Utilities don't start most wildfires, but the ones they start happen at predictable places on predictable days. This is the full fusion your wildfire plan promises and your systems can't compute: fire-weather forecasts, km-scale wind physics, soil and terrain, fuel type, live fuel moisture from satellite, and LiDAR strike-tree analytics, blended into a per-span Fire Potential Index that runs four days ahead. Watch it drive three decisions in one red-flag event: the vegetation strike team cutting the 14 spans that matter (not the ones the trim cycle says are due), sensitive relay settings arming on exactly nine feeders as the index crosses threshold, and a surgical PSPS that de-energizes 4,100 customers instead of 38,000. Then the gust hits the limb, and nothing burns.
| Without | With GridCORTEX | Δ |
|---|
| Without | With GridCORTEX | Δ |
|---|
The stage is a fictional utility territory heading into four days of dangerous fire weather; the simulation clock runs 96 hours, Day 1 through Day 4. The stakes are simple: a wildfire started by utility equipment can mean burned homes, lawsuits, and years of liability, so one prevented ignition is worth the whole program. The opening picture is the seasonal risk map. It layers aerial laser scans (LiDAR) that have located 2,300 individual trees tall enough to strike a power line if they fall, satellite readings of vegetation dryness, soil and slope data, and vegetation types. The northeast hills are the danger zone: the brush there holds only 61% of the moisture level considered safe, and it is still drying. GridCORTEX blends all of it into one Fire Potential Index, a 0 to 100 risk score computed for every span of wire and projected four days ahead. The dashboard opens with the index at 31 and the small dead twigs on the ground at 9% moisture.
The first decision point comes at hour 3. The fused data ranks 14 spans of line as this week's real ignition risk; three of them were not scheduled for tree trimming for another two years. The software asks a human operator to approve pulling two contractor crews off routine trimming for 72 hours to cut those 14 spans, worst consequences first. A person decides because it means real crews and real budget. By the end of Day 1 six spans are cleared, including a leaning gray pine over the line called FDR-118 that the normal schedule would not reach until 2028; all 14 are cleared and verified by hour 26. At hour 30 a red flag warning arrives, the weather service's formal alert for dangerous fire weather: dry wind from inland, gusts forecast above 45 mph, humidity at 8%. The second decision point asks the operator to arm fast-trip settings on exactly 9 power lines: extra-sensitive breaker settings that cut power at the first flicker of trouble, with automatic re-energizing switched off so a line is never powered back into a fallen branch. Protection engineers review every setting before it goes live. The trade-off shows up at hour 40, when a squirrel touches line FDR-131 and 220 customers lose power for 40 minutes over a harmless contact the normal settings would have ridden through. That small outage is the honest price of the protection.
Day 3 is the peak. The wind model puts gusts of 47 to 48 mph through the Cedar Ridge canyons from 14:00 to 20:00, and the fire index crosses 90 on three hillside segments. The third decision point is the hardest call in wildfire operations: a Public Safety Power Shutoff, or PSPS, a planned blackout in which the utility turns lines off before the wind can knock them into dry fuel. Drawn with a blunt weather-zone boundary, the shutoff would darken 38,000 customers. GridCORTEX computes the boundary switch by switch instead and recommends turning off 4,100, keeping the Cedar Ridge community microgrid, a neighborhood of 610 homes with its own local power supply, lit, and contacting customers who depend on medical equipment first. A review board of people, not the software, owns that call. Then at hour 63 the gust arrives: 47 mph wind drops a limb across line FDR-115. The fast-trip settings cut power in 0.08 seconds and the line stays dead, so no spark reaches ground fuel that is at 4% moisture, dry enough for one spark to become a fire. A patrol finds the limb 22 minutes later. At hour 68 an hourly heat-sensing drone patrol spots a smolder in the hills, left by a lightning strike from last week. The control center is alerted in 40 seconds, the fire department gets exact coordinates, one line section in the fire's modeled path is switched off as a precaution, and the smolder is put out at 0.2 acres.
With all three approvals, the wind breaks after hour 76. The sensitive settings come off line by line as the index falls, the shutoff areas are patrolled and re-energized, and every customer is back by 06:00 on Day 4. The wrap at hour 90 is the whole story: zero utility-caused ignitions, 4,100 customers dark for 19 hours instead of 38,000 for 24, one 40-minute nuisance outage, and 14 dangerous trees gone for good, with the reasoning behind every decision recorded. Decline the recommendations and the same gust starts a wind-driven fire that reaches 1,850 acres and destroys 11 buildings by Day 4.
Every input exists, but nothing combines them. The 14 dangerous spans stay on the routine 2027-2028 trimming schedule, because the schedule ranks work by calendar date, not by risk. Breakers keep their normal settings everywhere, because nobody can say which lines need the sensitive ones. The shutoff debate, facing only a blunt 38,000-customer outline with schools and a hospital inside, stalls at "monitor conditions," because the cost of the blackout is obvious and the risk is a guess. At hour 63 the limb drops across line FDR-115, and the breaker automatically re-powers the line twice, straight into the fallen branch, throwing sparks into tinder-dry fuel on a 47 mph wind.
The camera network confirms a fire 9 minutes later, moving toward Cedar Ridge; evacuations begin. With no drone patrols, the lightning smolder in the hills is found by a 911 caller three hours later, a second fire front on a day that already has one. By Day 4 the tally is 1,850 acres and 11 buildings, and the cause investigation opens on the breaker's own records. The utility's data becomes the other side's best evidence in court.
The software fuses seven data sources into one fire risk score for every span of wire, four days ahead, and turns the score into three recommendations that people approve: cut the 14 spans that matter most, ranked by risk instead of by calendar; arm the extra-sensitive breaker settings on exactly 9 lines, with automatic re-powering blocked; and shrink the safety shutoff to a surgical 4,100 customers instead of 38,000, with the neighborhood microgrid kept lit. Every gate has a person in charge: managers approve the crews, protection engineers approve the settings, and a review board owns the shutoff.
The payoff: the limb strike is cut off in 0.08 seconds with no re-powering and no fire, the reliability cost is one 220-customer nuisance outage, the drone patrols catch a lightning smolder at 0.2 acres, and the event closes with zero utility ignitions and an automatically kept record of the justification behind every setting, cut, and boundary.
| Measure | Without GridCORTEX | With GridCORTEX | The difference |
|---|---|---|---|
| Utility ignitionsfires started by the utility's own equipment, the number the whole program exists to keep at zero | 1; wind-driven, 1,850 acres | 0 | the whole ballgame |
| The FDR-115 limb strikewhat happened when the gust dropped a tree limb onto the power line named FDR-115 | breaker re-powered the fallen line twice | power cut in 0.08 seconds, stayed off | armed by the index |
| Strike-tree spansthe 14 stretches of line where a tree tall enough to hit the wire stood over dry fuel | waiting on the 2027-28 schedule | cut 3 weeks early | ranked by risk, not calendar |
| PSPS scopePSPS is a Public Safety Power Shutoff, a planned blackout ahead of dangerous wind; scope is how many customers it darkens | 38,000 blunt (never called) | 4,100 surgical + microgrid lit | 89% fewer customers dark |
| PSPS customers dark (headline)the shutoff as the public experienced it | 0; then evacuations | 4,100 × 19 hrs | risk retired, lights mostly on |
| Nuisance-trip costthe price of extra-sensitive settings: harmless contacts, like a squirrel, now cause small outages | n/a | 1 outage, 220 customers, 40 min | the honest price |
| Detection to patrolhow quickly someone laid eyes on the spot where the limb hit the line | camera at +9 min (a fire) | patrol at +22 min (no fire) | crews staged in advance |
| Lightning smolder in the hillsa small fire left by last week's lightning, nothing to do with the utility's lines, but burning in its territory | found by a 911 caller, 3 hours later | drone heat camera; out at 0.2 acres | hourly patrols + lines pre-armed |
| Event outcomehow the four days end | burned buildings, evacuations, lawsuits | a logbook entry | everything |
| Liability recordwhat the utility can show in court about the choices it made | breaker record is the plaintiff's best evidence | every setting justified, traced | defensible |
| Veg budget effectwhat the tree-trimming budget actually bought | 80% spent on low-risk spans | every crew-hour aimed at real risk | same money, real risk cut |
| Reliability cost of fire opshow much everyday reliability was given up to run in fire-safe mode, which causes more small outages | sensitive settings all season long | 9 lines, 52 hours, only while the index demanded it | surgical |
| WMP / regulator filingthe WMP is the Wildfire Mitigation Plan, the safety plan the utility must file with its regulator every year | written narratives | computed evidence, every step logged | writes itself |
| Community trusthow customers feel about the shutoff once it is over | "why was my power off?" | microgrid lit, notice early | earned |
The exposure that changes is patrol exposure during wind events. Instead of patrolling a whole district because the district sits under a warning, crews patrol the circuits the model ranks highest, which cuts driving in high wind and after dark and cuts the operating posture changes that put people in the field at the worst hours.
Counted in units you already track:
The duty officer's daily assembly hours come back and the wildfire team's inspection targeting is produced rather than debated, while the risk call itself stays with the duty officer.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Duty officer and analyst labor | assembly hours avoided x your loaded rates x fire season days x cycles per day |
| Targeted patrol labor | patrol hours avoided by patrolling ranked circuits instead of whole districts x your loaded patroller rate x events per season |
| Vehicle and travel | patrol miles avoided x your fleet cost per mile |
| Avoided PSPS scope | customers not de-energized because their circuit scored below threshold x hours of the event x 60 x your value per customer minute. This one counts only if you actually use the twin to inform the call, and the call stays yours |
| Avoided ignition | your own modeled cost per ignition event x the share you believe circuit level targeting would prevent, which is your number and one you should keep conservative |
You pay for the weather modeling and the twin, for integration to your vegetation findings, conductor condition records, GIS, and outage history, and for your team's time validating the score against actual near misses and ignitions over at least one season. The quality of your conductor and vegetation records will decide how much of year one goes to data cleanup.
The payback a reviewer will accept is driven by duty officer and patrol labor and by PSPS customer minutes avoided. Avoided ignition is the largest number and the least auditable, so keep it out of the base case.
Part of this benefit is indirect and it is fair to say so. The direct piece is real: routine clearance measurement puts a forester under energized conductor, in brush, on ROW access roads and highway shoulders, for hours at a time, and a measurement taken from the point cloud does not. The indirect piece is that violations found sooner are trimmed as scheduled work rather than as an emergency response after a contact.
Counted in units you already track:
Forester field hours and GIS data entry hours come back to the vegetation program, and the forester's remaining field time goes to verifying flagged spans and talking to landowners instead of measuring spans that turn out to be fine.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Forester field labor | ground survey hours avoided x your loaded forester rate, plus miles avoided x your fleet cost per mile |
| Data processing | your current service bureau processing fee per mile or per square mile x the volume you process per year, compared against processing the same volume in house |
| GIS data entry | records keyed per year x minutes each x your loaded GIS analyst rate |
| Rework and mobilization | crew mobilizations sent against a stale span list x your contractor mobilization cost, which is the cost of finding the tree already gone or the wrong span flagged |
| Avoided outage | your own cost per tree caused distribution outage x the share of them you believe a current inventory would have caught in time, which you set, not us |
The dominant cost is data acquisition, the flights or the satellite tasking, and you are largely paying that today. On top of it you pay for the GPU processing capacity or the hosted service, for the GridCORTEX analytics layer, for integration into GIS and the work management system, and for your program staff to validate the first inventory against known ground truth spans before anyone plans a season from it.
Payback is driven by processing turnaround and ground survey labor, not by avoided outages, because acquisition cost stays roughly flat either way. If the pilot only proves you can refresh the inventory in the same season you captured it, that alone is usually the case.
The public safety effect is the point: fewer utility ignitions from faults that fire mode is designed to catch. There is a workforce exposure effect too, and it comes from a mechanism people miss. Because fire mode blocks reclosing, every trip becomes a locked out circuit that has to be patrolled end to end before it can be re energized, often at night in bad conditions. Tighter zones and validated coordination mean fewer devices lock out for a given fault, so fewer circuit miles get patrolled.
Counted in units you already track:
Scarce protection engineering hours come back, and the engineer moves from producing settings to judging the ones that were flagged, which is the correct place for that expertise.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Protection engineering labor | engineer hours avoided x your loaded protection engineer rate, and separately x your outside consultant rate for the portion you contract out, which is usually the higher of the two |
| Patrol labor and travel | circuit miles not patrolled after avoidable lockouts x your patrol hours per mile x loaded crew rate, plus miles x your fleet cost per mile |
| Design basis documentation | hours to assemble the documented rationale per zone x your loaded rate or your consultant rate x zones per cycle |
| Customer interruption from fire mode | customers de energized by fire mode lockouts that better coordination avoids x hours out x your own value of customer interruption |
| Avoided ignition | your own modeled cost per utility ignition x the share you attribute to faults fire mode settings would have caught, which you set, not us |
You pay for the scoped engagement that builds and runs this and for integration into your protection settings management system, but the cost that decides whether this works is network model quality. Your model has to be accurate and complete, with current load, fault duty, and device data, and closing those gaps is real engineering time. Your protection engineers also review and approve every setting group, and that review does not shrink much, because it should not.
Payback here is usually measured in districts completed per season rather than in dollars, because protection engineering capacity is the binding constraint and you cannot buy more of it quickly. If you want a dollar figure, use the consultant rate you would otherwise pay for the same design volume.
The safety change is mostly in the restoration tail. Every de-energized mile must be patrolled before it is re-energized, so a smaller scope is directly fewer patrol miles, fewer switching operations, and fewer field entries during and after a high wind event. There is also the exposure a de-energization creates for customers running backup generation, which a narrower scope reduces.
Counted in units you already track:
Engineering and restoration patrol hours come back, and the PSPS decision authority gets scope options with their risk and customer impact already priced instead of waiting on an engineer to build one.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Customer minutes | customers spared x hours de-energized x 60 x whatever value your commission or your own case places on a customer minute. This is the dominant driver and it is your number, not ours |
| Restoration patrol labor | patrol hours avoided on miles never de-energized x your loaded patroller rate |
| Engineering labor | scope building hours avoided per event x your loaded planning engineer rate x events per season |
| Event support cost | reduction in customer notification volume, community resource center staffing, and PSPS related claims x your own per event cost for each |
| Switching labor | switching operations avoided x hours per operation x your loaded crew rate |
You pay for the modeling service, for the ADMS and GIS integration that lets a proposed scope be expressed as a real switching plan, and for a serious amount of your own engineering and control room time validating proposed scopes against plans your operators would actually execute. Expect that validation to run a full season before anyone signs off on a narrower scope during a live event.
Payback is dominated by customer minutes not interrupted, and everything else is rounding. That makes the whole case rest on your own value per customer minute, so settle that number internally before you model anything else.
The exposure removed is a crew driving toward a smoke column to find out what it is, in wind, in poor visibility, often at night, on the same roads the fire agencies need. Faster and more precise de energization also narrows the footprint that has to be patrolled and re energized afterward.
Counted in units you already track:
Duty officer watch hours and confirmation patrol hours come back to wildfire operations, and the duty officer stops scanning and starts adjudicating alerts, which is the part only a person should do.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Watch labor | camera watch hours avoided x your loaded duty officer rate x red flag days per year |
| Confirmation truck rolls | confirmation trips avoided x your loaded crew rate per trip, plus miles x your fleet cost per mile |
| PSPS scope | customers kept energized because scope narrowed x hours de energized x your own cost per customer hour of PSPS interruption, using whatever value you already apply in your PSPS cost benefit work |
| Patrol and re energization | circuit miles not patrolled because the de energized footprint is smaller x your patrol hours per mile x loaded crew rate |
| Avoided fire loss | your own modeled cost per grid ignition that reaches reportable size x the share you believe seconds level detection would have kept small, which you set, not us |
The dominant cost is the field build, poles or mounts, power, and backhaul to the camera sites, not the inference. On top of that you pay for the edge devices, the GridCORTEX detection layer, integration into your ADMS switching queue, and duty officer time through the first fire season tuning thresholds. Be ready for false positives early. The honest way to run that period is to keep every alert human adjudicated and measure how the false positive rate falls, rather than promising it starts low.
Payback is dominated by narrowed PSPS scope and by avoided fire loss, and only the first of those is auditable from your own records, so build the case on PSPS customer minutes and watch labor and treat avoided loss as upside.
The two highest consequence exposures in line inspection are low altitude manned flight and climbing energized structures. Both are replaced for routine condition assessment, and the climber only goes up once the defect is already confirmed and the repair is scoped.
Counted in units you already track:
Patrol hours and desk triage hours come back to the line department, and the inspection engineer stops reading every image and starts reading only the flagged ones.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Patrol labor | patrol hours avoided x your loaded patroller rate |
| Aviation | manned inspection hours avoided x your all-in helicopter hourly cost, whether owned or contracted |
| Engineering triage | triage hours avoided x your loaded engineer rate |
| Avoided failure | your own cost per unplanned transmission outage x the share of failures you believe earlier detection would have caught, which you set, not us |
| Vehicle and travel | road miles avoided x your fleet cost per mile, plus per diem on remote patrols |
You pay for the flight operations, whether you own the fleet or contract it, for the GridCORTEX intelligence layer that turns imagery into ranked findings, for the integration into your asset and work systems, and for your own staff time to validate detections during the first inspection season. The intelligence layer is the smaller line item. The flight operations dominate.
Payback is usually driven by aviation and patrol labor, not by avoided failures, because avoided failure is the number you will trust least. Build the case on the two you can audit and treat avoided failure as upside.
What is this, exactly? It is AI software: intelligent agents and models built and delivered by SoftServe, running on NVIDIA accelerated computing. It is not a hardware appliance and it does not replace the systems you run today. It deploys in your own cloud or on your premises, connects read-only to your existing systems, and recommends; your people approve every action, starting in shadow mode until it earns trust.
A continuously updated risk model for wildfire and operations leadership: a circuit-and-span-level ignition risk map, refreshed each forecast cycle, that feeds PSPS calls and inspection targeting. The demo above uses synthetic data; everything below describes what the real deployment needs from your organization.
| Your system | Typical products | How we connect |
|---|---|---|
| Weather and environment | National Weather Service feeds, commercial forecast services, satellite and LiDAR imagery | read-only API |
| Geographic Information System (GIS) | Esri ArcGIS Utility Network, GE Smallworld | scheduled file export (CSV or CIM XML) |
| SCADA historian | AVEVA PI System, GE Proficy | historian mirror (one-way feed) |
| Asset / work management (EAM/CMMS) | IBM Maximo, SAP PM, Oracle WAM | database replica refreshed nightly |
| Outage Management System (OMS) | GE PowerOn, Oracle NMS, ADMS outage module | database replica refreshed nightly |
| Planning and study tools | CYME, Synergi, WindMil | scheduled file export (CSV or CIM XML) |
| Protection settings management | ASPEN OneLiner, CAPE, relay settings databases | scheduled file export (CSV or CIM XML) |
Runs in your cloud account on GPU instances to handle weather model downscaling, with an on-premises option; all feeds are read-only with no connection to control systems, and PSPS decisions stay with your wildfire leadership. The 30-day pilot runs in shadow against actual conditions.
The Approve button you just clicked in the demo above is the real workflow. This is what it looks like on the screen of the wildfire duty officer in the GridCORTEX console:
Approve publishes the accepted forecast to the PSPS planning workspace and creates draft inspection tasks in your work management system in pending status; your wildfire team confirms and assigns them. PSPS calls remain with the decision authority; the twin only informs them.
The trigger is automatic from weather, fuel, and asset feeds; the duty officer enters nothing. Field observations logged in the work system feed back into the score.
Rescored each forecast cycle, with weather every 15 minutes and SCADA continuously; every circuit card shows the forecast run time it came from.
Lives as a console risk map refreshed each forecast cycle; mobile push when a circuit crosses the threshold. The console runs in a browser beside your existing screens on day one; embedding into your own systems is a roadmap step once the read-only phase has earned trust. Approve, Modify, and Decline are all captured in an audit trail your compliance team can pull, and GridCORTEX never blocks or overrides anything in the systems you run today.
The fair question from any wildfire officer: "We buy fire-weather forecasts, we fly LiDAR, we have a wildfire mitigation plan and PSPS criteria, what's new here?" Here's the honest answer.
When someone asks "what did it actually calculate?", this is the list. In the simulation these factors drive the storyline; in a pilot they are computed from your LiDAR, weather, fuel, and protection data.
Presenter's one-liner: "Seven data sources nobody had ever multiplied together became one number per span, four days ahead. The number sent the tree crews to fourteen spans, armed fast-trip on nine feeders, and drew a PSPS boundary around four thousand customers instead of thirty-eight. Then the 47-mile-an-hour gust put the limb on the wire, the feeder cleared in 80 milliseconds, nothing reclosed into it, and nothing burned. That's ignition risk as engineering instead of dread."