The promise of smart thermostats saving you money is one of those claims that sits in a frustrating no-man's-land: plausible enough to sell hardware, vague enough to never be falsifiable. Manufacturer claims of 25% savings come wrapped in so many caveats — house type, insulation, baseline behavior, energy price fluctuations — that they function more as marketing than engineering. One journalist decided to actually run the experiment. The test bed was a 1930s solid-brick house with a recent insulated loft conversion, a 10-year-old Worcester Bosch combi boiler, and 13 radiators. Winter one (November 2024 to May 2025): manual control with a Hive thermostat's smart features disabled, basic scheduling, aggressive TRV management, and the kind of domestic miserliness that earns you cold stares from family members huddled under blankets. Result after controlling for hot water: 14,316.7 kWh across 213 days, roughly 10-15% below the household's previous winters. Winter two: a full Tado X system — 13 smart radiator thermostats, two wireless room thermostats, a wired programmer — running all AI features including geofencing, open-window detection, automatic hydronic balancing, and per-room scheduling. The system costs approximately £930 for hardware plus £29.99 per year for the AI subscription. After the same 213-day period with relatively similar outside temperatures and the same hot-water adjustment, the AI system consumed 11,735.7 kWh — 18% less gas than aggressive manual control, saving £205.70 at the current price cap. What makes this test genuinely useful is the baseline. The journalist wasn't comparing AI to a thermostat left at 21°C around the clock — the standard trick that inflates every manufacturer's savings claim. He was comparing it to an already-optimized manual regime that was itself 10-15% below his previous consumption. The AI still won by 18%, and the house was warmer. The system's room-by-room control, which shuts off radiators in unoccupied rooms and modulates boiler demand accordingly, appears to be the primary mechanism — not some opaque learning algorithm. The economics are straightforward but have a long payback. At £205.70 annual savings against a £930 hardware cost plus roughly £30 per year subscription, you're looking at roughly five years to break even — assuming energy prices, boiler efficiency, and the hardware itself all hold steady. For households with less disciplined manual control, the savings could be substantially larger and the payback faster. The system does not support OpenTherm on the journalist's boiler, which would have enabled flow-temperature modulation and potentially even greater savings. The side benefits reported are worth noting: per-room humidity monitoring, the ability to heat a bathroom before a shower without warming the entire house, and the elimination of household thermostat wars. These are real quality-of-life improvements that don't show up in kWh numbers but matter to actual adoption. The data also flagged an uninsulated porch radiator as an energy sink within the first week — a diagnostic function that pays for itself once. Two caveats the piece doesn't fully address: the test is n=1 with no blinding or independent verification, and the comparison winters may not have had identical heating-degree days despite "relatively similar outside temperatures." Still, as a real-world controlled experiment against a genuinely tough baseline, this is far more informative than anything in the manufacturers' marketing materials.
