AI for energy and utilities
AI for energy and utilities
Trim where the risk is, pre-stage crews before the storm, forecast the load that is coming. 5x ROI in 30 days, or we work free.
- Distribution utilities
- Utility contractors
- Renewables operators
- Energy services firms
- Solar and wind O&M
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Proof, honestly
No published utility case study yet, and we say so. Founding clients get priority scheduling, direct founder involvement, and a co-published case study once the numbers are real. The 5x ROI guarantee applies from day one.
We guarantee 5x ROI inside 30 days of deployment, in writing, measured against a baseline you sign before we build. If the system misses the bar, we keep working for free until it clears.
Energy & utilities, before and after
The manual path is dashed: Trim cycles run on the calendar, Crews staged on judgment, Forecasts trail demand. The system path replaces it, and a person approves before anything ships: Trim budget follows risk, Pre-staging from live signals, Load models that keep up.
Before: by hand
- 01Trim cycles run on the calendarhuman
- 02Crews staged on judgmenthuman
- 03Forecasts trail demandhuman
After: the system
- 01Trim budget follows risk
- 02Pre-staging from live signals
- 03Your approvalhuman
- 04Load models that keep up
The research
North American summer peak demand is forecast to rise by over 224 GW in the next 10 years, 69% above the previous year's projection, with new data centers driving most of the increase (NERC, 2026, a planner forecast).
North American Electric Reliability Corporation · 2026 · Load-serving entity and system planner forecasts submitted to NERC in mid-2025, covering the contiguous US, Canada and northern Baja California
Vegetation drives your outage minutes.
Trim cycles run on the calendar, not on risk, so the same span gets cut on schedule while the strike tree three feeders over waits its turn. US customers averaged about two hours a year of interruptions outside major storm events in 2024, and 11 hours with them (EIA, 2025, US national average). The regulator counts every one.
The pre-stage call is made with yesterday's weather.
Guess low and restoration runs days while the commission watches. Guess high and you pay standby for crews that watched it rain. Either way the call was made from a forecast map and memory.
Load forecasts built for a different decade.
Data-center and electrification demand moved faster than the planning models did, and the interconnection queue is full of load your forecast never saw coming.
The contractor flew the line in April. It is October.
The photos landed in a shared drive, nobody scored them, and the defect they caught is still on the pole. The registry never heard about the flight.
Filing season is a data hunt.
The interruption records live in the OMS, the cause codes in spreadsheets, and the explanations in whoever worked the storm. Assembly takes weeks and the signer inherits the risk.
The warranty clock outruns the paperwork.
The turbine tripped on the 4th. The month-end report explains it on the 30th, and by then the OEM claim window has burned most of a month on a fault nobody wrote up.
Run your numbers.
Your operations
Savings use the low end of our hours-reclaimed range. The math is conservative on purpose.
The math
Calculated at the low end of every range.
Every first build is covered in writing: 5x ROI in 30 days. Or we work for free.
The numbers in energy and utilities
North American summer peak demand is forecast to rise by over 224 GW in the next 10 years, 69% above the previous year's projection, with new data centers driving most of the increase (NERC, 2026, a planner forecast).
North American Electric Reliability Corporation · 2026 · Load-serving entity and system planner forecasts submitted to NERC in mid-2025, covering the contiguous US, Canada and northern Baja California
CenterPoint Energy cut vegetation-related outage minutes 50% year over year with a LiDAR-based AI digital twin (T&D World, 2025).
CenterPoint Energy (via T&D World) · 2025
88% of agentic AI pilots never reach production. That is why every build here gates on scope and a signed baseline (IDC, 2026).
IDC · 2026
Who this is built for
You know which feeders will fail first
The storm plan used to be a map, a forecast and the memory of the last blow. Now the outage model reads the forecast track against feeder history and asset condition, and hands your duty supervisor a ranked pre-stage proposal. He still makes the call. He just makes it with evidence, and the standby bill is a decision instead of a coin flip.
The worst span gets cut first
The trim cycle treats every span the same. The outage record does not. Scoring the LiDAR and the outage history by span sends the budget to the strike trees and the fast-growth corridors first, and the defence at the budget table becomes a ranked list with sources instead of a binder and a feeling.
The reliability filing shows its work
Filing season used to mean weeks of pulling interruption records out of the OMS and cause codes out of spreadsheets. The draft tables now assemble continuously, every minute traced to its source record, and your name goes on a filing you can defend line by line. The signature stays yours. That is the design, not a limitation.
The downtime story is written the same day
A turbine trips on the 4th and the explanation used to surface in the month-end report, after the warranty window had burned three weeks. Now the trip is flagged with its fault code the day it happens, and the lender report assembles from numbers that were checked while they were fresh, whether they live in Power Factors or a settlement export.
What stays human
- She walks into the budget call with the span ranking
- The duty supervisor owns the crew plan
- She throws out the cedar shadow
The steps the day below leaves to a person, by design.
A storm watch lands on trim week
She walks into the budget call with the span ranking
humanThe quarterly program review opens at 9am and she brings one list: every span in the district scored from the last LiDAR flight and five years of outage history, worst first, each score citing the flight date and the feeder. The trim miles she asks for are the ones the list defends.
A storm watch reroutes the afternoon
Environment Canada posts a wind warning for Friday night. The model reruns the outage forecast against the track, the feeder histories and asset age, and drafts a pre-stage proposal, which depots, which feeders, listed by expected trouble.
The duty supervisor owns the crew plan
humanShe takes the proposal down the hall. The duty supervisor moves one crew because a bridge closure makes the proposal's depot unreachable, approves the rest, and the plan is his. The model suggested. Nothing staged itself.
The day's flight lands scored
The contractor's drone photos from the morning patrol arrive. Vision scoring flags encroachments ahead of cycle and writes each one into the asset registry with the photo, the span id and the confidence attached, ranked by consequence.
She throws out the cedar shadow
humanOne flag is a shadow across the conductor, not a limb. She rejects it, and the rejection lands in the grading set so the false-alarm rate stays a number her general foreman can watch, not a feeling. Two real encroachments move up the trim schedule in Clearion.
The draft filing takes the day's outage minutes
The month's interruption records assemble toward the reliability filing, every outage minute carrying its OMS record and cause code. The tables wait as drafts for the named signer. Nothing files itself.
Friday's storm will have a paper trail
Every ranking, proposal and approval from today is logged. When the commission asks after the storm why crews were where they were, the answer is a record, not a reconstruction.
What we build for energy and utilities
Aim the crews
Vegetation-risk prioritization.
LiDAR and imagery scored by ignition and outage risk, so the trim budget goes where the risk actually is.
Predictive outage analytics.
Weather, asset condition and outage history combined, so crews pre-stage on signal instead of hunch.
Wildfire risk flagging.
Imagery and line-sensor data watched for the conditions that precede ignitions, flagged for your engineers.
Keep the record straight
Asset inspection document automation.
Field reports and photos structured straight into the asset registry, instead of dying in folders.
Regulatory reporting prep.
Reliability data assembled and checked. Your team reviews and files, and a named person signs.
Availability and production reporting.
Metered production assembled into the monthly lender and offtaker reports, every number traced to its source, the operator signs and sends.
See the load coming
Load forecasting for new demand.
Forecast models that account for the data-center and electrification demand already in your queue.
Where the data comes from
Asset registry
The GIS and the maintenance system hold the physical truth: spans, poles, transformers and feeders in Esri ArcGIS, work and condition history in IBM Maximo or SAP PM. Every score the system produces lands here as a structured record against a span id, which is what makes the ranking auditable instead of a heat map nobody can trace.
Where it stops. The registry stays the system of record under your permissions. Writes are structured inspection results and ranked work proposals a person can trace to their source, never control data, and critical infrastructure information stays inside your environment.
Inspection media
The LiDAR flights, drone photos, thermal shots and field photos your crews and contractors already produce. Today they are the most expensive data you own and the least used, scored once by a tired reviewer or not at all. Vision models read every frame and return flags with the span id and the photo attached.
Where it stops. Media stays in your storage and moves under no-training API terms. Detection is graded against defects your crews verified in the field, with misses and false alarms reported separately, and a flag is a proposal for an engineer, never a work order.
Outage systems
The OMS, whether that is Survalent or Milsoft or the utility's own, holds the interruption records, cause codes and restoration times that the storm model and the reliability filing both depend on. The system reads them to learn which feeders fail under which weather, and to keep the filing tables current.
Where it stops. Read only, always. Nothing dispatches a crew, opens a switch or touches SCADA. The OMS feeds the analysis, the analysis feeds drafts, and the drafts wait for people. The line between the data side and the control side is the design.
Built around your rules
| Regime | What it demands here | How the system complies |
|---|---|---|
| Critical-infrastructure security (NERC CIP) | Cyber assets in CIP scope carry strict access, change and electronic-security-perimeter controls, and a vendor who blurs the boundary fails the security review. | Nothing connects to operational technology. Systems deploy inside utility-controlled environments on the data side of the boundary, and the data-flow documentation lands with your security team before the build starts. |
| Utility commission reliability reporting | Distributors report interruption statistics to their commission, the OEB and BCUC among them, and accountability for accuracy travels with the filing. | Draft tables assemble continuously with every outage minute traced to its OMS record and cause code. A named person reviews and signs. Nothing files itself. |
| PIPEDA | Customer personal information at a utility means meter, account and billing data, handled under Canadian privacy law with residency and retention answered before anything is built. | Customer data stays inside your environment, scoped per role at the database layer, and the risk models this page describes run on assets and weather, not on people. |
| Occupational safety (WorkSafeBC and provincial OH&S) | Work near energized conductors runs under provincial OH&S law, utility safety rules and utility arborist certification requirements. | The system never directs field work. It ranks spans and drafts schedules for a supervisor to assign, and a safety call made in the field beats a ranking every time. |
| Vendor posture | Utility procurement and security reviews expect attestations, a clean data-flow story and no claims that outrun the paper. | Our core vendors publish SOC 2 reports on their trust pages. The attestations enter your procurement file as they stand, and no answer on the security questionnaire claims what the data-flow map cannot show. |
The regimes that govern energy & utilities, what each demands, and how the system complies
Regime
- Critical-infrastructure security (NERC CIP)
What it demands here
Cyber assets in CIP scope carry strict access, change and electronic-security-perimeter controls, and a vendor who blurs the boundary fails the security review.
How the system complies
Nothing connects to operational technology. Systems deploy inside utility-controlled environments on the data side of the boundary, and the data-flow documentation lands with your security team before the build starts.
- Utility commission reliability reporting
What it demands here
Distributors report interruption statistics to their commission, the OEB and BCUC among them, and accountability for accuracy travels with the filing.
How the system complies
Draft tables assemble continuously with every outage minute traced to its OMS record and cause code. A named person reviews and signs. Nothing files itself.
- PIPEDA
What it demands here
Customer personal information at a utility means meter, account and billing data, handled under Canadian privacy law with residency and retention answered before anything is built.
How the system complies
Customer data stays inside your environment, scoped per role at the database layer, and the risk models this page describes run on assets and weather, not on people.
- Occupational safety (WorkSafeBC and provincial OH&S)
What it demands here
Work near energized conductors runs under provincial OH&S law, utility safety rules and utility arborist certification requirements.
How the system complies
The system never directs field work. It ranks spans and drafts schedules for a supervisor to assign, and a safety call made in the field beats a ranking every time.
- Vendor posture
What it demands here
Utility procurement and security reviews expect attestations, a clean data-flow story and no claims that outrun the paper.
How the system complies
Our core vendors publish SOC 2 reports on their trust pages. The attestations enter your procurement file as they stand, and no answer on the security questionnaire claims what the data-flow map cannot show.
Systems deploy inside utility-controlled environments, control systems stay untouched, and CIP boundaries are respected by design where they apply.
The economics
Before and after economics
Line
- Manual hours in scope
Before
At the page defaults, 8 people spending 8 hours a week on inspection filing, outage write-ups and report assembly is 64 hours, $5,440 a week at $85 loaded cost
After
The calculator above prices your own baseline, and the 5x ROI guarantee is measured against a baseline you sign before we build
- Trim planning
Before
Cycle-based, every span on the same clock regardless of what the outage record says about it
After
A risk-ranked span list with sources, stated qualitatively on purpose, no utility outcome number is published yet
- Storm pre-staging
Before
A judgment call made the afternoon before, from a forecast map and memory
After
A ranked proposal from the forecast and the feeder histories, and the supervisor still makes the call
- Reliability filing prep
Before
Weeks of assembling interruption records and cause codes at filing time
After
Draft tables maintained continuously with every minute traced, a named person reviews and signs
Advizr publishes no utility client outcome yet, and this table does not pretend otherwise. The only dollar figure is arithmetic on this page's calculator defaults, 8 people at 8 manual hours a week each at $85 loaded hourly cost, which is 64 hours or $5,440 a week in scope. Every other row states the change qualitatively, and the founding-client terms near the top of the page are the honest substitute for a case study.
The objections
CenterPoint runs a LiDAR digital twin with a budget we will never have.
The CenterPoint number is on this page because it proves the direction, not the price of admission. You do not need a digital twin to stop trimming on the calendar. You need the LiDAR or contractor photos you already pay for, your outage history by feeder, and a ranked span list your general foreman trusts. The mechanism scales down to a forty-crew contractor. The moat was never the model, it was doing the scoring at all.
A model that cries wolf costs us truck rolls.
It is graded before it costs you one. Detection is scored against defects your crews already verified in the field, and misses and false alarms are reported as two separate numbers, never blended, because a missed strike tree and a wasted truck roll are different failures with different prices. Every rejected flag goes back into the grading set, so the false-alarm rate is a curve you watch, not a promise you took.
Our line crews will read this as headcount software.
Software does not trim a cedar or splice a conductor, and qualified crews are the scarcest thing in this sector. The system ranks spans and drafts paperwork. The general foreman still runs the crews. What a crew notices after the build is fewer wasted rolls to spans that did not need them, and a work order that arrives with the photo and the span id already attached.
Half our inspection history is paper sheets and photos stuck in email.
Then we start with one district, not with a data project. The pole photos, last spring's contractor flight and the OMS extract get read as they are, and detection is graded against defects your crews already verified before anything ranks a live span. If the record cannot support the ranking, the audit says that before you spend on it, and what the build produces from day one is the structured registry you were missing.
A wrong load forecast ends up in a rate case.
Which is why the forecasting here is classic statistics, backtested against your own load history, not a language model improvising. The forecast lands as planning scenarios with stated assumptions that a planner owns and defends. Nothing reaches the commission that a person did not sign, and the backtest record is part of the deliverable, so the defence of the number exists before the number is used.
How we build it for energy and utilities
Index the asset history first.
A question about a specific span or substation should return its own record, not a general answer about equipment of that type. That is the whole build in one sentence.
Report misses and false alarms separately.
A missed defect and a false alarm have very different costs here. Reporting one blended accuracy number hides the one that matters.
Keep the system beside operations, never inside it.
Inspection scoring and risk sweeps produce ranked work proposals. Nothing connects to operational control systems, by design and by architecture.
What we will not automate
Any operational control action, and any filing submitted without a licensed person. The system ranks work by consequence and the people who own that call decide.
How the system is built for energy and utilities
See the full capability mapRetrieval
Asset records, inspection history and regulatory filings indexed together, so a question about a span or a substation returns its own history rather than a general answer about equipment of that type.
- pgvector
- Full-text BM25
- Per-asset indexing
Agents and orchestration
Agents score inspection imagery and run risk sweeps on a schedule, then open a work proposal ranked by consequence. Nothing touches operational technology. The recommendation goes to the people who already own that call.
- agent-worker
- Scheduled runs
- Proposal queue
Evaluation
Detection is graded against inspections your crews have already verified in the field. In this sector a missed defect and a false alarm have very different costs, so both are reported separately rather than as one accuracy figure.
- Ground-truth grading
- Separate miss and false-alarm rates
- quality-worker
Models
Vision models on inspection imagery. Classic load and outage forecasting, which is a mature statistical field and does not need a language model. Frontier models for the filings and the field-report drafting.
- Vision models
- Load and outage forecasting
- Frontier models, one gateway, routed per task
Data boundary
Grid and asset data is critical infrastructure information. It stays in your accounts, scoped per role, with audit logs on access, under no-training API terms. Nothing connects to operational control systems.
- Supabase row-level security
- Access audit logs
- No OT connectivity
Asset registry, Inspection media, Outage systems feed a hybrid index. The agent runtime works from that index, and every consequential action passes a human approval before it reaches Workflow builder, Task board, Cost per outcome.
Your systems
- 01Asset registry
- 02Inspection media
- 03Outage systems
The system
- 01Hybrid index
- 02Agent runtime
- 03Your approvalhuman
Where your team works
- 01Workflow builder
- 02Task board
- 03Cost per outcome
Asset registry, Inspection media, Outage systems and the system run inside a boundary labeled your accounts. The only path that crosses the boundary is the audited egress to the model API, under no-training terms. The boundary is what answers Critical-infrastructure security (NERC CIP) and Utility commission reliability reporting.
Inside your accounts
- 01Asset registry
- 02Inspection media
- 03Outage systems
- 04System in your cloud
- 05Audit log
Outside, through the audited port
- 01Model API
What that means in practice
Classic and predictive ML
Not every problem needs a language model. Predicting numbers, churn, demand, fraud risk, is usually solved better, cheaper and more explainably with proven statistical ML.
Where we stop. When the input is language, judgment or unstructured documents, classic ML underperforms and we say so. The discipline runs both ways: if your problem is a prediction problem, you will hear that it does not need an LLM from us before you pay for one.
MLOps and infrastructure
The plumbing that keeps AI running in production: deployment, scaling, monitoring and retraining, so it does not quietly degrade after launch.
Where we stop. We do not park your system on infrastructure only we understand. If your platform team runs Kubernetes on AWS, we deploy there. The architecture has to survive us leaving; that is the point of it.
Security, privacy and governance
Keeping your data yours, and your AI safe to put in front of customers.
Where we stop. We do not claim certifications we do not hold, and we will not ship a customer-facing agent without a human gate and an eval suite. If a vendor cannot offer no-training terms, it does not get into the stack.
Where your team works
Tour the platformWorkflow builder
Describe the workflow. Watch it assemble.
Say what should happen in plain language and the builder assembles the automation on a canvas you can read, run, and change.
Task board
People and agents, working the same board.
Every piece of work lives on one shared board, whether a person or an agent owns it. Handoffs between the two are explicit, so nothing falls in the gap.
Cost per outcome
Every dollar of spend traced to the work behind it.
Outcomes delivered, cost per outcome, value attributed, return on spend. The same measurement the guarantee is settled against, live on one page.
Approval queue
The AI proposes. You approve. Nothing sends itself.
Every consequential action arrives as a proposal with the risk, reversibility, and expiry spelled out before you say yes. Control stays in the room.
The hard questions
Free · 3-5 days
Know your number in five days.
We map your operations, find the highest-ROI automations, and hand you a ranked plan with the payback math attached. Yours to keep, whoever builds it.
Prefer to talk first? Book 15 minutes with James. No pitch deck.