VMD Energy Intelligence
Energy data that
carries a decision.
Measured energy behaviour turned into operational and economic understanding: what happened, what it cost, which intervals created the exposure, and what the evidence behind that answer is.
Why it exists
A meter reading is not a decision.
Most energy software answers how much and when. Those are measurements, and a company that has them still has to work out what to do. The questions that decide anything sit one layer further in, and they are the ones this product is built to investigate:
- Why did that peak happen, and what was running when it did?
- Did it cost enough to be worth acting on?
- Which intervals actually created the exposure?
- Was it avoidable, or did a physical limit make it unavoidable?
- Would a battery have changed that outcome — and by how much?
- What evidence supports the answer?
Not every question has a universal answer, and the product does not pretend otherwise. What it does is make each one investigable against a company’s own measurements rather than argued from experience.
How it works
Measure, analyse, explain, quantify.
Five movements, and each one only exists because the one before it did. The fourth is where energy software usually stops short: knowing what an event cost is what turns an observation into something a person can decide about. The fifth is the newest, and the only one that makes a claim about the future.
Measure
Sites, meters and validated interval measurements, with power and energy kept explicitly apart — kW is not kWh, and a model that blurs them produces confident nonsense. Tariff data sits alongside the measurements rather than being applied as an afterthought.
Analyse
Peak-demand analysis and deterministic anomaly detection over the stored series. Deterministic: the same data produces the same finding, and no model is involved in deciding whether something is anomalous.
Explain
Every result carries what it was derived from — the contributing intervals, the tariff context that applied, and where a physical constraint limited what was possible. A finding that cannot be traced back is not usable in an argument.
Quantify
What the event was worth in money, whether it was material enough to act on, and what a feasible alternative would have produced instead. This is where an energy observation becomes an economic one.
Project
Expected demand for the period ahead, issued as a claim that can be checked afterwards rather than as a confident number, and priced the way a demand charge actually settles. The first four movements describe what happened; this one is the only place the product speaks about what has not happened yet.
Today
What is built.
These are working areas of the product, not intentions. Everything else on this page that is not built is marked as such. All of it is deterministic: the same data produces the same result, and no language model sits in any of these paths.
Energy data foundation
A model of sites, meters and validated interval measurements, with explicit kW and kWh semantics and tariff data held against it. Historical energy data is ingested into that model and persisted.
Peak and anomaly analysis
Peak-demand analysis across the measured series, and deterministic anomaly detection that flags what departs from the established pattern.
Economic exposure
Energy events costed against the tariff context that applied, and assessed for materiality — which of them carried enough economic weight to be worth attention.
Evidence and provenance
The intervals that contributed to a result, the assumptions applied, and the physical reason a limit was reached, retained alongside the result itself.
Historical counterfactual analysis
What a past period would have looked like under a different feasible configuration, including battery feasibility and state-of-charge–aware reasoning over the stored demand series.
Demand forecasting
Expected demand for the next twenty-four hours at the meter’s own interval, issued as a versioned claim with a prediction interval derived from observed residuals. A forecast may only use intervals that had already closed when it was issued, and that boundary is enforced rather than trusted, so accuracy can be measured against what actually happened.
Forward economic projection
What a proposed change to a future period would be worth, priced the way a demand charge actually settles — two projected monthly maxima compared, never a reduction multiplied by a rate — and afterwards checked against what the period turned out to cost.
Constraints and bounded search
A declared production schedule with a declared dispatch policy, and the grid connection limit, applied as verdicts on an alternative rather than as penalties. A bounded, deterministic search over a peak-cap grid records which alternatives it evaluated and what each was worth.
Economic intelligence
From an event to what it was worth.
A peak expressed in kilowatts is a fact about equipment. The same peak expressed in money is a fact about the business, and only the second one gets a decision made. The analysis carries an event along this chain:
- The event, located in the measured series
- The intervals that produced it
- The tariff context that applied at that time
- What it cost
- Whether that was material enough to act on
- What a feasible alternative would have produced instead
This chain is historical, computed against one company’s own measurements and tariff context: it describes what did happen and what could have happened instead. Forecasting is a separate capability with its own section above, and its claims are marked as forecasts wherever they appear. One period’s result is not a saving another company should expect.
Evidence
Every result says where it came from.
A number nobody can interrogate does not survive its first meeting. Where the implementation supports it, a result carries the working behind it:
- Which measurements contributed
- Which intervals mattered, and by how much
- The tariff context used in the calculation
- The assumptions the analysis applied
- The physical constraint that limited an alternative, where one did
- Why the result was classified as material
This is provenance at the level of the analysis: the data, the method and the assumptions are recoverable. That is possible because there is no language model in any of it — the costing, the anomaly detection and the forecast are computed by stated methods, and a stated method is the only kind of reasoning that can be handed to somebody else and checked.
Counterfactual analysis
What would have happened instead.
A counterfactual asks a plain question about the past: given what was actually measured, would a different feasible configuration have produced a different result? For battery feasibility the analysis runs like this:
Historical demand
The measured series for the period, interval by interval.
Battery constraints
Capacity, power limits and the state of charge the battery would have been in.
Feasible intervention
What the battery could actually have done at each interval, given that state of charge — not what it would have needed to do to produce the best answer.
Resulting difference
What the period would have cost under that alternative, against what it did cost.
This is analysis over stored measurements. Energy Intelligence does not connect to a battery, does not dispatch one and does not control one — there is no battery integration and none is claimed. The feasibility model reasons about what a battery would have done; a physical battery is never in the loop.
Modelled example
The shape of one analysis.
Not a customer and not a case study — the sequence a battery-feasibility analysis actually follows, with the figures left out because publishing one without its conditions would be a claim rather than an example.
Input
A historical interval series for one site, with the tariff context that applied.
Finding
A demand peak identified in that series, and the intervals that produced it.
Exposure
What that peak cost under the tariff in force, and whether it was material.
Alternative
Whether a battery within stated capacity and state-of-charge limits could have reduced it, and where a physical limit stopped it going further.
Difference
The gap between what the period cost and what the alternative would have cost.
A result of this kind describes one site, one period and one tariff. It is a historical analysis, not a projection and not a guaranteed saving.
In development
Where this is going.
None of the following is built. It is the direction the product is being developed in, and it is listed separately from what is built so the two cannot be confused:
- ERP contextualisation, so an energy event can be read against what the business was doing
- Production-aware planning — the product judges an alternative against a schedule somebody declared, and generates no schedule of its own
- Scenario and decision simulation
- Recommendations drawn from the analysis — alternatives are evaluated and priced, and none of them is put forward as the one to take
- A broader economic decision workflow around them
- An assistant over the modelled data
To be explicit about what is not being built toward either: there is no control of building systems, photovoltaics, storage or charging, no automatic production rescheduling, no autonomous writes into an ERP and no unattended operation. The product analyses and explains; a person decides and acts.
Where it sits
One domain of a wider direction.
Energy Intelligence is a domain product within VMD’s work on enterprise decision intelligence — the same principle applied to one subject: model the data properly, compute deterministically, explain the result, and leave the decision with the person accountable for it. The wider platform is under development and is described where it is described; nothing here depends on it being finished.
Bring your energy data.
The product is in development and works against real measurement data. If you have interval data, a tariff and a question about a period, that is the conversation.