A regional electric utility we worked with had spent five years deploying advanced metering infrastructure across its service territory. The meters worked. The data flowed. The head-end system happily produced billing files on schedule. What it could not do — and what nobody had budgeted for at deployment — was answer planning questions.

Every quarter, a planning engineer would file a ticket asking for interval data on a specific feeder so they could study a distributed energy resource interconnection or evaluate transformer loading under an EV-adoption scenario. Every quarter, the head-end vendor’s extract took two to three weeks. By the time the data arrived, the question had moved on, the assumptions had drifted, or the regulator had asked something new. Meanwhile billions of 15-minute interval reads sat in a system that treated analytic access as an exception case.

The fix was not to replace the head-end. It was to copy it.

We stood up a partitioned analytics warehouse alongside the head-end and built a near-real-time replication of the meter-event and interval-read streams into it. The first hard problem was time: meters reported in local clock with daylight-saving boundaries, the head-end stored in a mix of UTC and operator time, and downstream systems each made different assumptions. We normalized everything to a single utility-wide time standard and documented the rule. The second hard problem was identity: meters move, premises change, feeders get reconfigured during storm restoration. We conformed the meter-to-premise-to-feeder relationship against the GIS system of record and built a slowly-changing dimensional model so that historical analyses could pin to the topology that existed on the date being studied, not the topology of today.

With the foundations laid, the rest was leverage. Standard load-shape aggregations — feeder, substation, rate class, hour-of-year — were materialized on a daily schedule. Planning engineers got a SQL workbench and a small library of parameterized notebooks for the most common studies: DER interconnection screening, feeder-load forecasting, EV-load impact, transformer loading. A governance layer enforced the regulator’s data-handling requirements through automated row-level controls rather than human review.

Studies that took weeks now take hours. EV-load forecasts at the feeder level now drive the utility’s ten-year capital plan with assumptions traceable directly to the underlying interval data. Quarterly regulatory filings ship on time with documented lineage from the filing back to the meter read.

The next phase, currently in flight, extends the same model to non-metered grid-edge data — substation SCADA, weather, distributed generation telemetry — to support real-time situational awareness for the operations group.