three nights of not sleeping properly.
LSD sells dehumidifiers, heating, and cooling appliances across three sites. These are big-ticket products, but cooling in particular behaves like an impulse buy the moment the temperature climbs. Demand is driven almost entirely by the weather, and it’s heavily regional. The Midlands can be sweltering while Scotland sits at 17 degrees in the drizzle.
We faced two core challenges:
1. The account was built around margin, not demand. Every product was labelled High, Medium, or Low margin based on the gap between the cost of goods and the selling price. It worked in theory—budget flowed to the products that returned the most—but it grouped products by what they made the business, not by what customers wanted today. A heater and a fan could sit in the same margin bucket, so when a cold snap or a heatwave arrived, we couldn’t push one without pushing the other.
Google’s algorithm compounded this issue. Like most large catalogues, spend concentrated on a narrow slice of the range. LSD’s catalogue runs into the thousands, but before the restructure, a highly concentrated fraction of the catalogue was generating 100% of the Shopping revenue. Every past attempt to widen that visibility came back with the same answer from the platform: you need more budget.
2. Our bidding was always a day behind. Smart bidding is good at what it does, but it learns from conversions that have already happened. It sees the spike after the spike, and it doesn’t read a forecast. The alternative—a person manually adjusting bids across seven regions, three sites, and every Shopping and Search campaign—takes most of the morning. And 9am to midday are LSD’s highest-converting hours. By the time manual adjustments were finished, we had already missed the window we were trying to catch.
The engagement
Step One: Rebuild the account around what people actually buy
Working with LSD, we identified the key product types across all three sites (cooling, heating, ventilation, dehumidifying, and so on) and wrote rules that classified the entire feed automatically, using each product’s type or name. Every product was given a strategic home, including the long tail that Google had been ignoring.
We were careful about depth. Split too finely, and budget gets spread too thin to hit targets or give the algorithm enough data to learn from. We kept the categories broad enough to carry meaningful spend, with the option to split further by brand where volume justified it.
The result was an account with levers that matched how demand actually moves. We could push cooling when it was hot, heating when it was cold, and react to competitor activity or industry trends without fighting our own structure.
Step Two: Let the forecast pull those levers
With the account organised by product type, we built a weather engine in a spreadsheet. It’s fast, transparent, and anyone on the team can see exactly why a bid moved.
Every morning, the sheet automatically pulls the forecast for seven UK regions. Those regions are mapped directly onto the location targeting across every Shopping and Search campaign. We agreed on the rules with LSD in advance: hard temperature thresholds and exactly how hard to push each product category when they’re crossed. For example, if the Midlands is forecast to hit 30°C or more, bids on cooling products in the Midlands automatically increase by 50%. The thinking happens once, upfront. The sheet just applies it.
The sheet converts those decisions into a bulk upload file. We paste it into Google Ads Editor and post. The entire account, across all three sites, is updated in under five minutes—well before the morning rush peaks. A human still reviews the changes before they go live, but the human is no longer the bottleneck.
The results
By shifting from a margin-based structure to a product-type structure, we achieved entirely new levels of efficiency and catalogue depth:
With 11% less budget available, we sold more and held total revenue flat. ROAS went up, not down, despite no longer optimising purely for margin.
The bigger story is catalogue depth. Products seeing impressions rose 63%, with 48% of the entire catalogue finally placed in front of customers. The number of unique products generating revenue grew by 56%. We broke the algorithmic assumption that a wider range requires a bigger budget.
Cost per click did rise by 22%—but that was the calculated price of competing for higher-intent searches on in-season products, and it paid for itself: fewer pounds in, more sales out.
The Weather Engine (Peak Months, YoY)
When peak season hit, we didn’t add a penny of budget. We took the same spend and pointed it at the right products, in the right regions, on the exact right days.
Our team of experts are standing by to discuss how we can help to optimise and transform your business through digital-first marketing.
23.png)
