What a live industrial energy fleet reveals
WattEY, the industrial electricity-monitoring platform Ali Ahmed built at BijliBachao.pk, measured 1.92 GWh net across 43 meters and 16 live organisations in Pakistan over 284 days. Rooftop solar offset ranged from 2.7% at the weakest site to a net-exporting 232% at the best — the same technology, wildly different outcomes; peak-hour exposure ran from 8.9% to 32.5% of grid import; and one site drew 229 kWh in the 3 AM hour. Every figure below is measured, fleet-level and anonymised.
Source: the WattEY production database at BijliBachao.pk, read on 25 August 2026 (pinned cutoff). Energy is net of solar export. Licensed CC BY 4.0.
The dataset
How much data is this?
| Metric | Value |
|---|---|
| Meters | 43 deployed (34 live at sites) |
| Organisations | 16 live (19 with meters assigned) |
| Energy measured (net) | 1.92 GWh (1,918,068 kWh) |
| Days of continuous data | 284 |
| Readings stored | 767,000+ |
| Readings per day | ~9,700 |
| Highest instantaneous load | 635.76 kW |
| Busiest single fleet day | 31,235 kWh (30 July 2026) |
| Largest single site to date | 460,827 kWh |
Energy is net — grid import minus solar export — which is what a site is billed on; gross throughput is higher but double-counts exported solar.
Fleet patterns
How differently do similar sites behave?
These are the patterns no single monthly bill can show — the reason to measure at all.
Peak-hour exposure varies nearly fourfold
Share of grid import that lands inside the expensive evening window, across 20 sites.
| Site | Peak-hour exposure |
|---|---|
| Lowest | 8.9% |
| Fleet average | 18.7% (20.1% energy-weighted) |
| Highest | 32.5% |
| Spread | 3.7× (nearly fourfold) |
Solar offset ranges from 3% to a net-exporting 232%
Rooftop-solar export as a share of a site’s grid import over 90 days — the best site is a genuine net exporter, sending back more than twice what it draws.
| Measure | Value |
|---|---|
| Weakest system | 2.7% |
| Best system | 232% (net exporter — sends back 2.3× its draw) |
| Largest exporter by volume | 45,699 kWh in 90 days |
| Systems excluded (exported under 1%) | 4 of 17 |
Half the fleet runs under 49% of its own peak
Load factor — average load against a site’s own 95th-percentile peak. You pay for the peak.
| Site | Load factor |
|---|---|
| Least efficient | 8.8% |
| Median site | 48.6% |
| Most efficient | 77.7% |
The surprises
What runs when nobody's watching?
| Observation | Value |
|---|---|
| Highest overnight draw vs own average | 182.3% |
| One site, draw in the 3 AM hour | 229 kWh |
| Highest weekend vs weekday consumption | 128.2% |
| Lowest weekend vs weekday consumption | 16.1% |
| One site typical draw | 61.82 kW |
| That site’s peak event | 96.53 kW (1.56× its typical draw) |
Data honesty
How often does a meter go quiet — and what then?
Every reporting gap is shown, not filled in. The energy is kept; the timing nobody observed is not invented.
| Metric | Value |
|---|---|
| Silences detected (any length) | 1,170 |
| Gaps over one hour | 497 |
| Average gap over one hour | 5.0 hours |
| Longest gap | 125.3 hours (over 5 days) |
| Largest single backfill | 15,860 kWh (after a 56.9 hours silence) |
The meter mix
The 34 meters live at sites, by connection type.
| Meter type | Count |
|---|---|
| Standard grid | 15 |
| Solar / net-metered | 15 |
| Solar + generator | 2 |
| Dedicated generator | 2 |
Going deeper
What else the read showed
Further fleet-level patterns the verified 25 August read computed directly from the telemetry.
One fifth of grid import lands in the peak window
| Window | Share of grid import |
|---|---|
| Peak | 20.1% |
| Off-peak | 79.9% |
Over half the measured sites sit below 0.90 power factor
Of 29 sites with enough loaded samples (measured only while a site drew over 1 kW), 15 run below 0.90 — the band where many tariffs add a penalty.
Energy by meter class
Net energy measured over the record, split by the class each meter resolves to.
| Meter class | Energy (kWh) |
|---|---|
| Standard grid | 1,168,666 |
| Solar / net-metered | 747,510 |
| Dedicated generator | 1,892 |
Monthly energy trend
Net metered energy per month across the record. January is genuinely high — three large industrial meters were live then and have since gone quiet. The final month is partial.
| Month | Energy (kWh) |
|---|---|
| Nov 2025 (partial) | 5,480 |
| Dec 2025 | 48,905 |
| Jan 2026 | 607,851 |
| Feb 2026 | 55,341 |
| Mar 2026 | 114,683 |
| Apr 2026 | 123,314 |
| May 2026 | 128,597 |
| Jun 2026 | 110,113 |
| Jul 2026 | 268,716 |
| Aug 2026 (partial) | 455,069 |
Methodology
How this was measured
Every figure is read directly from the WattEY production database, not estimated or projected — each traceable to a query and pinned to a hard cutoff so it reproduces exactly. WattEY reads a three-phase meter every three minutes — sixteen values per reading — and computes the fleet-level aggregates above from that raw telemetry. The snapshot was taken on 25 August 2026; the numbers grow every day, so they are dated rather than presented as timeless facts. Energy is measured net — grid import minus export — for solar and net-metered meters, and as total throughput for grid and generator meters.
The dataset holds 2.30 million raw telemetry rows. Raw readings are kept for 90 days; the hourly aggregate keeps everything, so 213,180 kWh (11% of the lifetime total) — from meters that stopped reporting before the raw window — survives only in the aggregate. It is real measured energy that can no longer be re-derived from raw readings, and it is counted.
Where the platform cannot compute a figure, this dataset says so rather than estimating one: there is no grid / solar / generator source split (the meters measure grid import and export, not solar generation directly), and no sector breakdown (the platform stores no industry field). A dated measurement is honest; a guessed one is not.
Everything here is fleet-level and anonymised: no client, site name, serial number or per-site figure appears, by design. A “site” is one anonymous metered organisation. The platform, and the analysis, were built by Ali Ahmed at BijliBachao.pk.
The data is released under Creative Commons Attribution 4.0. Use it freely with attribution to Ali Ahmed / BijliBachao.pk. Download the CSVSee also the full data index.