Short answer
For a single site, the value is finding waste rather than optimizing a system. Sub-meter by zone or by major equipment first — without that, no analysis can say more than the bill already does. Then build a baseline of expected consumption given weather, day and activity, and alert on the gap. In LYVIA’s own engagements the first findings are usually equipment running when nothing is happening, they cost nothing to fix, and they arrive early — client observation rather than a published benchmark. Automatic control of heating comes last, if at all.
Visibility before optimization
A single meter and one invoice is a common starting point for a company of ten to a hundred people. It tells you the total and nothing about the cause, so the question that matters — which part of the site is responsible for the increase — has no answer.
That is why software layered on top of a single meter disappoints. It can present the total more attractively, compare it to last year, and project it forward — all of which is a restatement of information you already had. The actionable version starts one level down, with consumption attributed to something you can switch off, reschedule or repair.
Sub-metering is the whole prerequisite
Splitting the site into a handful of measured circuits is the step that unlocks everything else. Not every circuit — the major consumers and the obvious zones, which in most buildings means heating and cooling, compressed air if you have it, the main production equipment, lighting, and refrigeration.
This is electrical work rather than software, and it is the part of the project people try to skip because it is unglamorous and involves a contractor. It is also where the first savings come from, frequently before any analysis: simply seeing a circuit drawing power at three in the morning answers a question nobody had thought to ask.
A test for any energy proposal: ask what it will tell you that the invoice does not. If the answer does not involve measurement below the main meter, the honest version of that product is a dashboard for a number you already know.
The three things a model adds
| Use | What it produces | What it needs |
|---|---|---|
| Baseline | Expected consumption given weather, day of week and activity — the reference everything else is measured against. | A period of readings, plus outside temperature and a measure of activity. |
| Anomaly detection | An alert when actual diverges from expected: equipment left running, a failing unit drawing more, heating and cooling fighting each other. | The baseline, and a named person who receives the alert. |
| Schedule optimization | Proposed start and stop times — pre-heating against a forecast rather than a fixed clock, shifting flexible load. | A tariff or a comfort target that makes timing worth something. |
Why the baseline is the real product
Raw consumption is almost uninterpretable. A high week might be a cold snap, a busy production schedule, or a fault — and without a way to separate those, an alert on absolute consumption fires every winter and gets muted.
A baseline models what the site should have used given the conditions. Once that exists, the useful signal is the residual: consumption that the weather and the workload do not explain. That is where waste lives, and it is also what makes savings provable afterwards — you can show that consumption fell relative to expectation rather than because the season changed, which is the difference between a claimed saving and a demonstrated one.
It is the same comparison-against-expected pattern that makes monitoring work in other domains, and it fails in the same way when the inputs are wrong — the general version is in our note on AI in decision making.
Heating and cooling, and the comfort constraint
In the offices LYVIA has instrumented, heating and cooling is usually the largest controllable consumer, and the easiest savings are in timing rather than temperature. A system that starts on a fixed schedule heats an empty building on a mild morning and arrives late on a cold one; a forecast-aware schedule fixes both.
Handing over live control is a different level of commitment. Comfort complaints are immediate, personal and escalate to whoever authorized the project, and one uncomfortable Monday can end it regardless of the arithmetic. The sequence that works is to let the system propose, apply changes manually for a period, and only consider automatic control once the proposals have been right for long enough — with hard limits a person sets and can override.
Shifting load, where the tariff allows it
Moving consumption to cheaper hours only pays if your tariff distinguishes between hours. Many small business contracts do not, or do so weakly, and in that case load shifting produces operational disruption for no benefit.
Check the contract before designing anything around it. Where a time-of-use structure does exist, the candidates are the loads that genuinely do not care when they run — batch processes, charging, some refrigeration duty cycles, water heating. Anything tied to people or to a production schedule is not flexible, whatever a model calculates. And if you generate on site or hold storage, the arithmetic changes enough that it deserves its own analysis rather than a rule of thumb.
What not to expect
- Savings without a change. Every reduction comes from someone altering a schedule, repairing something, or replacing equipment. Monitoring finds the opportunity; it does not take it.
- A generic payback figure. It depends on your bill and on what the monitoring finds. A vendor quoting a percentage before seeing your site is quoting a brochure.
- Insight from the main meter alone. Covered above, and the most common disappointment in this category.
- Regulatory reporting as a by-product. If you report under a scheme — the UK’s Energy Savings Opportunity Scheme, an energy audit obligation in your own jurisdiction, or a voluntary certification — its definitions and boundaries are specific and vary by country and by year. Check them against the scheme itself rather than assuming a monitoring tool matches.
A first month that pays for itself
Sub-meter the five or six biggest consumers. Record for a few weeks without building anything clever, and look at the overnight and weekend traces — the period when the site should be quiet is where unexplained consumption is most obvious and cheapest to fix.
Then build the baseline and alert on the residual, sending it to a named person who can act. Everything beyond that — schedule optimization, tariff work, automatic control — is a second phase that the first one will have justified or not. Where this sits against other priorities in a first year is in our AI strategy roadmap, and if the same sensors are also watching equipment health, that is a related but separate project — see predictive maintenance.
Energy is a narrow vertical and rarely the first thing worth doing — the cross-industry candidates are in our list of use cases that survive the pilot.
Where the same consumption figures are being requested by a customer rather than a regulator, the reporting side is in AI for ESG reporting.
Frequently asked questions
Can AI reduce a small company energy bill?
It can find where the energy goes and catch consumption that should not be happening — a compressor running overnight, heating fighting cooling, a line drawing power at the weekend. What it cannot do is reduce consumption on its own: every saving comes from a change someone makes to a setting, a schedule or a piece of equipment. The model shortens the distance between waste starting and someone noticing, and on the sites LYVIA works with that distance is usually measured in months rather than days — an observation from client work, not a published figure.
What has to be in place before starting?
Sub-metering. A single meter for the whole site tells you the total and nothing about the cause, which means any analysis on top of it can only describe a number you already had. Splitting consumption by zone or by major equipment is what makes everything else possible, and it is an electrical job rather than a software one — the honest first step for most sites, and often the only one needed to find the first savings.
What is a consumption baseline and why does it matter?
It is a model of what your site should consume given the conditions — outside temperature, day of the week, production volume, occupancy. Without it, a high reading is ambiguous: cold week or a fault? With it, you compare actual against expected and the difference is the signal. Almost everything useful in this area is built on that comparison rather than on raw consumption.
Does this need new hardware?
Usually some. Recent building management systems and industrial equipment often already expose readings, in which case the work is integration rather than installation. Older sites generally need sub-meters, and the cost sits mostly in the electrical work rather than the devices. Treat any proposal that requires no hardware at all with suspicion — it can only be re-describing the bill you already receive.
What about controlling heating and cooling automatically?
Optimizing a schedule and pre-heating against a weather forecast is realistic, and on the office sites LYVIA works with it is often the largest single saving available — client observation, not a published figure. Handing over live control is a bigger step: comfort complaints escalate quickly, and one cold Monday morning can end a project. The workable sequence is to let the system propose schedule changes, apply them manually for a period, and only then consider automatic control, with limits a person sets.
How long before it pays for itself?
It depends entirely on your bill and on what the monitoring finds, so any generic payback figure is marketing. What is reliable is the shape: the first savings usually come from things nobody knew were running, they arrive within the first weeks of visibility, and they cost nothing to fix. As a working rule from LYVIA's own engagements rather than a published benchmark, if a site has never been sub-metered, the initial findings are typically the cheapest energy savings it will ever make.
If your site has one meter and a rising bill, the first step is measurement rather than software — and it is a short conversation to scope. Book a call.
