A retail energy provider in a deregulated, ERCOT-style market, BrightPlain Energy, fictional, 240,000 customers, heads into a heat-dome week. The hedge book covers 1,560 MW; the AI forecast says the real peak is 1,840 MW. That 280 MW gap, priced at a $5,000/MWh cap, is the difference between a good week and a company-ending one. Watch the desk close the position before the market prices the heat, activate a churn-safe demand response portfolio, and ride the scarcity hour with zero open exposure, while the competitor's customers get the panic email.
| Without | With GridCORTEX | Δ |
|---|
| Without | With GridCORTEX | Δ |
|---|
BrightPlain Energy is a fictional retail energy provider, a company that buys power on the wholesale market and sells it to 240,000 homes and businesses. It operates in a Texas-style deregulated market, where retailers compete for customers and wholesale prices can legally spike to a cap of $5,000 per megawatt-hour, roughly a hundred times a normal price. The simulation runs one extreme-heat week, Monday 06:00 through Friday midnight. The Monday position review sets the stakes. The company has bought 1,560 megawatts of power in advance to cover what its customers should use: 1,200 megawatts of steady around-the-clock supply plus 360 shaped to the daily usage curve. The forecast vendor's weekly email says the week's peak demand will be 1,610 megawatts. The GridCORTEX forecast, built meter by meter from smart-meter data and street-level weather under the building heat dome, says Thursday peaks at 1,840 megawatts, and with 90% confidence no worse than 1,905, the hottest Thursday in nine years. The difference is a 280 megawatt gap with no power bought to cover it, and at the $5,000 cap a single uncovered peak hour costs $1.4 million.
Three decision points follow, each executed by people: the trading desk makes every trade, the portfolio team runs every activation. Monday morning, close the gap now: the heat dome is visible in weather models but not yet in market prices, so power for Thursday and Friday afternoons (13:00 to 21:00) still sells at a $142 per megawatt-hour average. The desk buys 240 megawatts for $0.68M, cutting the uncovered gap from 280 to 40 megawatts. By Tuesday the same block trades at $310, and Monday's purchase is already $1.1M ahead. Wednesday, activate the customer portfolio: 92,000 enrolled smart thermostats (pre-cooling homes, then easing them 2 degrees, with bill credits sized per customer group) plus contracts with large business customers to cut usage on request. Together they take 118 megawatts off the Thursday-Friday peak at an equivalent cost of $71 per megawatt-hour, designed so customers stay comfortable and stay customers. Thursday at 15:40, with the live price at $4,900, the scarcity-hour call: the last 22 uncovered megawatts can be bought at essentially the cap (about $0.33M), or covered by the company's own coordinated fleet of customer batteries and precision usage cuts, thousands of small devices acting together like one power plant, for about $0.02M. The fleet is dispatched; zero megawatts are bought at the cap.
The ending, with all three approvals: Thursday's 16:00 to 19:00 scarcity window and Friday's second extreme afternoon pass with zero uncovered demand, every scarcity hour supplied below an effective $150 per megawatt-hour, week profit at $1.9M above plan, customer losses held to 0.4%, retention offers accepted by 61% of the at-risk customers they reached, and the risk committee receiving a complete decision log on Monday, recorded step by step in Relay, the built-in audit trail. Ignore the recommendations and the same weather produces a $21.4M loss: the desk discovers the gap live on Thursday, buys 180 megawatts at a $4,100 average while the rest rides the cap, and a $14M collateral demand from the market's clearing house arrives Friday morning.
The vendor forecast lands in one inbox and the record of what power the company has bought lives in the trading system, so the 280 megawatt gap exists for days before anyone computes it. Nothing is bought while the price is low; by Tuesday the block costs $310 and the desk is still watching. On Thursday at 14:00, with live prices at $3,800 and rising, the desk buys what it can, 180 megawatts at an average $4,100, and the rest rides the cap through 10 exposed scarcity hours at an average supply cost of $2,340 per megawatt-hour, against retail prices set assuming a fraction of that. The market's clearing house demands $14M of additional collateral by Friday 10:00. Bill-shock projections leak onto social media, the retention call center queue hits 4,000 calls, and 2.8% of all customers leave, because frightened customers shop around. Week result: a $21.4M loss.
The software compares the meter-level demand forecast against the company's power purchases continuously (use case UC 21.2), so the gap is visible Monday at 07:00, three days before the market prices in the heat, and closing it costs $142 instead of $4,100 per megawatt-hour, 29 times cheaper. The device-fleet coordinator (UC 6.4) turns enrolled thermostats and batteries into the cheapest peak-hour resource the company owns, 140 megawatts of peak reduction. And the customer-loss model (UC 21.1) runs during the event, replacing the mass panic email with targeted credits that 61% of recipients accept. People stay in charge: the desk executes every trade and the portfolio team owns every activation. The week ends $1.9M above plan, a $23.3M swing, with 0.4% customer losses and no collateral demand.
| KPI | Without GridCORTEX | With GridCORTEX | Delta |
|---|---|---|---|
| Open position into Thu peakexpected customer demand with no power bought to cover it; every uncovered megawatt rides the spike | 280 MW | 0 MW | fully covered |
| Gap detectedwhen the company learned its purchases would not cover the heat wave | Thursday, live | Monday 07:00 | 3 days of warning |
| Forward cover costthe price paid to buy the missing power, in advance versus in a panic | 180 MW @ $4,100 avg | 240 MW @ $142 | 29× cheaper |
| DR / VPP contributionDR (demand response) pays customers to use less at peak; a VPP (virtual power plant) is thousands of small customer devices coordinated by software to act like one power plant | none | 140 MW peak | the customer base acts as a power plant |
| Scarcity hours exposedhours of cap-level prices faced with demand still uncovered | 10 hrs | 0 hrs | −10 |
| Effective supply cost (Thu-Fri peak)the average cost of every megawatt-hour supplied to customers during the peak | $2,340/MWh | $147/MWh | −94% |
| Forecast basiswhat the demand forecast is built from; AMI means the smart meters on every home | vendor weekly email | AMI + km-scale weather | meter-level detail |
| Week P&L vs planthe week's profit and loss against budget | −$21.4M | +$1.9M | +$23.3M |
| Margin / collateral callcash the market's clearing house demands upfront to cover a losing trading position | $14M scramble | none | treasury slept |
| Customer churn (week)the share of the 240,000 customers who left during the week | 2.8% of book | 0.4% | ~5,800 customers kept |
| Save-desk call volumecalls into the retention center from customers threatening to leave | 4,000 calls | baseline | no panic |
| Retention offer conversionthe share of at-risk customers who accepted a targeted credit and stayed | n/a, blast email | 61% targeted | targeting beat the mass email |
| Risk committee packagethe decision record the risk committee reviews after the event | reconstructed in Excel | Relay-traced decision log | the audit trail is built in |
| Next scarcity event posturehow ready the company is for the next price spike | same blindness | same system, better priors | each event makes it smarter |
The safety effect is indirect. A VPP that actually delivers what it promised on a peak day reduces the emergency operations that follow when it does not: manual load transfers, callouts, and field staff working a hot afternoon into the night.
Counted in units you already track:
Event day hours come back to the program director and the monthly reporting week comes back to the program analyst.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Capacity value delivered | incremental megawatts delivered against commitment x your capacity price or your own avoided capacity cost |
| Program labor | event and reporting hours avoided x your loaded program staff rate |
| Underdelivery exposure | your penalty or shortfall charge per megawatt x the shortfall megawatts you currently incur in a typical season |
| Market revenue | megawatt hours bid into energy or ancillary products x the settled price in your market, for the hours the portfolio was previously idle |
| Incentive efficiency | your incentive payment per enrolled device x the devices you no longer need to call because the dispatch is better ordered |
You pay for the orchestration platform, for an integration to each vendor DERMS dispatch interface, and for the contract work to get the vendors to expose those interfaces at all, which is often the slow part. Your program staff also need time to validate the combined forecast against a season of real events before they will offer against it.
Payback is usually dominated by the capacity value of delivering the commitment plus the penalty exposure you stop carrying. Program labor is real but it is the smaller line.
There is no field exposure in this use case and we will not manufacture one. This is a balance sheet risk case. The one genuine human exposure it touches is the risk desk working continuously through an extreme weather event, which your own timesheets already show.
Counted in units you already track:
Rebuild and stress test hours come back to the load analysts, and the open position becomes a daily product instead of a weekly project.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Analyst labor | rebuild, stress test and reconciliation hours avoided x your loaded load analyst and risk manager rates |
| Imbalance exposure | megawatt hours of forecast error reduced x your own realized imbalance price spread, taken from your settlement statements and not from any market study we run |
| Scarcity tail | the change in your tail loss, measured by backtesting your own positions against your own settled prices, against your own board approved loss tolerance |
| Hedge efficiency | premium paid for block hedges a shaped position would not have bought, priced at the costs you actually executed |
| Collateral | margin or collateral you can stop posting because the position is better shaped x your own cost of capital |
You pay for the scoped engagement that builds and runs this, for weather ensemble data, and for integration into your energy trading and risk management system and your settlement data feeds, plus your risk manager's review time. The desk still owns every trade: the recommendation arrives as a draft ticket, and GridCORTEX holds no market access and executes nothing.
Payback is dominated by imbalance exposure and tail loss, both computed from your own settlement statements, and the analyst labor line is real but small. Backtest on your own worst month before you size the program, because that is the only number your board will believe.
This is a software product for a retailer that owns no wires, so there is no field exposure to remove and we will not invent one. The only honest physical connection runs through arrears: customers on a plan that fits their usage fall behind less often, and fewer disconnect for nonpayment requests means fewer trips by the wires company's technician to a meter, which is a visit that carries real confrontation risk. That benefit lands on the utility, not on you.
Counted in units you already track:
Analyst and save desk hours come back, and the retention team stops building lists and starts running campaigns.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Retained gross margin | accounts saved above your current baseline x your gross margin per account per month x your expected remaining tenure in months |
| Avoided acquisition | accounts saved x your all in cost to acquire a replacement customer |
| Discount efficiency | your average retention discount x the number of safe renewals that no longer receive one, which is the line that usually pays for the whole thing |
| Analyst and save desk labor | list building hours, churn reporting hours, and save desk minutes avoided x your loaded rates |
| Bad debt | your write off rate x the balance that no longer ages on accounts moved onto a plan that fits their usage, using the share you attribute to plan mismatch |
You pay for the GridCORTEX predictive service, for integration to your billing system, interval data feed, customer relationship system, and market price data, and for your own analyst time to define margin correctly and run the backtest on your own book before anyone acts on a score. If you cannot state gross margin per account today, that is the first project, not this one.
Payback is usually carried by discount efficiency, meaning the offers you stop giving to accounts that were never leaving, rather than by saved accounts, because saves are the number your finance group will argue about. Prove it with a holdout control group, not with a before and after comparison.
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 forecasting and risk service for a competitive retailer's supply and risk desk. It produces cohort-level load forecasts from weather ensembles, projects the book with churn adjustments, measures the open position against hedges daily, recommends shaped hedges, and runs extreme-event stress tests, as a daily position report and recommendations. The demo above uses synthetic data; everything below describes what the real deployment needs from your organization.
| Your system | Typical products | How we connect |
|---|---|---|
| Energy trading and risk (ETRM) | ION Allegro, openlink Endur | scheduled file export (CSV or CIM XML) |
| Customer Information System (CIS) / retail billing | retail billing platforms | database replica refreshed nightly |
| Market and grid operator interfaces | ERCOT or other ISO settlement data | read-only API |
| Weather and environment | National Weather Service feeds, commercial services | read-only API |
| Customer relationship and service systems | Salesforce, Zendesk | database replica refreshed nightly |
| Retail market transactions | EDI switch, drop, and enrollment records | scheduled file export (CSV or CIM XML) |
| Metering (AMI usage data) | Smart Meter Texas or utility-provided interval usage | scheduled file export (CSV or CIM XML) |
Runs in the retailer's own cloud account with GPU instances. Position and book data is commercially sensitive and stays in your environment with need-to-know access. All connections are read-only; the desk executes every trade through its existing brokers.
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 retailer's supply and risk desk lead in the GridCORTEX console:
Approve posts the recommendation as a draft ticket in the trading and risk system; the desk sizes, executes, or declines it under its own limits and controls. GridCORTEX never trades and holds no market access.
Position measurement is automatic from trading, billing, and market feeds; the desk types nothing. A deal done by phone flows in once the desk books it in the trading system.
The book and hedge position net daily against overnight billing and trading system syncs, with weather ensembles each cycle; every report shows the as-of timestamp of its positions and forecasts.
Lives as a daily position report in the GridCORTEX console beside the trading system; a position beyond limits pushes a mobile notification to the risk officer. 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 REP: "We have an ETRM, a forecasting vendor, and traders who lived through the last one, 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 meter data, book, and market.
Presenter's one-liner: "The forecast saw the heat three days before the market priced it, the desk closed 280 megawatts while it was cheap, the thermostat fleet became the cheapest peaker in the portfolio, and the churn model protected the book while the competitor blasted panic emails. Same weather, plus twenty-three million dollars. That's what you just watched."