AI system for automatic water, gas, and electricity metering

Automated resource metering: where the manual scheme breaks down

With manual meter reading, the metering cycle depends on people.

A meter reader records the number from the meter into a log or table, then it is transferred into the accounting system — and each step introduces its own error: a missed node, a transposed digit, a reading taken not on the scheduled date.

Readings arrive fragmented, some backdated, and until reconciliation no one knows exactly where accuracy was lost.

Discrepancies in the statements surface only after consolidation: the sum across consumers does not match the reading at the inlet, and the cause is unclear. A data entry error, debt from a previous period, and an actual leak all look the same — as an imbalance of a few percent. It has to be investigated manually, pulling primary logs, and this takes days.

Resource losses under such a scheme are detected with a delay of an entire reporting period. A leak in a section or unauthorized withdrawal becomes visible only when the data has already been consolidated — that is, when the resource is gone and cannot be returned. The later the signal, the more losses are absorbed into the company's balance.

Closing the reporting period stretches out: until all nodes reconcile, the account cannot be closed, and discrepancies accumulate.

The balance across metering nodes also fails to reconcile because readings are taken on different days and at different intervals — it is easier to write off the imbalance as measurement error than to find its source.

We examine how this is closed: continuous automatic reading, a single time scale, and balance reconciliation across nodes at the moment data arrives.

What switching to automated meter reading delivers

Manual rounds set the upper speed limit: while the operator walks the sites, some meters have already produced new values, while others still have old ones.

Automated reading removes the window period — the reading collection time shrinks to one polling cycle, and the billing period closes on a single snapshot rather than on data from different days.

Reliability of metering data rests on the fact that the value enters the system from the meter, not from a transcribed log.

Manual entry with rounding and edits disappears, and the discrepancy between the sum across consumers and the inlet node is visible the same day, not at the end of the month in a reconciliation act.

Leak detection is based on minimum night flow: when consumption in a building falls to the base level, deviation from the accumulated profile becomes a signal.

The system compares night values with the site's own history and with neighboring nodes, so a leak at the inlet is visible long before the next payment document.

Unauthorized withdrawal manifests as a persistent discrepancy: the volume at the common node is higher than the sum across individual meters, and the difference repeats from period to period.

Balance control turns the search for illegal taps and interference into a regular procedure instead of occasional checks.

A separate class of failures is “the operator did not reach the meter”: an inaccessible basement, denied access, a missed transmission deadline.

With automated reading these gaps disappear, and the dispatch summary shows node status, emergency events, and daily balance on one screen — the dispatcher works with deviations rather than with data collection.

Meter reading methods: what to choose for your installed base

A heterogeneous installed base is a common situation: at one site there are meters with pulse output, next to them mechanical meters read by a person, and some points have no regular reading at all.

Therefore, choosing a collection method is not a race for the most modern technology but a calculation for the specific installed base: how many points, how often readings are needed, and what is cheaper to maintain over several years.

The key criterion is the total cost of ownership. It consists of rounds and manual entry, meter replacement, communication channels, resolution of disputed readings, and downtime when data does not reach the system on time.

In practice, a hybrid metering scheme more often wins: recognition covers the old installed base, while meters with data transmission are installed at new and problematic nodes.

Modernization of the meter fleet proceeds in waves — first the sections with the greatest losses, then the rest, without stopping metering and without replacing everything at once.

Implementation stages: from installed base survey to pilot zone

Implementation begins not with equipment procurement but with a survey: it is important to understand how many metering points actually exist, which of them already transmit data, and which will need retrofitting.

An error at this step is the most expensive — an incorrectly counted installed base leads to reworking the collection scheme after installation, when nodes are already mounted and consumers have been notified.

Work sequence

1. Inventory of metering points. We visit sites, record the meter type, presence of a pulse output or industrial interface, condition of communication lines, and access to panels.

   The result is a registry with addresses, binding to consumers, and a mark “ready for connection / needs retrofitting”.

2. Data collection scheme. We determine where to place concentrators, how to route channels, and what to do with sites without communication. We choose polling frequency for the task: one for commercial metering, another for balance and loss detection.

3. Reading processing setup. We define validation rules: outlier rejection, gap recovery, mapping to tariff and customer account. We lock these rules before the pilot zone; otherwise discrepancies will be written off as a system error.

4. Pilot zone. We connect 20-50 points of different types — this tests real conditions rather than ideal ones: weak signal, old meters, nonstandard wiring.

5. Accuracy verification. We compare automated readings with manual readings and control measurements. We investigate the discrepancy at each point down to the cause rather than averaging it out.

6. Handover to operation. We hand over round procedures, dispatcher instructions, the failure escalation scheme, and the manual entry procedure when a node fails.

What stays with you: configurations, diagrams, procedures

We hand over the system not as “a service on our side” but as a set of artifacts with which metering continues to work without us. From then on, support is handled by your team or any other contractor — without reverse engineering or reconstructing logic from system behavior.

• Metering scheme — list of nodes, meters, and communication channels, their binding to sites and tariff zones, and the procedure for recalculating balance when a meter is replaced.
• Access policies — roles for dispatcher, engineer, and accounting, rights to adjust readings and tariffs, action log tied to the user.
• Maintenance procedure — frequency of checks, actions when communication with a node is lost, verification and replacement procedure for meters, escalation points.
• Dispatcher instruction — step-by-step scenarios for each event class: underconsumption, consumption spike, emergency shutdown, manual reading adjustment.
• System documentation — data structure description, reference data, integration points with billing and dispatch, and deployment procedure from scratch.
• Handover package — access to the configuration repository, test environment, shift training, and incident review for the first period of operation.

The practical point of the list is simple: any breakdown can be fixed by your own team without waiting for approval from us. We remain available for support, but you are no longer hostage to a single vendor.

Case study: how we eliminated manual rounds on a water utility network

Water utility site: 340 metering points, mixed fleet — meters from three manufacturers, including Energomera, of different generations. Some points had pulse output, some transmitted data only via an optical port, and some had no communication channel at all.

Readings were taken by meter readers: route on a tablet, log on paper, manual transfer to a table. The period closed in 9–12 days, and consumption discrepancies surfaced already in billing.

What we changed

We installed a unified reading module at the site: it reads both pulse outputs and the optical port, as well as the meter's open protocol, brings everything to a single data model, and sends it at night when the channel is free.

We added two rules: if there is no data for a point for six hours, the system automatically creates a ticket, and a discrepancy above the threshold between flow at the site inlet and the sum across consumers is flagged as a suspected leak. Commissioning took six weeks without interrupting supply.

What participants say

Customer: “Meter readers are no longer the bottleneck — the period closes the next day instead of a week and a half later, and disputed points are visible before billing.” Project engineer: “The main work was not in the meters but in bringing four different data formats to one — after that, connecting a new point takes hours, not a crew visit.”

How automated reading differs from manual rounds

Manual rounds are a discrete sample: readings are taken once a month along a route, and the reading frequency is set not by the need for data but by logistics — how many metering nodes the operator can cover in a shift.

Between two visits, consumption is not visible at all: a leak, night draw-off, or node failure shows up only in the final figure, when there is no longer any way to explain it.

Human factors come into play at every step. Transferring numbers to the statement, rounding, a mixed-up node, a locked room — and the reliability of readings rests only on rechecking.

Comparison with the reference and balance is done after the fact, after billing, so a discrepancy surfaces as a dispute with a consumer rather than as a signal on the control panel.

Automated reading works differently: meters with pulse or digital output — including the widespread Mercury — transmit readings on schedule, and an interval series with a step of an hour or a day lands in the storage.

Labor costs for rounds are reduced to installation and rare maintenance, while comparison with the reference, balance across nodes, and anomaly detection run automatically — a deviation is visible the same day, not a month later.

The schemes do not compete directly: manual rounds cover billing, while automated reading also covers loss control with anomaly review for each node.

Where comparison with a reference and data frequency higher than monthly are needed, manual rounds physically cannot produce the required series.

Common technical doubts before starting a project

A project usually starts with the same questions: what to do with existing meters, how to account for seasonal peaks, what happens when communication is lost, whether the system can be introduced in parts, and how data will reach the billing system. Below are short answers on each topic.

Will seasonal consumption fluctuations break the logic?

Thresholds and the model are calculated from the seasonal profile — irrigation, heating season, night load. Deviation is compared with the expected value for that period, so a summer peak does not turn into false alarms.

What happens when communication is lost?

Readings accumulate in the meter buffer with a timestamp and go to the server when the channel returns. The gap is closed retrospectively and recorded as a separate event — the dispatcher sees “no data,” not “zero.”

Can the system be launched in stages?

Yes, we proceed in waves: first one node or one resource type, then the rest. Each wave is handed over autonomously, so metering already produces data while the next part is still being installed.

How will data reach the billing system?

Via scheduled export or exchange through a software interface in an agreed format. We take reconciliation upon ourselves: any discrepancy between meter readings and billing is resolved before transfer.

Let's discuss the task and come back with an estimate and plan

To make the estimate substantive, send what already exists: a list of metering points and resource types, diagrams of inlet nodes, a list of devices with communication interfaces, and the regulator's reporting requirements.

If some data is missing, we will record the gaps in the problem statement and close them during the site survey.

The review is conducted as an engineering session with your tech lead and operations service: we clarify the methodology for calculating balances and losses, who receives reports and in what form, which actions are allowed automatically, and which only after operator confirmation.

We come back with an estimate and a plan showing:

• scope of work for each metering node — installation, collection setup, integration with accounting systems;
• target architecture: levels of data collection and storage, composition of redundant circuits;
• stage timelines with checkpoints — pilot circuit, trial operation, commissioning into permanent operation;
• work on the customer's side: access, installation windows, approvals with the resource-supplying organization;
• collaboration format: team composition, areas of responsibility, acceptance and documentation handover procedure;
• risks of each stage and ways to mitigate them — communication channel redundancy, fallback to manual reading;
• post-commissioning support procedure and personnel training on working with the system.

Describe the task — resource types, number of sites, what is already installed, and what result is needed. We will come back with a scope estimate, implementation plan, and stage timelines.