Consumer AI agents can now call retention desks to negotiate bills on subscribers’ behalf. The Totogi Ontology enables an equivalent response that won’t give away your margin. ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­    ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­  
View in browser
Newsletter Banner Refresh

You’re talking to a bot

 

Last month a guy named Mike P told his Grok bot to figure out how to pay for itself. It called Xfinity and knocked $43 a month off his bill, then promised to audit next month’s statement. He put it out on X and the post got 1.7 million views in four days.

 

Mike’s bot won because retention has always been priced on a bet that most subscribers never call. It’s true for humans, but bots always call, and they always audit. Even scarier, they’ve read your playbook before they dial, so they know your moves before your rep does.

 

For a telco, moving slowly used to mean safe. Now it means exposed. In this week’s blog, I lay out three moves to get a machine on your side of the call, because the bots on the other side are getting better every week.

Ep152 Max Gasparroni Fastweb + Vodafone Promo

Episode 152

Fastweb + Vodafone charts a new course

 

Every merger creates a new entity with two of everything and years of work streamlining systems. Max Gasparroni, chief technology and security officer at Fastweb + Vodafone, is in the thick of it right now. He inherited up to 450 systems to rationalize and tens of thousands of nodes in the mobile core, and he’s rolling out AI at the same time. Is he struggling? No way. He’s more than halfway through and ahead of plan. In this episode, we cover how Max is approaching the hundreds of overlapping systems, why AI is “an existential journey” for the team, and why he wants less data, not more.

 

LISTEN NOW: Apple Podcasts, Spotify, YouTube, TelcoDR website

What I am doing-1

Hyperscaler event season is approaching fast. Microsoft Ignite is in two weeks, November 17-20, in San Francisco. AWS re:Invent is two weeks later, November 30-December 4, in Las Vegas. These shows are a great way to catch up on the latest with public cloud and AI. Even if you’re not going in person, you can watch a lot of the content online. The session catalogs for both shows allow you to filter by telecommunications industry, making it easy to sift through the hundreds of options to find what you want. Most importantly, be sure to watch the announcements that come out during the events, as they’re generally the biggest of the year.

Moves in the cloud-1

SpaceX just keeps acquiring more spectrum. This time, it’s the three AWS-3 licenses from Tristar License Group, covering the 1695-1710 MHz band across 62 counties in eight US states. The FCC approved the sale, and the licenses, which cover about 4% of the US population, fill the gaps in SpaceX’s existing AWS-3 holdings from EchoStar and the FCC auction, bringing it close to nationwide coverage in the band. This news is hot on the heels of the Grain Management spectrum sale I covered in my last newsletter.

 

And that’s not all: SpaceX is reportedly talking with US tower companies, including Crown Castle, to secure access to at least 80,000 cell sites as it develops Starlink Mobile. The planned network will combine satellite connectivity with ground-based infrastructure, with a direct-to-device (D2D) service launch targeted for early 2028. Then, this morning, the Washington Post reported that Musk met with EchoStar’s Charlie Ergen in October about buying Boost Mobile or using its network, after AT&T, T-Mobile, and Verizon all refused him an MVNO deal. Analysts are still arguing about whether Musk is building the fourth national carrier or angling for an MVNO deal. I think he’ll take whichever gets him ground coverage first. Spectrum, towers, a prepaid brand with 7.4 million subscribers: every piece is a means to the same end. Satellites overhead plus small cells on the ground is the only network architecture that works on Mars, a planet with no towers and no buildout crew. Paying subscribers in rural Texas are how SpaceX funds and tests it before the first settlers land. I made the full case in my last blog. Check it out!

 

AI is raising the stakes on both sides of the telecom business. Security firm Calif used OpenAI’s GPT-5.6 Sol to find two critical PHP flaws that let unauthenticated attackers run code on vulnerable servers. Roughly 70% of the world’s websites run PHP, and Cisco estimates about half of the internet-facing instances are on versions too old to receive the patch. Governments in Finland and the Netherlands took sites offline until they could fix them. Meanwhile, a subscriber used a Grok agent to negotiate a lower phone bill. Attackers can now turn a patch into an exploit within hours, and customers now send agents to your call center. Both stories point to the same imperative: operators have to run AI at the speed of their adversaries AND their subscribers. For more on this, see my latest blog.

 

Telcos don’t need perfect data for autonomous operations, according to Ericsson’s head of cloud software and services for EMEA, Lucky La Riccia. You just have to build guardrails around the messy data and prioritize the domains where automation delivers clear business value. Those guardrails sound a lot like the Totogi Ontology I’ve been promoting for years, with one difference: a guardrail catches the agent after it has already reconciled 200 definitions of “customer” at runtime, while an ontology decides what actions are allowed before the agent ever runs. I guess it’s good that Ericsson’s figuring AI out, but Lucky is still advocating for human oversight, a “manager in the loop” model, which will save you exactly nothing and leave you tied to the same old slow processes. Again, use an ontology to nail your AI architecture.

 

At the same Ericsson summit in Zurich, Swisscom CTIO Mark Düsener gave the keynote, and he went further than his host. AI agents without data and context, he said, “are blind and they will hallucinate by default.” He had the numbers to show why: Swisscom has 88,000 APIs, about 40,000 of them catalogued and about 30,000 up to date, so the agents have no idea what most of the estate does. His fix is six years in and counting: 15 IP networks converging to one by March 2027; a common Swisscom data model across BSS, OSS, and process IT; and a BSS rework to cut service integration from months to a week. Asked when Swisscom will be done, he said never. He’s right, and that is the argument for buying the 90% of the context layer every operator shares instead of building all of it yourself. Catch the recent interview I did with him on the Telco in 20 podcast.

 

KT announced an AI-focused dedicated-line service offering up to 800 Gbps to connect distributed AI data centers and other major sites. The service combines high-capacity fiber connectivity, optimized routes, and an all-photonic architecture designed to minimize latency. It echoes what Shawn Hakl and I talked about on the podcast a few months ago following AT&T’s AWS Interconnect last-mile announcement: AI needs more than computing power. It needs better connections. KT is selling links between AI sites and AT&T is selling the enterprise-to-cloud path, and both are selling the one part of the AI stack that holds its value when the GPUs inside the data centers get written down. Expect every Tier-1 with a fiber footprint to have a product like this within a year. This is what AI CapEx looks like when a telco spends it on the asset it already owns.

 

Gartner warns that the rapid expansion of AI data centers is financially unsustainable, with more than $3.5 trillion of infrastructure spending exposed to impairment by 2030 if investment keeps outpacing revenue. The math is simple: data center owners need $8.5 trillion in cumulative revenue through 2030 to cover their spending and a 12% return, and Gartner forecasts no more than $3.6 trillion. Demand for AI is real, but intense competition is keeping prices low, and businesses and consumers are struggling to find enough value to justify the spend. Gartner also expects the current generation of AI chips to be replaced before most providers recover their investment, and more than 80% of neoclouds to lose their independence by 2030. For a preview of how this plays out for a telco, look at Indosat Ooredoo Hutchison, one of the best-run AI-native operators in Asia. It will book about $100 million in AI cloud revenue this year, up from $28 million in 2025, with $300 million already contracted for 2027. It also put roughly $570 million into AI cloud CapEx in nine months, pushed its CapEx-to-revenue ratio to 63%, and watched its stock fall 6.7% on results day. The hyperscalers and neoclouds are following a “double ’til you lose” betting strategy. Telcos: Are you sure you want to play this game?

 

Data centers are a hot-button issue among voters these days, but they aren’t the number one topic at public city council meetings across the US. What is? Pickleball courts! Turner Novak shared the data on X, pulled from Hamlet, a site that tracks public comment at local government meetings. Neighbors turn out to fight the noise from a few new courts the same way they turn out to fight a new server farm, and the same two complaints come up in both: too loud, too close. For the record, I’m pro-AI AND pro-pickleball. At least the data center doesn’t cheat on line calls. It just occasionally hallucinates. 🏓

Newsletter-2

You are receiving this email because you opted in via our website.

TelcoDR, 3801 N Capital of Texas Hwy. Suite E240, #76 Austin TX United States 78746 

Totogi LLC, 3 Germay Drive, Suite E240, #76, Wilmington,DE,19804,United States,

Unsubscribe Manage Preferences

Copyright (C) 2024 TelcoDR All rights reserved. Privacy Policy