> For the complete documentation index, see [llms.txt](https://docs.layerlens.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.layerlens.ai/more-in-this-section-2/scenarios-11.md). # Telecommunications scenarios Five canonical scenarios where wireless carriers, MVNOs, ISPs, and cable operators use Stratix evaluation to gate AI features. ## 1. Customer service automation **Setup:** Customer interacts with a chatbot or voice assistant for billing, plan changes, troubleshooting, outage status. AI grounds responses in the customer's account and the carrier's policy. **Why it's hard:** * Account changes affect bills directly — wrong plan promise becomes a chargeback nightmare * Outage-status accuracy must be near-real-time * TCPA / CAN-SPAM consent rules govern any outbound contact * Multi-language parity is a regulatory expectation **Stakes:** * FCC complaint volume directly tied to carrier performance scores * State PUCs enforce service-quality rules * Class actions on misleading plan changes are common * TCPA per-violation damages: $500-$1,500 per call/text **Stratix evaluation criteria:** | Aspect | What you measure | Pattern | | ----------------------- | ----------------------------------------- | ------------------------------ | | Plan-promise accuracy | What AI promised matches policy | Hard rule + faithfulness judge | | Outage-status freshness | Cited outage data is current | Hard rule | | Authentication rule | Account changes require verified identity | Hard rule | | Policy citation | Decisions cite the policy section | Citation rule | | Multilingual parity | Same accuracy across languages | Per-language judge | ## 2. Network operations and incident response **Setup:** AI assists NOC engineers — surfaces alarm correlations, suggests root-cause hypotheses, drafts customer-impact statements, recommends mitigation steps. **Why it's hard:** * Network state is high-velocity; stale guidance is wrong guidance * Cascading-failure dynamics defeat naive correlations * Public safety dependencies (911, FirstNet) raise the stakes * Incident communications carry legal exposure (FCC reporting, securities disclosures) **Stakes:** * FCC Network Outage Reporting System (NORS) timelines * 911 outage rules — 30-minute notice to PSAPs in many cases * Material network outages can be SEC Form 8-K events * Reputation cost compounds with social-media velocity **Stratix evaluation criteria:** * **Mitigation-recommendation faithfulness** — Output grounded in runbook * **NORS-trigger detection** — Hard rule: 911-affecting outages flag NORS path * **Customer-impact-statement faithfulness** — Plain-language judge against actual incident facts * **Advisory-only rule** — Hard rule: AI never auto-applies network changes * **Stale-data rule** — Cited monitoring data freshness within threshold ## 3. Plan recommendations and upsell **Setup:** AI recommends upgrades, add-ons, plan changes based on usage patterns. Customer-facing or rep-assisted. **Why it's hard:** * "Negative-option" enrollment risk under FTC + state UDAP * Bait-and-switch perception when promotions don't match reality * Personalized-pricing rules vary by jurisdiction * Senior / vulnerable population protections **Stakes:** * FTC negative-option enforcement is high-priority * State AGs aggressive on telecom upsell practices * FCC has used Section 201(b) "just and reasonable" authority on misleading plans **Stratix evaluation criteria:** * **Promotion-realizability rule** — Hard rule: every promotion must be actually achievable * **Negative-option guardrail** — Hard rule: no enrollment without affirmative consent * **Senior-protection rule** — Identified senior accounts route to human for material plan changes * **Disclosure-completeness rule** — Required terms (price, term, ETF) present * **Personalized-pricing fairness** — Group fairness on offered prices ## 4. Fraud and abuse detection **Setup:** AI scores accounts and traffic for fraud — SIM swap, account takeover, robocall origination, IRSF (international revenue share fraud). **Why it's hard:** * Dynamic adversaries; signatures shift * False positives lock out legitimate customers — high CSAT cost * ATIS / FCC STIR/SHAKEN environment for caller-ID * Enterprise customers expect visibility into fraud actions affecting their traffic **Stakes:** * IRSF losses to industry are billions annually * SIM-swap fraud has driven multi-million-dollar individual losses + class actions * FCC enforcement of robocall mitigation * Wireless industry settlements over SIM-swap practices **Stratix evaluation criteria:** * **Top-decile precision** — Tracked against confirmed fraud * **Fairness scorer** — Group-level lockout rate parity * **Reason-code coverage** — Every action cites contributing signals * **Account-recovery rule** — Hard rule: lockout includes a documented recovery path * **STIR/SHAKEN compliance** — Caller-ID attestation handled per A/B/C-level rules ## 5. Lawful intercept and law-enforcement requests **Setup:** AI assists compliance/legal team with intake of lawful-intercept requests, subpoena/warrant interpretation, response drafting. **Why it's hard:** * CALEA + state-specific assistance obligations * ECPA Title III (wiretap) vs. Title II (subscriber info) vs. SCA distinctions critical * Over-disclosure violates customer privacy + carrier policy * Under-disclosure violates court order; contempt exposure **Stakes:** * Criminal exposure for non-compliance with valid orders * Civil exposure under ECPA / state analogs for over-disclosure * Reputational cost when over-disclosure surfaces **Stratix evaluation criteria:** * **Order-type classification rule** — Hard rule: subpoena / 2703(d) / warrant / Title III correctly classified * **Scope-limitation rule** — Hard rule: response is bounded by order scope * **Attorney-review gate** — Hard rule: no production without attorney sign-off * **Audit log** — Every intake/response logged with who-when-what ## See also * [Telecom eval patterns](/more-in-this-section-2/eval-patterns-11.md) * [Telecom compliance](/more-in-this-section-2/compliance-11.md) --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.layerlens.ai/more-in-this-section-2/scenarios-11.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.