Without a change in approach, there is a risk that telco websites simply become price comparison mechanisms for AI agents. That is the warning from Tom Cox, founder and CEO of Humara, who has spent the best part of 15 years building software that sells mobile contracts online for UK and US operators.
“For prospects, you become a checkout shop,” he told TelcoForge recently. “And that’s obviously terrible for a telco, because then you’re just a commodity, you’re just a price. You haven’t been able to market to them.”
Most of the industry’s AI conversation is largely about what the technology can do inside the network. Cox is concerned with how AI plays out for the consumer. More to the point, what happens when the one selecting and arranging a service contract is no longer even a person.
Painful Contractions
Cox’s vantage point is unusual. He started 15gifts in 2010 as a gift finder that people could only talk to: no browsing, no catalogue, just a conversation. Virgin Media saw it and asked whether it could be licensed, and paid for the first build. Over the following decade it became a guided selling tool used by increasing numbers of telcos. The product was relaunched as Humara in April 2025, and the company took the name this February.
As you can see, Cox has spent a long time watching millions of people try, and often fail, to buy a phone contract online. His view of AI starts from that problem rather than from the model. Too much of the industry, he says, is “leading with AI and agentic” rather than “starting with the problem and then seeing if it fits or not.” Examples of AI “practically applied to a real telco problem, and then it being proven,” he adds, “are rarer to find.”
The problem Cox kept running into was confidence. Signing up for a handset on a multi-year contract is a real commitment, and the buyer has to stay above what he calls a ‘confidence bar’ at every step. Telco buying processes are long and full of friction, and most people dip below the line somewhere. “Either it’s upper funnel or lower funnel, they will end up dropping away,” he said. “They’ll either reach out to a human or they’ll just procrastinate.”
The company’s answer was to study what the best human sales staff actually do – how they build rapport, run discovery, recommend, handle objections – and rebuild that in software across a variety of specialised agents.
The multi-agent design is becoming more common, but most operator AI spending on customer engagement goes into service: deflecting service calls, fixing bills, resetting routers. Cox is building for the sale in an industry whose digital channels have treated buying as a form to fill in rather than a conversation. Most objections, he says, are about price, and the agent that handles them has become an expert in price negotiation.
Losing Control
Cox sees a divide among the operators he works with. A minority say they are seeing “material, noticeable declines in traffic” as discovery moves into AI assistants. They are treating it as a major initiative, going beyond optimising content for AI search to asking how they could “publish our own agents within those platforms and control the conversation.” The majority, he says, regard AI-referred traffic as “absolutely tiny… negligible,” and are putting their effort into the existing base.
Cox explained why the majority may be looking in the wrong place, to do with the nature of AI-led discovery: “It’s all prospects, by default,” he said. An operator focused on its base will notice the change least, right up until it needs to grow.
Cox expects the visitors who do arrive via AI to do so later in the buying journey, having done their research elsewhere. Adobe’s data supports that in the retail sector, where AI-referred visits now convert 42% better than other traffic. That sounds like good news, but Cox thinks it is the opposite.
“You lose control, basically. It’s just a price game,” he said. “These big telcos, they’re never the cheapest. They’ll always lose on price. They rely on marketing to win. And if you take marketing away from them, and it becomes just the price in an agent search, then they’ve got nothing.”
This is the part the industry has not yet reckoned with. Operators know that emotional loyalty lies elsewhere. “Your loyalty is with the hardware provider. It’s with Apple or Samsung,” as Cox puts it. While Ofcom found only 16% of UK mobile customers switched provider in the past year, much of that can be explained by a combination of two-year contracts and sheer distaste for the process of contract switching. Agents choosing to switch on a customer’s behalf removes both the impact of marketing and inertia. It is not moved by the brand campaign, and it does not find switching tedious.
The price argument needs one qualification. Many major telecoms brands have low-cost brands or MVNOs of their own. An agent hunting for the cheapest deal may still land the customer with the same group, but it still reduces their margin; the opposite of what we’re looking for.
Entering The Chat
Cox’s suggested response is twofold. The first is to recognise a visitor who has come from an AI assistant and keep talking to them the way they were just being talked to. “They’ve just come from conversational, they’re going to want to carry on,” he said. So instead of dropping them into a product grid, the site makes conversation the main interface, “handholds them to the end,” and uses that conversation to put back what the agent stripped out: the “why Vodafone,” as he put it.
Retail has already learnt that buying inside an AI is harder than it looked. OpenAI moved purchases out of ChatGPT and back to merchants’ own apps in March, after finding that people researched in ChatGPT but did not buy there. Telco purchases involve credit checks, identity verification and number porting, so buying entirely through an assistant is a long way off. For now the pattern is “discover in AI, buy on site,” which makes the operator’s own site the place where the final purchase argument gets made.
The second part is to go where the conversation is happening, by publishing a sponsored agent inside the AI platforms so customers can talk to them directly. That has only just become possible. OpenAI announced Sponsored Agents on 18 September as a US test with selected advertisers; the agents can advise and link out, but cannot complete a purchase. “That’s so new and evolving,” Cox said. “That’s a real area of discovery for us.”
If an AI’s recommendation replaces the shop window, the brand needs its own agent where the choice is made. That agent also has to be good enough at persuasion to hold its own against a sheer price comparison engine.
Protecting the Bottom of the Barrel
Cox’s argument is about sales, but TelcoForge has been seeing a similar shift play out elsewhere.
In August we argued that networks are becoming increasingly separable from the customer relationship, pointing to brand-led MVNOs from Klarna, Lidl and Revolut. AI agents are a second route to the same place: the operator can keep the network but lose the customer.
Meanwhile Rakuten’s David Soldani warned of a “control illusion” in network operations, where vendor AI runs on logic the operator “cannot see, cannot change, and does not own.” Cox is describing the same illusion at the shop front. An operator may believe it owns the customer relationship while the decision is shaped by a recommendation engine it has no view into.
Soldani’s suggestion of a “guarantee economy,” pricing on verifiable outcomes rather than megabytes, was pitched as a revenue model for 6G; but verifiable outcomes are exactly what an AI agent can read and compare. A brand campaign does not last through an agent’s price comparison; a published, measurable performance guarantee might.
We may also be missing tricks to build those guarantees and good feelings. Tecnotree’s Prianca Ravichander recently described an operator’s AI agent sorting out her broadband at 2am. Her reaction: “I’d pay for that, but it was completely free.” That is the pattern Cox’s argument exposes. Operators deliver value that never becomes a reason to choose them.
Scam protection is a similar example, from the author’s experience. My UK phone flags likely scam calls, and I have no idea whether that is my operator, my handset or regulation.
(In practice it could be any of them. Ofcom has required providers to block overseas calls spoofing UK landline numbers since January 2025. EE and Vodafone sell scam-call labelling as £2-a-month add-ons, while Samsung’s handsets offer similar labels from Hiya, the same company behind EE’s service.)
Without clarity on who is helping me filter out time-wasting and fraud, I don’t know who to be grateful to and the chance for relationship improvements are lost.
Operators have shown they can turn protection into brand value. O2’s “Daisy,” the AI granny who wasted scammers’ time, raised satisfaction with O2’s fraud efforts by 42% on £20,000 of paid media. But that was a campaign, not a lasting reason to pick a network. At a recent industry meeting in Rio, TNS reported SMS phishing up 14 times in a year. Protection is becoming more valuable at exactly the moment agents threaten to render invisible any value an operator is not shouting about in the moment.
TelcoForge’s 6G Value Season starts from the premise that “6G technology isn’t valuable. What we do with it in the 2030s might be.” One of the questions it asks is what the 5G era can teach the next generation’s commercial proposition. We might think about this: 5G struggled to gain a consumer premium because it wasn’t clear what changed between 4G and 5G for them. 6G will likely be sold into a market where the customer may not look at all.
Operators need a voice of their own wherever someone is making a purchase choice, whether the chooser is a person or a machine. They also need to turn what their networks do into claims agents can verify. Otherwise, there is a risk that the price race to the bottom will only intensify.
