Industries
AI in telecom operations
AI in telecom usually means network optimization, but a carrier's most expensive manual work sits in the customer-facing back office: order fallout, billing inquiries and disputes, service changes, collections and retention. That is where AI, including agentic AI that carries a case through several systems, can take over routine cases and leave people the judgment calls.
Search for AI in telecom and most of what comes back is about the network: traffic prediction, self-optimizing radio, predictive maintenance on towers. Those are real programs, run by engineering. Ask a carrier's head of operations where the manual work is, and the answer is different. It is in the orders that fall out of provisioning, the bills customers do not understand, the service moves that need four systems updated, and the queues of disputes and payment arrangements.
What operations problems do carriers face?
Contacts that self-service cannot close. A carrier's customer questions are about one account: this bill, this order, this outage at this address. Generic self-service does poorly with that, because it answers from help articles and the customer is asking about their own account. For carriers the unresolved remainder lands in the contact center.
Order fallout. An order passes through sales, credit, inventory, provisioning, number porting and billing systems. When one step rejects it, the order sits in a fallout queue until someone works out what went wrong. Customers notice a missed activation date long before the carrier does.
Moves, adds and changes. A move to a new address or a plan change touches the CRM, provisioning, field scheduling and billing. Each hand-off is a chance for the change to be half-made: the new service is active and the old one still bills.
Billing inquiries and disputes. Every bill run, every price change and every promotion that expires produces a wave of questions. Questions not answered well become billing disputes. The telecom billing page covers this domain in depth.
Collections and retention. Payment arrangements, hardship requests and cancellation calls each need account context and a policy decision, and each is handled one call at a time.
Which processes are good first candidates, and why?
| Process | Why it suits AI | What stays with a person |
|---|---|---|
| Billing inquiries | High volume, answerable from account data, measurable | Disputes above limits, hardship, regulatory complaints |
| Order fallout | Failure reasons repeat; fixes follow known patterns | Credit exceptions, unusual business orders |
| Moves and changes | Many systems, clear target state to verify | Field scheduling conflicts, address disputes |
| Payment arrangements | Written policy with clear eligibility | Hardship and vulnerable customers |
| Dispute investigation | Evidence scattered across systems | The decision on contested charges |
The first candidate is usually billing inquiries, because volume is easy to measure and a better answer shows up in the next bill cycle. The billing inquiry automation page sets out the design.
What does agentic AI change in telecom?
An AI agent can carry a case across systems: read the fallout error, check the inventory record, correct the provisioning request, resubmit it, confirm activation and tell the customer. That is a bounded agent workflow with tools, limits and stop conditions. The agent owns the routine path. When the error does not match a known pattern, or the fix needs a credit decision, it stops and hands the case to a person with everything it found. The AI agents for business guide covers how to draw that line before building.
What human-in-the-loop constraints are specific to telecom?
The billing rules (call detail disclosure, truth in billing, broadband labels and consent for AI-voiced outbound calls) are covered on the telecom billing page. Outside billing, three constraints shape where the person sits.
- SIM changes. 47 CFR 64.2010(h) requires wireless providers to authenticate a customer with secure methods before executing a SIM change, without relying on readily available biographical information, account information, recent payment information or call detail. It also requires that employees handling inbound contacts cannot see customer data until the customer is authenticated, and training on SIM swap fraud. The eCFR notes that compliance with paragraph (h) is not required until a compliance date is set, but a carrier designing an AI agent today should build to it. An agent can collect the request and run the authentication; a person trained on SIM swap fraud handles anything that looks wrong.
- Changes that move service or money. Port-related order errors, cancellations, payment arrangements and service credits after an outage change what the customer pays or receives. The agent prepares them; within written limits it executes them; beyond those limits a person approves.
- Vulnerability and complaints. Hardship, suspected fraud or account takeover, and complaints filed with a regulator go to a person every time.
Where should a carrier's first pilot sit?
One process, one customer segment, one or two channels. A billing inquiry pilot runs across at least one bill cycle; an order fallout pilot covers one product line, such as fixed broadband installs, for a few weeks of orders. At first the agent only proposes fixes for a person to approve, and it earns the right to submit them itself once its proposals stop being corrected. Baselines for call deflection, time to resolution and repeat contacts are set before day one.
That narrow start is also how Manthle's engineers work with a carrier: one process, engineered around the existing stack and run alongside the carrier's own team, with the next process added only once the first holds.
You know what happens before we start.
One process, five steps. A pilot is measured against goals agreed at the start, and once it meets them, it becomes a monthly operations contract.
- 01
See
We spend time inside the process: its history, its systems, and the people who settle the hard cases by habit.
- 02
Rethink
With your process owner, we redesign before anything is automated: which decisions belong to a model, which to a rule, and which to a person.
- 03
Engineer
We build the AI system around the redesigned process, connected to the systems you already run, with the exception queue designed alongside it.
- 04
Harden
We test it on the cases that break things: incomplete inputs, odd edge cases, systems that go down.
- 05
Run
We operate it with your team, and keep it dependable as models, systems and policies change.
Questions
Carriers use AI in network planning and maintenance, and it can also run operational work such as billing inquiries, order handling, collections and customer service.
Agentic AI in telecom refers to AI agents that carry a customer case through several steps and systems, such as checking an order, fixing a provisioning error and confirming with the customer.
Billing inquiries, order fallout, service moves and changes, and collections suit AI first because they are high in volume, governed by written policy and measurable.
US carriers must apply FCC rules on customer proprietary network information, including the SIM change rules in 47 CFR 64.2010, truth in billing and broadband labels, and the TCPA's consent rules for artificial voices.
AI agents can handle inbound calls with proper authentication, but AI-voiced outbound calls count as artificial voice calls under the TCPA and need prior express consent unless an exemption applies.