Smart thermostats that actually save money (and ones that don't)

The smart-thermostat market has done a remarkable thing — it has converged on roughly the same feature set, while persuading three different consumer audiences that each manufacturer is uniquely capable of saving them money. The truth is that the basic technology is now commoditised, the savings claims are largely repeating each other, and the difference between a good thermostat install and a bad one has very little to do with which device you bought.

I'm a proponent of smart thermostats, with caveats. We've measured before-and-after on three different houses now (one our own, two for friends) and the savings come out reasonably consistently — but they come from a few specific behaviours, not from the generic 'AI learning' marketing claims. This piece is about which features actually deliver, which are theatre, and how to set one up so the payback is real.

Headline: a competently set-up smart thermostat saves a typical UK household around £80–£180 a year on heating. The device itself costs £150–£300. Payback is 1–4 years. That's a good return, but only if you actually use it correctly.

What savings actually come from

The genuine savings from a smart thermostat come from four things, in roughly this order of contribution:

1. Better setpoint discipline. Most UK households heat to 21–22°C in occupied rooms when they actually feel comfortable at 19–20°C. A smart thermostat with a good schedule and an honest review tends to drop the average setpoint by 1–1.5°C, which translates to about 6–10% of the heating bill.
2. Tighter scheduling. The traditional dumb timer turns the heating on at 06:30 and off at 22:00 regardless of whether anyone is home. A smart thermostat with proper schedule management can shave 1–3 hours off most days, particularly on weekends, worth another 5–8%.
3. Geofencing turn-down (when properly configured). Worth 2–4% if the household actually has predictable away-from-home patterns. Worth nothing if everyone is at home most of the time.
4. Weather compensation (where supported). Worth 3–6% on combination boilers and considerably more on heat pumps. Reduces flow temperature when outside is mild, dropping standing losses.

Add those four together for a typical UK household and you get 16–28% off the heating bill, which on a £1,400 annual gas spend is £225–£390. The lower end of that range is what most households actually realise; the higher end requires committed configuration.

Features that are mostly theatre

And here are the features I would not pay extra for:

• 'AI learning' / 'self-programming' algorithms. These have been marketed for over a decade and the published evidence shows they save somewhere between 0% and 4% over a well-configured manual schedule. They save a lot more compared to a household that has no schedule, but that's not the comparison the marketing makes.
• Voice control. Charming. Useful for novelty. Doesn't save energy.
• Energy-savings reports / dashboards. They tend to compare your current usage to your past usage, which sounds informative but doesn't help you decide what to change. The decisions are made on setpoint, schedule, and zoning, not on a heatmap.
• Proprietary multi-zone systems requiring TRV replacements throughout the house. Worth doing in selected cases (large house with very different occupancy patterns by zone), rarely worth the £600–£1,500 retrofit cost in a typical 3-bed.
• Holiday-mode learning. A simple manual override does the same job.

The pattern: any feature that promises to save you money without you doing anything is over-promising. Any feature that helps you understand and adjust the system is earning its place.

What to look for in 2026

Strip away the marketing and the genuinely useful features in a 2026 thermostat are:

• OpenTherm or eBus modulation for combination boilers — lets the boiler actually modulate down rather than cycling on/off. Worth 3–5% on its own.
• Weather compensation curves with outdoor sensor input or weather-API estimation, manually adjustable. The default curves are usually too steep.
• Per-zone TRVs that report room temperature back to the central thermostat — lets you actually know if a room is at setpoint without standing in it.
• Schedule clarity: a UI you can actually configure without three support calls. Weekly and weekday/weekend modes minimum.
• Local control without internet dependency: when the cloud goes down (and it does, regularly), the thermostat should keep doing its job. This one matters more as you live with the system longer.
• Heat pump support, if relevant — including support for the lower flow temperatures and the longer warm-up profiles that heat pumps need.

The setup that actually delivers savings

Buying the device is 20% of the work. Setting it up properly is 80%. Here's the configuration approach we use:

1. Pick an honest setpoint. Sit down with the household and agree a real target — usually 19–20°C in living rooms during occupied hours, 17–18°C in bedrooms at night. Drop these by 0.5°C every two weeks for the first month and see if anyone actually notices. Most households can drop 1–1.5°C without complaint; some can drop more.
2. Build a tight schedule. Map the actual hours someone is in each room. Heat zones during occupied hours, set back to 16°C during unoccupied hours rather than off entirely (so the building stays close to setpoint and the recovery is fast).
3. Enable weather compensation with a moderately steep curve and tune over the first heating season. The default is rarely right.
4. Set hot-water heating to one or two specific windows per day rather than continuous re-heat, particularly if you have a cylinder. Two 45-minute heating sessions cover most households.
5. Review the schedule monthly for the first three months. The first version is always wrong somewhere.
6. Set a budget alarm — most thermostats now support this — at the level that would mean something has gone wrong (window left open, fault). It's the diagnostic that catches drift early.

Special case: smart thermostats with heat pumps

Heat pumps are a fundamentally different control problem from gas boilers. Where a boiler is happy being switched on and off, a heat pump prefers to run continuously at low flow temperatures. A thermostat that treats a heat pump like a boiler — turning it off and on, blasting flow temperature up to 55°C+ — actively damages the SPF and the running cost.

For heat-pump houses, the requirements are stricter:

• Weather compensation must work, not be ignored
• The thermostat shouldn't try to use deep setbacks; 1–2°C is plenty
• Hot-water priority should be configurable
• Direct OpenTherm or modbus communication with the heat pump is preferable to simple on/off control

The honest answer is that many off-the-shelf smart thermostats are not built for heat pumps, and using them with one will give you boiler-style behaviour and bad running cost. Manufacturers' own controls, while less polished, are usually better matched to the system.

What I'd buy in 2026

For a typical UK gas-boiler home, I'd look for a thermostat with OpenTherm support, a clear schedule UI, weather compensation, and at least three TRV-equipped zones. Budget £200–£350 for the central unit plus £50–£70 per TRV. Expect a 2–3 year payback against a typical £1,200–£1,600 annual heating bill.

For a heat-pump home, I'd use whatever the heat pump manufacturer's controls are, supplemented by their own app for visibility, and skip the third-party smart thermostat entirely. The native controls are imperfect but the integration matters more than the polish.

For a small apartment with no zoning needed, the simplest internet-connected thermostat with a phone app and a decent schedule does most of the work for under £150 and pays back in 18 months.

What I would not do: spend £500+ on a single smart thermostat for a single-zone system. The marginal saving over a £200 device is rarely justified.

Privacy, lock-in, and the long term

One thing the marketing rarely discusses: smart thermostats are also data devices. They report your occupancy patterns, your setpoint behaviour, and your heating profile to a manufacturer's cloud. Most manufacturers are commercially reasonable about this; some are less so. Read the privacy policy before buying, particularly if you care about where your home-occupancy data is stored.

Lock-in matters too. A thermostat that requires its proprietary cloud to operate fundamental functions is one that becomes a paperweight when the manufacturer changes strategy or goes out of business. It's not a hypothetical — the smart-home industry has a record of orphaning products. Devices that support a local-control protocol (OpenTherm, Z-Wave, Matter, or simple HTTP) have a longer practical life.

The longer-term question — covered in our piece on smart-home systems that survive a decade — is whether the smart thermostat you buy in 2026 is going to outlast the company that made it. Plan for the answer being 'maybe not' and prefer products that give you fall-back local control.

What I'd flag for households with vulnerable occupants

One audience the smart-thermostat marketing handles badly is households with elderly residents, young children, or anyone with a chronic health condition where indoor temperature stability matters more than energy savings. The default behaviour of a smart thermostat — aggressive setbacks during 'unoccupied' hours, geofencing-based turn-down, learning algorithms that drop the schedule a degree if no one complains — is the wrong behaviour when the cost of a cold morning is more than a discomfort.

The configuration to look for in those cases is a constant-temperature mode (no setbacks deeper than 1°C), with a moderate setpoint (19–20°C in living areas, 18°C minimum in bedrooms), and any savings driven by weather compensation rather than schedule deepening. The arithmetic on energy savings is worse, but the safety argument outweighs the £50–£80/year you'd give up. Document this configuration somewhere, because the next person to look at the thermostat — partner, carer, installer — will assume the default-savings setup and re-enable the wrong things.

Where geofencing is used in such households, set it conservatively — never turn the heat off entirely when 'unoccupied', drop only by 1°C, and avoid relying on a single phone's location signal when alternative occupants may also be present. The honest framing: smart thermostat in this configuration is a comfort and convenience tool first, energy-saver second. That trade-off is fine, just don't pretend otherwise.

A smart thermostat is one of the better small retrofit purchases a UK household can make, with a 1–4 year payback and a meaningful contribution to the comfort-and-cost balance. But the saving comes from setpoint discipline, schedule tightness, and weather compensation — not from any vendor's branded AI. Buy a competent device, configure it carefully, review the schedule monthly for the first quarter, and the payback is real.

What you should not do is treat the device as the answer in itself. The device is a tool. The savings come from the household being honest about how warm it actually wants to be and when it actually needs the heating on.