The Brief
Everyday AI 5 min read

18% Less Gas, More Comfort: What a Two-Year Heating Test Reveals

NAVION

Share

Smart thermostats have promised savings for years. The marketing language is consistent: cut your bills, stay warm, let the technology handle it. But promises from manufacturers are easy to make and hard to verify, especially when energy prices shift regularly and home conditions vary so widely. What actually happens when someone runs a controlled, two-year comparison between disciplined manual heating control and an AI-governed system? The answer, it turns out, is more instructive than any product brochure.

The Baseline: What Careful Human Control Actually Looks Like

Before evaluating what AI can do, it helps to understand what a motivated, attentive human can achieve. The test described here used a 1930s solid-brick house with a recently insulated loft conversion as the testing environment. During the first winter, the approach was deliberately rigorous: a basic schedule limited heating to a couple of hours in the morning and evening, unused rooms had their radiator valves closed, and the heating was switched off manually whenever the house was empty.

The result was a gas consumption of 14,316.7 kilowatt hours across 213 days, from November through May, after accounting for hot water usage. That figure was already roughly 10 to 15% lower than in previous years when the same household had been less disciplined. This is the baseline that matters. It was not a comparison against wasteful habits. It was a comparison against someone actively trying to minimise consumption, at the cost of comfort and, by the tester’s own account, some domestic tension.

This is what most coverage of smart home technology misses. The relevant question is not whether AI beats careless heating. It is whether it beats careful heating.

What the AI System Actually Does Differently

The AI system tested was Tado X, installed in the summer of 2025. The setup replaced standard thermostatic radiator valves with 13 smart versions, added wireless thermostats in the two largest rooms, and connected a programmer to the boiler. The total hardware cost was approximately £930, with an additional subscription of £3.99 per month or £29.99 per year to unlock the AI features.

The core difference between this system and a standard smart thermostat is granularity. Rather than managing the whole house from a single central point, each room is controlled individually. The AI learns how quickly each room heats up and cools down, then uses that information to time heating more precisely. It performs automatic hydronic balancing, which optimises how hot water flows through the radiator circuit. It detects open windows and turns off the relevant radiator automatically. It switches off heating when the last household member leaves, and preheats the home before their predicted return.

The AI is built using anonymised data from more than one million homes. It does not sell that data to third parties. The system also includes an energy monitoring feature called Energy IQ, which breaks down consumption costs by individual room. In the first week of use, this feature identified that a radiator in an uninsulated porch was consuming a disproportionate amount of energy for a space that is occupied for only seconds at a time. A simple schedule adjustment followed.

After that, the system ran without intervention. The house felt warmer. The boiler seemed to run more frequently. The intuitive conclusion was that consumption had risen. It had not.

Why 18% Matters, and What It Signals More Broadly

Across the same 213-day period, with comparable outdoor temperatures and the same hot water adjustment methodology, the AI-controlled system used 11,735.7 kilowatt hours of gas. That is 2,581 kilowatt hours less than the manual approach, a reduction of 18%. At the energy price cap in effect at the time of reporting, that translated to a saving of £205.70.

Here is why that number deserves attention. The manual baseline was already optimised. The human running it was motivated, consistent, and willing to accept discomfort to keep bills low. The AI system beat that baseline by 18% while simultaneously making the home more comfortable. This is not a story about replacing human effort with automation. It is a story about what happens when a system can process more variables, more continuously, than a person reasonably can.

A human can set a schedule. A human can close a valve. A human cannot simultaneously monitor the thermal dynamics of 13 rooms, adjust flow rates in real time, anticipate departure and return patterns, and recalibrate based on outdoor temperature trends. The AI does not replace judgment. It handles the volume and precision that human attention cannot sustain at scale.

The broader implication extends beyond heating. As AI systems become embedded in physical environments, from homes to buildings to industrial facilities, the value they offer is not dramatic transformation. It is consistent, granular optimisation across many small decisions that compound over time. An 18% reduction in gas consumption, achieved quietly in the background of ordinary domestic life, is a concrete illustration of that principle.

Tado is described as the largest manufacturer of smart heating controls in Europe. Alternatives exist, including Hive, Drayton, and Honeywell. Google’s Nest, which had offered smart-learning thermostats, withdrew from the European market in 2025.

In Short

A two-year controlled test found that an AI-governed heating system reduced gas consumption by 18% compared to disciplined manual control, saving £205.70 over a single heating season, while improving comfort. The AI’s advantage came not from doing something humans cannot understand, but from doing something humans cannot sustain: continuous, room-by-room optimisation across an entire home, every hour of every day.

Based on reporting from The Guardian.

Written by

NAVION