AI for Telecom · Consumer care · Activations · Network and enterprise
Answer at volume. Tell them before they call.
- Saudi Arabic, day and night
- Your IVR, your numbers
- Provisioned in BSS, not a ticket
- Spend limits and approvals
04 · WHERE THE WORK LANDS
Care
→ Answered at peak
logged · recording kept
Activations
→ eSIM live
confirmed on network
Billing
→ Invoice explained
from charging
Network
→ 3,884 told first
before they dialled
Enterprise
→ SLA answered
against the ticket
Field
→ Visit confirmed
slot re-confirmed
03 · HARNESS
VOHO TELECOM AI HARNESS
ANSWER · VERIFY · PROVISION · PRE-EMPT
VOICE
Listen · speak · barge-in
TOOLS
Charging · CRM · HSS
MODELS
Arabic speech · LLM · risk
AGENTS
One per desk
AUTHENTICATION
Harder bar on risky actions
SPEND LIMITS
Anything chargeable
EVALUATIONS
Scored at peak volume
MODEL ROUTING · CONTEXT · RAG · GUARDRAILS · PII REDACTION · HUMAN APPROVALS · AUDIT TRAIL
02 · SYSTEMS OF RECORD
01 · THE EDGE OF YOUR NETWORK
Swipe the drawing sideways.
One AI layer. Every channel a subscriber uses and every system behind it.
The Voho Telecom AI Harness
A model with no charging system is an expensive hold message.
It cannot see the balance, the bundle, the device or the site that just dropped, and it should not be allowed to change a plan on its own judgement. The harness gives models your systems, your authentication bar per action, and spend limits on anything that costs money.
A model on its own
- Guesses when the bundle ran out
- Cannot provision anything it promises
- Treats a SIM swap like a balance query
- Knows nothing about the incident already open
- Leaves your fraud team no trail
The same model, in the harness
- Reads charging and tells the subscriber the actual date
- Completes the bundle and the eSIM on the network
- Applies a harder verification bar to the risky actions
- Sees the same OSS alarm the NOC does
- Writes every attempt, approval and refusal to the audit trail
Operator systems
What you already run
- BSS / billing
- Online charging
- CRM
- Provisioning · HLR / HSS
- OSS / NOC alarms
- Number portability
- Retail POS
- Device catalogue
- Field dispatch
- IVR / contact centre
- SMS gateway
Voho Telecom AI Harness
The runtime between your systems and the models
- Tool calling
- Model routing
- Telecom connectors
- Subscriber authentication
- Memory
- Context
- RAG
- Workflow orchestration
- Permissions
- Spend limits
- Human approvals
- Evaluations
- Guardrails
- PII redaction
- Audit trails
Model layer
Routed per task, never one model for everything
- Saudi Arabic speech
- Speech synthesis
- Foundation models / LLMs
- Network anomaly models
- Churn and propensity models
- ID and document models
Agent layer
Scoped to one desk each
- Care Agent
- Activation Agent
- Billing Agent
- Network Agent
- Retention Agent
- Enterprise Agent
- Field Dispatch Agent
Where the work lands
Finished in the system that owns it
- Calls answered
- SIMs activated
- Plans changed
- Outages notified
- Tickets enriched
- Engineer visits booked
01
Answer.
Peak hours, Eid, the evening after a price change: every call picked up in Saudi Arabic or English, on the IVR and numbers you already run.
02
Verify.
The subscriber authenticated against your own rules before anything on the account is read or changed, with a harder bar on the actions fraudsters want.
03
Provision.
Bundles added, plans changed, eSIMs activated, roaming enabled, executed in BSS and the network during the call, inside the spend limits you set.
04
Pre-empt.
When a site degrades, the affected subscribers are told before they dial, the IVR says so, and the ticket carries who is impacted.
01 · Consumer care
Finished on the network before the call ends.
Balance, bundle, device, activation: the four reasons behind most of the queue. This one is answered, charged and provisioned in about ninety seconds, with the risky actions deliberately left on the other side of the line.
Transcript
Authentication
Nothing on the account is readable until these pass.
Tool calls
waiting…
Policy
- Renewing a bundle the subscriber already uses, inside your spend limit, completes in the call.
- A SIM swap, a port-out or a credit above the limit takes a harder check and a person, and is flagged to fraud.
Illustrative call · recording, transcript and every tool call on one record
02 · When the network breaks
Four thousand people, told before they dial.
Your NOC knows at 19:04. Your contact centre finds out from the ninetieth caller. The harness reads the same alarm, works out who is affected, tells them in their own language and hands the field team a ticket that says who is waiting.
Subscribers on the site
—
Notified before calling
—
Enterprise lines called
—
Calls that never arrived
—
Network Agent · response
Alarm read from OSS
OSS / NOCCell down at RUH-114, transmission fault. The same alarm the NOC is looking at.
Affected subscribers resolved
OSS + BSSSubscribers currently homed on the site, matched against BSS for contact policy and language.
IVR message updated for those numbers
IVRCallers from the affected cells hear the real cause and the estimate, not the generic menu.
Proactive notification sent
SMS gatewayIn the subscriber’s own language, on the channel they have accepted, once, not every ten minutes.
Enterprise lines called
VoiceBusiness accounts on the site get a call, in SLA order, with the ticket reference read out.
Ticket enriched for the field team
OSS ticketCustomer impact, enterprise exposure and call volume attached to the incident before dispatch.
Restoration confirmed and everyone told
ProvisioningService state re-checked on the network, then a single closing message. No second outage of noise.
Across the operator
One harness. Every desk that answers a subscriber.
The IVR integration, the connectors, the authentication bar and the evaluation set are built once. The next call reason is configuration.
Consumer care
The largest queue in the country
- Balance, bundle, invoice and fair-usage questions answered from charging, not from a script
- Plans changed, add-ons bought and roaming enabled during the call
- Complaints logged, classified and escalated inside the regulatory clock
- Data and coverage problems checked against the live network before a ticket is raised
- Retention offers presented only where your rules and margins allow
Activations and provisioning
From order to service, in one call
- eSIM and physical SIM activations completed and confirmed on the network
- Identity documents read and checked for new lines
- Number portability requests raised and tracked to completion
- Fibre orders qualified by address and engineer visits booked into dispatch
- Retail and reseller orders assisted without a person on the counter phone
Network and enterprise
Where care and operations meet
- Outage notification to affected subscribers before the queue builds
- Inbound calls about a known incident answered with the real restoration estimate
- Enterprise SLA queries answered against the ticket and the contract
- Field visits scheduled, confirmed and re-confirmed with the customer
- Ticket quality improved automatically: impact, history and customer context attached
Connectors
One AI layer. Every system behind the queue.
Read access first, write access exactly where you allow it, and nothing touching network configuration at all.
BSS and billing
Accounts, invoices, adjustments
Online charging
Balance, bundles, fair usage
Provisioning
HLR / HSS, eSIM, service state
OSS and NOC
Alarms, incidents, restoration
CRM
History, complaints, cases
Portability
In and out, with status
Field dispatch
Engineer slots and confirmations
IVR and SMS
Your numbers, queues and gateways
Built for an operator
What fraud and security ask, answered in the product.
The bar is per action, not per product, because a balance query and a SIM swap are not the same risk and should never take the same check.
A harder bar on the dangerous actions
SIM swap, number port-out and beneficiary-style changes require stronger verification than a balance question, can be held for a cooling period, and are flagged to fraud. You set the bar per action.
Spend limits on anything chargeable
Add-ons, credits and adjustments run inside configured limits. Above them the action is drafted and waits for a person, and every attempt is logged.
Inside your own network
On your infrastructure or in a Saudi region. Recordings, transcripts and subscriber data stay where your policy says and the models run where the data is.
Never trained on subscriber data
Calls, messages and documents are used to serve that request. They are not used to train models, ours or anyone else’s.
It reads the network, it does not touch it
Alarms, incidents and service state are read. Configuration changes on network elements stay with your NOC and your change process.
Evaluated on your own calls
Scored against real past calls before go-live: what it would have answered, what it would have provisioned, and where it would have handed over.
From the library
What this would look like at companies you know.
Demonstrations, not customer work. We built five agents for each of these organisations from public knowledge of how they run; none of them asked, and none of them run this today. Open one to see this page’s argument applied to a operator you recognise.
stc (Saudi Telecom Company)
Telecom
The Kingdom’s largest telecom operator: mobile, fibre, enterprise connectivity and a payments arm.
Run the demos →Mobily (Etihad Etisalat)
Telecom
The Kingdom’s second mobile operator, and a fibre business whose hardest calls are about a house that cannot get connected.
Run the demos →Zain KSA
Telecom
The challenger operator, which means winning customers off other networks, and the number-porting desk that decides whether that works.
Run the demos →Integrated Telecom Company
Telecom
Wholesale and infrastructure connectivity, where the customer is another operator and a fault report is a peering dispute waiting to happen.
Run the demos →Salam
Telecom
A challenger operator whose growth is in enterprise connectivity, where an outage is a customer’s branch network rather than a phone.
Run the demos →How a project runs
Start with the top call reason. Keep the harness.
The first project proves itself on numbers you already report: answer rate, average handling, first-contact resolution and repeat calls.
01
Take the top call reason
Balance and bundles, or activations. The one that fills your queue and already has a number attached to it.
02
Connect inside your network
Charging and CRM read access, provisioning for the actions you allow, your IVR, reviewed by security and by the fraud team.
03
Replay peak-hour traffic
Scored on real calls at real volume, including the ones that should have gone to a person and the ones that smell like fraud.
04
Go live on a slice, then widen
A percentage of traffic with handover always available, then the next call reason on the same harness.
What care, fraud and network ask first.
If your question is not here, bring it to a technical session with our engineers.
01Can it actually change a plan, or does it just raise a ticket?
It completes the change in BSS and the network during the call, for the actions you allow, inside the spend limits you configure. Where something is outside those limits it is drafted and handed to a person rather than promised and dropped.
02What about SIM swap? That is how accounts get stolen.
It is treated as a high-risk action, not a routine one: stronger verification than a balance query, an optional cooling period, notification to the registered contact and a flag to your fraud team. You set the bar, and every attempt is on the audit trail.
03Does it work with our IVR, or do we replace it?
It works with what you have, Cisco, Avaya, Genesys or SIP, on your existing numbers and queues. Nothing is ripped out to answer the first call.
04How does the outage notification work?
The harness reads the same OSS alarm your NOC sees, resolves which subscribers are affected from the network and BSS, notifies them on the channels they accept, updates the IVR message for those numbers and attaches customer impact to the ticket. Sending is governed by your own contact policy.
05Can it touch network configuration?
No. It reads alarms, incidents and service state. Changes to network elements remain with your NOC and your change process.
06Our customers switch between Arabic and English mid-sentence.
So does the agent. Najdi, Hijazi and the rest, with numbers, bundle names and dates read back the way a person would say them.
07What happens at peak, when volume triples?
Capacity is not a hiring problem here. The queue is answered at the same speed at nine in the evening as at nine in the morning, and anything the agent cannot finish goes to your people with the context already gathered.