The outage prediction model flagged the failing transformer three days early. The crew still didn't know, until they arrived on site, that the substation it fed also served the regional water treatment plant.
A forecast that arrived early and still landed too late
Picture a utility whose grid analytics platform correctly predicts a transformer is likely to fail within the week. The alert reaches the operations team with days of lead time enough to plan a controlled replacement instead of an emergency response. What the alert doesn't say is which customers are actually fed by that transformer, or that one of them is a water treatment facility whose backup power only covers a few hours. That relationship existed somewhere in the utility's systems. It just wasn't connected to the forecast the operations team was acting on. (This is an illustrative scenario reflecting a pattern that can emerge across grid operators, not a specific customer engagement.)
This illustrates a pattern that can emerge in energy and utilities: the data required to make a good decision usually exists somewhere in the organization, and the forecasting or detection model built on top of it may be doing exactly what it was designed to do. What breaks down is the last step, connecting an accurate technical signal to the operational and business relationships that determine what should actually happen next.
Three places the gap shows up
1. An outage that doesn't know which customers it affects
The operational question: "This feeder is about to be de-energized for planned maintenance. Which customers does it actually serve, and are any of them hospitals, water treatment facilities, or other critical infrastructure that need advance notice?"
Answering that means tracing a chain from the feeder, to the substation and circuit it belongs to, to every customer connection downstream, to whichever of those customers are flagged as critical infrastructure — a chain that typically spans a grid topology system, a customer information system, and a separate critical-facilities registry, maintained by three different teams with no routine reason to reconcile with one another.
2. A wildfire risk score that doesn't know what it's scoring
The operational question: "This span of line is showing elevated vegetation growth and sits in a high fire-risk zone under tomorrow's wind forecast. Does it also carry a circuit that feeds a hospital or a community with limited evacuation routes?"
A utility can have accurate vegetation data, a solid weather feed, and a defensible fire-risk score for a given span of line, and still be missing the one relationship that determines how urgently to act: whether that span also carries a circuit serving a facility or community where the consequence would be especially severe.
3. A curtailment decision that doesn't know what it's trading away
The operational question: "Grid capacity at this substation is constrained this afternoon. If we curtail generation from this solar facility, which power purchase agreements does that put us at risk of breaching, and what's the financial exposure?"
A grid operations team can have an accurate view of substation constraints, and a generation forecast can be entirely correct but deciding which generator to curtail means connecting that constraint to interconnection and power purchase agreements typically held by a commercial team, in a system grid operations has no routine reason to query.
A relationship problem, not a modeling problem
In each case, the forecasting or detection system may be doing exactly what it was designed to do. What's missing is a relationship connecting grid topology, customer data, geographic risk, or commercial agreements, spread across systems that are each reasonably well governed on their own, but were never connected to each other in a way an operations team could draw on directly.
The forecast was accurate. The relationship that would have told the operations team what it actually meant lived somewhere else entirely.
Grid topology, customer information, asset condition, and commercial agreements typically originate across SCADA and outage management platforms, customer information systems, GIS and asset tools, and separate commercial systems. Databricks provides a governed data and AI foundation through which utilities can bring together data from this operational landscape. The next challenge is making the relationships across those domains reusable, so a forecasting model, an operations dashboard, and a commercial risk process can all draw on the same connected understanding of the grid, its customers, and its constraints.
Making the relationship available where the decision gets made
Kobai helps utilities connect feeder-to-substation-to-customer chains, line-span-to-circuit-to-criticality relationships, and generator-to-contract relationships as Shared Business Context on Databricks — available to inform the models, dashboards, and workflows that plan maintenance, prioritize vegetation crews, and evaluate curtailment decisions. (Note: the specific technical pattern for how this context is made available to each workflow is being validated with engineering and will be described precisely once confirmed.)
With that relationship available, a planned outage can draw on critical-facility context before crews are dispatched, rather than discovering it on site. Vegetation and fire-risk prioritization can draw on downstream criticality alongside the risk score itself. And a curtailment decision can draw on the commercial agreements it touches before it's made, rather than surfacing as a dispute afterward. (The curtailment/PPA scenario in particular should be validated with someone familiar with energy-sector commercial mechanics before publishing, to confirm it reflects realistic practice.)
|
Systems Reasoning Independently |
With Databricks + Kobai |
|
A planned outage is scheduled without visibility into affected critical facilities. |
Feeder, substation, and customer relationships are available before crews are dispatched. |
|
A fire-risk score doesn't account for what a line span feeds downstream. |
Line-span, circuit, and criticality relationships are available alongside the risk score. |
|
A curtailment decision is made without visibility into the contracts it may affect. |
Generator and contractual relationships are available to inform the decision. |
|
Relevant relationships are rediscovered manually, after an incident. |
Relevant relationships are available proactively, as part of routine operations. |
The last mile is a relationship, not a report
Each team's data and models were accurate on their own terms. The relationship that would have turned a correct forecast into a well-informed decision simply didn't exist anywhere both could draw on it.
The value of an accurate forecast isn't simply knowing what's about to happen. It's having enough business context to decide what to do next.