Telecoms operators train front-line teams on the agent's screens and their retention targets, then ask them to review held actions at enormous volume. The predictable failures follow: SIM swap and port requests released because a caller sounded convincing, retention offers pushed past the agent's bounds, and bill credits approved in bulk. The fixes cost little next to the losses.
Why are SIM swaps the first thing to train?
A SIM swap or number port hands control of a customer's phone number to whoever asked for it. Because that number often receives one-time passcodes for banks and other services, a fraudulent swap can empty accounts far from the operator's own systems.
An agent can check the request against the account: the device history, recent changes, how the requester authenticated. It holds requests that do not add up, at depth 4 because the action is hard to undo. The reviewer then faces a caller who is often urgent, articulate and armed with personal details. A request backed only by what the caller says is declared; a passcode confirmed on the customer's registered device is far stronger. Train reviewers on real past fraud attempts, including well-prepared ones, and make it clear that refusing a genuine customer for an hour is a much smaller harm than releasing a fraudulent swap. Treat a missed red line here as the most serious error in the queue.
How do retention targets bend review?
Telecoms sales and retention teams are measured on keeping customers. An agent can propose retention offers within bounds the operator has set: a discount, a handset upgrade, a credit. Offers above those bounds are refused or held.
Untrained staff under target pressure find ways round. They split a large credit into several small ones, stack offers the bounds were designed to prevent, or ask the agent to rephrase until something passes. Each workaround defeats a pricing decision the business made deliberately. Training should explain why each bound exists, show what the record reveals when someone games it, and give staff a proper route to escalate a valuable customer's case. Pair it with a change in measurement: review quality should count alongside retention.
What goes wrong with credits at scale?
Billing disputes are constant in telecoms, and agents handle many of them: reading the bill, the usage records and the tariff, and drafting or issuing a credit within limits. The reviewer sees only what the agent held.
At volume, reviewers drift into approving held credits in bulk. The usage records from the operator's own network systems are sealed; the customer's statement that they never made those calls is declared. A held credit often means the two disagree. Reviewers need to know how to read usage evidence, when a dispute points to fraud or a fault rather than a billing error, and when to escalate a pattern rather than credit the individual case. A widespread outage is a special case: compensation drafts for many customers at once need a check on the outage record before release.
Why do outsourced and multi-site teams need one standard?
Most operators run front-line service across several sites and partners. If each site trains differently, the same held SIM swap may be released in one centre and refused in another, and the agent's record becomes inconsistent. That inconsistency hurts customers and weakens the evidence the agent needs to keep its depth when its certificate comes up for renewal after 90 days.
Set one assessment, one register of certified reviewers and one set of approval rules across all sites and partners. An operator can adopt the practice of tying approval rights for depth 4 actions such as SIM swaps to a current certificate, wherever the reviewer sits. Calibrate across sites regularly, using the same cases, and share the results so each site can see where its decisions differ from the rest.
What about security duties?
Providers of public electronic communications networks and services in the EU fall within NIS2, with cybersecurity risk management and incident reporting duties. A sudden rise in held port requests may be the first sign of an organised attack, not just a busy day. Reviewers and team leads should know how to report a pattern, not just decide the case in front of them. This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
What should a telecoms operator do instead?
- Make SIM swap and port review the first and most heavily assessed skill.
- Explain every retention bound and measure review quality alongside retention.
- Teach usage evidence and fraud patterns behind billing disputes.
- Run one standard across all sites and partners, with a shared register.
- Train team leads to spot attack patterns and know who to tell.
Agent certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day, so routine credits may suddenly arrive in the held queue. Teams should know this is the control working, not a fault to clear quickly.
The general order is in the guide to training teams for AI agents. For a telecoms version, see what an AI agent training programme for telecoms operators should include, the AI agent training plan template for telecoms operators, and how much AI agent training costs for telecoms operators. Banks see the downstream effect of SIM swap fraud, covered from their side in what banks get wrong about AI agent training.
Questions people ask
Can an agent approve a SIM swap on its own?
Only where the operator's own limits allow and the evidence is strong, such as authentication on the customer's registered device. Anything weaker should be held for a trained person. Many operators choose to hold every swap.
Should retail store staff be trained like contact centre staff?
Yes, if they approve the same actions. Store staff often process swaps and upgrades face to face and face the same social engineering. Hold them to the same assessment and register.
How do we keep training current as fraud changes?
Refresh the case library with recent anonymised attempts from your own queue and run short calibration sessions on them. Treat every confirmed fraud that passed review as training material for everyone.