AI for Utilities & Power · Generation and water · Networks · Customer service
Run the network. Answer the street.
- Saudi Arabic, day and night
- Read-only to SCADA and EMS
- Bills explained from the meter
- Crews dispatched, not promised
One AI layer. Every customer channel and every system behind the network. Swipe to see the whole diagram.
The Voho Utilities AI Harness
No script wins a high-bill call. Reading the meter does.
The customer is angry, the invoice is usually correct, and the reason is two estimated reads and a failing air conditioner. An agent who cannot see the interval data can only apologise. The harness gives models your meters, your network model and your credit policy, and keeps every hand off the network itself.
A model on its own
- Explains the tariff to somebody who asked about their meter
- Cannot see that June and July were estimated
- Promises a restoration time nobody in the control room agreed
- Offers an instalment plan outside your credit policy
- Treats a dialysis patient like any other account
The same model, in the harness
- Reads fourteen months of interval data before it answers
- Names the two estimated reads and what they moved
- Quotes the estimate the incident record actually holds
- Offers only the arrangements your policy allows
- Calls the medically dependent customers first, by voice
Utility systems
What you already run, left in place
- SCADA / EMS
- Outage management
- GIS network model
- AMI / smart meters
- Billing / CIS
- SAP PM
- Work management
- Customer portal
- IVR / contact centre
- SMS gateway
- Engineering archives
Voho Utilities AI Harness
The runtime between your systems and the models
- Tool calling
- Model routing
- Utility connectors
- Customer verification
- Memory
- Context
- RAG
- Workflow orchestration
- Permissions
- Payment and credit rules
- Human approvals
- Evaluations
- Guardrails
- Audit trails
- Observability
Model layer
Routed per task, never one model for everything
- Saudi Arabic speech
- Speech synthesis
- Foundation models / LLMs
- Consumption and load models
- Leak and anomaly models
- Asset failure models
Agent layer
Scoped to one desk each
- Customer Agent
- Billing Agent
- Outage Agent
- Metering Agent
- Connections Agent
- Reliability Agent
Where the work lands
Finished in the system that owns it
- Calls answered
- Bills explained
- Outages notified
- Meters checked
- Crews dispatched
- Connections progressed
01
Answer.
Through the August peak and the first cold night, every call picked up in Saudi Arabic or English, on the numbers and the IVR you already run.
02
Locate.
The account, the meter, and the transformer and feeder it hangs off, resolved from your GIS before anything is explained or promised.
03
Explain.
The bill against the actual meter reads, the outage against the actual fault. A customer accepts an explanation they can check; they do not accept an apology.
04
Restore.
Work orders raised and crews dispatched in the systems that own them, with the estimate the customer hears matching the one the control room holds.
01 · The high-bill call
Four times last month, and the invoice is right.
The most common call in the industry, and the one a script always loses. Here it is answered from the meter: the estimated reads named, the night-time load explained, a meter check raised anyway, and an instalment plan offered only because policy allows it.
Transcript
Verification
Nothing on the account or the meter is readable until these pass.
Tool calls
waiting…
Policy
- Explaining the reads, raising a meter check and offering an arrangement inside your credit policy complete on the call.
- A write-off, a tariff change or anything outside the policy is drafted for a person, and disconnection is never automated at all.
Illustrative call · recording, transcript and every tool call on one record
02 · When the supply goes
The control room knows at 14:02. So should the street.
The outage system and the GIS already hold which section is out and who hangs off it, including the customers who cannot wait. The gap between knowing and telling is the only thing standing between a quiet afternoon and twelve hundred calls.
Accounts on the section
—
Vulnerable, called first
—
Notified before calling
—
Calls that never arrived
—
Outage Agent · response
Trip read from the outage system
OMS / SCADAFeeder F-114 out at the 13.8 kV breaker. The same record the control room is working from.
The faulted section resolved in GIS
GIS network modelThree streets downstream of the transformer, not everyone who has ever complained about this feeder.
Vulnerable and critical customers first
CIS · critical registerMedical dependencies, a dialysis centre and two pumping stations identified before any bulk message goes out.
Those customers called, in person
VoiceA call rather than a message, in Arabic, with what to do if their equipment cannot wait for restoration.
Everyone else notified once
SMS gatewayOne message on the channel they accepted, with the cause and the estimate the control room actually holds.
The IVR stops arguing with them
IVRCallers from those streets hear the incident and the estimate instead of the main menu and a queue.
Crew dispatched with the picture
Work managementCustomer count, critical loads and call volume attached to the incident before the crew is assigned.
Restoration confirmed, then said
OMSSupply checked on the network before anyone is told it is back, and one closing message rather than five.
Across the utility
One harness. The network and the phone.
The connectors, the verification rules and the evaluation set are built once, and the generation side runs on the same industrial harness as a refinery.
Customer service
Two questions, most of the volume
- High-bill calls answered against the meter history, not a tariff leaflet
- Estimated reads identified and a meter check raised while the customer is still on the line
- Payment arrangements offered inside your own credit policy, never invented on the call
- Supply-off calls answered with the real cause, the real crew status and the real estimate
- New connections and capacity requests progressed and chased, with the missing document named
Network and outage
Between the control room and the street
- Affected customers resolved from the outage system and the network model, not from who called
- Proactive notification before the queue builds, once, in the customer’s own language
- Call volume, vulnerable customers and critical loads attached to the incident before dispatch
- Repeat faults on the same feeder surfaced as a reliability problem rather than five tickets
- Restoration confirmed against the network before anyone is told it is over
Generation, water and assets
The plants behind the supply
- Turbine, pump and desalination train health read from the historian and scored against the envelope
- Maintenance history and OEM manuals answerable in one question, with the page cited
- Work orders raised in SAP PM with the evidence that justified them attached
- Chemical, permit and inspection records read and checked against the procedure
- Shift handovers and event reports written from the alarm log and the operator’s own words
Connectors
One AI layer. Every system behind the supply.
Read-only wherever it touches the network, write access only in billing, work management and the case systems where a request belongs.
SCADA / EMS
Read-only. Alarms, states, load
Outage management
Incidents, crews, restoration
GIS network model
Feeders, transformers, mains
AMI and metering
Interval reads, estimates, events
Billing / CIS
Accounts, tariffs, arrangements
SAP PM
Assets, notifications, work orders
Work management
Crew slots and confirmations
Telephony
Cisco, Avaya, SIP. Your numbers
Built for a regulated utility
What operations and the regulator ask, answered in the product.
The first boundary is the one that matters most: nothing Voho does reaches the network. It reads, and everything it produces is for a person or a customer.
It never operates the network
SCADA, EMS and the outage system are read. No switching, no setpoint, no control action of any kind. Everything it produces is a message, a work order or a request for a person.
Nothing is disconnected by an agent
Disconnection is never automated. Reconnection after payment runs only inside the rules you configure, and anything touching a vulnerable or critical customer waits for a person.
Money moves inside your credit policy
Payment arrangements, adjustments and credits are offered only within configured limits. Above them the agent drafts and a person approves, and every attempt is logged.
Estimates that match the control room
Restoration times given to customers come from the incident record, not from a model’s guess, and they change when the control room changes them.
Inside your own network
On your servers or in a Saudi region. Customer, meter and network data stay where your policy says, and the models run where the data is.
Evaluated on your own peak
Scored against real calls from a real peak day and real past incidents before go-live: what it answered, what it raised, and where it should have handed over.
Already built
We have already done this for companies you know.
The library holds working agents for named Saudi organisations, built from public knowledge of how each one runs, unrequested. Five agents built for a named utility, running in your browser.
Saudi Electricity Company
Utilities
The Kingdom’s electricity network, and a call centre that receives its worst day and its hottest day simultaneously.
Run the demos →ACWA Power
Utilities
A developer of power and desalination plants across three continents. A business where the asset is a contract as much as a turbine.
Run the demos →Marafiq
Utilities
Power, water and wastewater for two industrial cities, where the customers are plants that cannot be interrupted and there is no alternative supplier.
Run the demos →National Water Company
Utilities
Water and wastewater for the major cities, where a customer with an impossible bill is usually a customer with a leak they cannot see.
Run the demos →How a project runs
Start with the high-bill call. Keep the harness.
The first project proves itself on numbers you already report: answer rate, repeat calls, complaints upheld and average handling time on a peak day.
01
Take the high-bill call
One call reason, the biggest one, with a number your contact centre already reports and a customer outcome you can measure.
02
Connect read-only first
Meter reads, billing and the outage system in, your telephony for the calls, reviewed by your OT security and customer teams.
03
Replay a peak day
The agents are scored against the calls and incidents of a real August day: what they would have explained, raised and escalated.
04
Go live out of hours, then widen
Start where the alternative is a queue nobody answers, then take the daytime volume and the next call reason on the same harness.
What operations, customer service and the regulator ask first.
If your question is not here, bring it to a technical session with our engineers.
01Can it operate the network?
No. SCADA, EMS and the outage system are read-only to Voho. There is no switching, no setpoint change and no control action. What it produces is a notification, a work order or an escalation to one of your people.
02Can it explain a high bill without making things worse?
That is exactly the case it is built for. It reads the interval data and the read history, identifies estimated reads, compares the period against the same period last year and against the weather, and says plainly which of those explains the bill. Where the meter looks wrong it raises a check rather than defending the invoice.
03Will it disconnect or reconnect customers?
It never disconnects. Reconnection after payment can be completed inside rules you configure, and anything involving a vulnerable customer, a medical dependency or a critical load waits for a person regardless of the rule.
04How does it know who is affected by an outage?
From the outage management system and your GIS network model, so the notification reaches the customers on the faulted section rather than everyone who has ever complained. Sending is governed by your own contact policy.
05What happens on a peak day when volume triples?
The queue is answered at the same speed at two in the afternoon in August as at midnight in February, and anything the agent cannot finish reaches your people with the account, the meter and the incident already in front of them.
06Do you work for water and district cooling too, or only power?
Both. The billing and outage arguments are the same, and the network model, the meters and the assets are simply different: mains, pumps and reservoirs instead of feeders, transformers and lines.
07What about the generation and desalination side?
That is the industrial harness, connected to your historian and SAP PM: turbine and pump health, maintenance history, work orders with evidence attached, and the engineering archive answerable with citations.