Estimated reads reported as actuals
Consumption reporting that does not distinguish actual from estimated reads, so billing disputes and volume forecasts rest on figures that are partly modelled.
Energy
Version-aware settlement, consumption reporting that states its estimated share, and reconciled asset records.
Energy and water data is high volume, highly regulated and full of gaps. Meters fail, reads are estimated, settlement runs are revised months later, and asset records disagree with what is physically installed. Analytics that ignores those realities produces confident, wrong numbers.
- Consumption reporting that states its estimated share honestly
- Settlement variance explainable between runs
- One reconciled asset view across GIS, asset and works systems
- Regulatory submissions produced from the same model as internal reporting
We build models that treat estimation and revision as first-class concepts: read quality carried as a dimension, settlement versions kept rather than overwritten, and reporting that can state how much of a figure is actual and how much is estimated.
We work with suppliers and retailers, distribution and transmission networks, independent generators and renewables operators, and water utilities, across both regulatory submission and operational reporting.
What goes wrong
Recurring failure modes in energy
Settlement revisions overwriting history
Later settlement runs replacing earlier ones in the warehouse, destroying the ability to explain why a previously reported figure has changed.
Asset registers that do not match reality
GIS, asset management and works systems each holding a different version of what is installed where, making condition and investment reporting unreliable.
Half-hourly volumes overwhelming import models
Interval data at national scale outgrowing a Power BI import model, pushing teams into aggregates that destroy the analysis they actually need.
Metrics
Metrics and their definition errors
Each metric with its definition and the error that most often distorts it. If our reading is wrong for your business, that disagreement is a productive starting point.
| Metric | Definition | The trap |
|---|---|---|
| Actual read coverage | Share of consumption volume supported by actual rather than estimated reads. | Must be volume-weighted; a count of meters flatters the figure because the largest sites read most reliably. |
| Settlement variance | Difference between successive settlement runs for the same period. | Requires version-aware storage; overwriting runs makes the measure impossible to compute at all. |
| Distribution losses | Energy entering the network less energy billed, over energy entering. | Period alignment between network and billing data dominates the result; without alignment the figure swings meaninglessly. |
| Customer minutes lost | Aggregate interruption duration multiplied by customers affected. | Customer counts per feeder must be versioned over time, or historical restatement is impossible. |
| Leakage rate | Water lost from the distribution network over water put into supply. | Night-flow estimation methodology must be recorded against the figure for regulatory submissions to be defensible. |
Use cases
What we typically build
Metering and consumption analytics
Interval and profile-class consumption with read quality carried through, exception reporting for failed and implausible reads, and volume reporting that states its estimated share.
Settlement reconciliation
Version-aware settlement storage with variance attribution between runs, so a change in a reported figure can be explained rather than argued about.
Network asset and condition reporting
Asset register reconciled across GIS, asset management and works systems, with condition, criticality and investment priority in one model.
Outage and reliability reporting
Interruption analytics with time-versioned customer counts, cause analysis and regulatory reliability submissions.
Demand and generation forecasting
Short and medium-term forecasting with weather and calendar effects, backtested, with accuracy published against the incumbent method.
Regulatory submission support
Submission datasets produced from the governed model with validation mirroring published guidance, and a full audit trail of derivations.
Integration
Source systems we connect to
- Meter data management systems
- Smart metering head-end systems
- Settlement and market data flows
- SCADA and network control systems
- GIS and network models
- Asset management systems (Maximo, Ellipse)
- Billing and CRM platforms
- Weather and market price feeds
Regulatory context
Regulatory context by market
- UK: Ofgem and Ofwat reporting, RIIO and price-control submissions, Elexon settlement processes
- EU: ACER and national regulator reporting, REMIT transaction reporting, Clean Energy Package data obligations
- Germany: BNetzA reporting, EnWG obligations, Marktkommunikation message formats
- Switzerland: ElCom reporting and Swissgrid balancing data requirements
- Belgium: CREG and regional regulator reporting, Elia balancing data
- Cross-cutting: NIS2 for critical infrastructure operators, CSRD emissions and energy disclosure
Questions
Energy: common questions
In a lakehouse rather than an import model. Interval data lands in Delta tables partitioned by date and aggregated into serving layers at the grains analysts actually use, with drill-through to interval level for investigation. Direct Lake or a SQL endpoint serves the detail when it is needed, which keeps the semantic model small and fast.
Yes. Settlement versions are stored as a slowly changing fact with run identifiers, partitioned by settlement date, with older runs compacted and moved to cheaper storage tiers under a lifecycle policy. The retention period is agreed against your regulatory obligation rather than kept forever by default.
We build the data layer, validation and submission datasets, and the audit trail that supports them. Submission itself, and the regulatory interpretation behind it, stays with your regulatory team. We make sure the figures they submit are reproducible and traceable.
From activity data with versioned emissions factors, so figures can be restated when factors are revised, and with the calculation method documented against each metric. Under CSRD assurance the method trail matters as much as the number.
Services that apply here
Azure & SynapseAzure pipelines, warehousing and lakehouse engineering built to run unattended.Microsoft FabricLakehouse architecture sized against measured consumption rather than headcount.GovernanceOwnership, cataloguing, classification and access evidence, implemented in the tools teams already use.
Next step
Analytics work in energy
A 45-minute working session with an engineer rather than a sales call. You describe the problem; we set out what we would build, in what order, and what it is likely to cost.
Answered within one working day, by the engineer who would do the work