British Gas Says Customers Prefer Chatbots. Its Own Numbers Say Something Else.

Centrica, the owner of British Gas, confirmed this week it is cutting around 1,300 jobs over the next two years: roughly 500 call centre roles across Glasgow, Edinburgh, Cardiff, Leicester, Stockport and Leeds, plus a further 800 in support functions. That's about 14% of its customer operations workforce. Announcing the cuts alongside half-year results, chief executive Chris O'Shea's explanation was that customers prefer digital channels. Over 90% of customers now go digital first, calls are down around 20% year on year, and nearly nine in ten enquiries are handled through self-service: the website, the app, the chatbot, WhatsApp.

At the same time, the company insists AI isn't the cause. O'Shea said the investment in AI and technology "isn't driving these particular job reductions", and a company spokesman went further: "The suggestion these roles are being replaced by AI or new chatbot technology is simply wrong." This, in the same announcement cycle where 800 of the cuts sit alongside what the company describes as a targeted deployment of AI tools.

For me, this announcement is worth studying not because the operational logic is wrong, but because the way it's being explained is doing the company real damage. There are three separate problems here, and they're ones I see repeated across almost every large customer-facing AI programme.

Channel usage is not the same as preference

The core claim, the one the headlines ran with, is that customers prefer digital. The evidence offered is usage: 90% go digital first, call volumes are falling.

Anyone who has run a large consumer operation knows what's wrong with this argument. Customers use the channels you put in front of them. If the app is prominent, the phone number is buried three menus deep, and the IVR queue is 40 minutes at peak, digital usage will rise. That's channel design working as intended. It tells you nothing about what customers would choose if the options were equally easy.

The survey data we do have points the other way. Research published by 8x8 in late 2025 found 83% of UK consumers prefer speaking to a real person for customer support, and only 4% prefer a virtual agent or chatbot. The same research found something more useful for an operator: over 30% said they'd accept AI handling their query if it reduced their costs. That's the honest trade sitting underneath this whole story. Customers don't prefer chatbots. They prefer cheaper bills, and some of them will tolerate chatbots to get there.

There's a version of this announcement built on that honest trade. It sounds like: routine contact volumes are falling, we're automating the transactional work, and the savings are part of how we keep standing charges down. That's a defensible position a CEO holds under questioning. "Customers prefer chatbots" is not, because everyone listening, including every British Gas customer who has sat in a call queue, knows it isn't true of them.

You cannot claim the AI savings and deny the AI cause

The second problem is the internal contradiction. Centrica wants credit with investors for deploying AI at scale, and it wants the job cuts understood as unrelated to AI, driven purely by changing customer behaviour. The GMB union's national secretary called the cuts "an absolute disgrace", saying jobs are being handed to chatbots. The company's response was to call that suggestion "simply wrong".

Both positions are doing work the facts won't support. If nearly nine in ten enquiries now resolve through self-service channels, and one of those channels is a chatbot, then automation is part of why fewer people are needed. Pretending otherwise insults the intelligence of the workforce and hands the union an easy win, because the contradiction is visible to everyone. The 31% fall in inbound contact since 2023 didn't happen independently of the technology the company built to absorb the contact.

This matters beyond public relations. The employees still in those contact centres are reading the same coverage. When leadership says the cuts are unrelated to AI while announcing AI deployment in the same breath, the remaining workforce concludes leadership will say whatever is convenient when their turn comes. Every future change programme at Centrica now starts from a lower base of trust. I wrote about this dynamic in the Kroger surveillance pricing case: when you leave a gap between what you're doing and what you're saying, someone else fills it with the worst available interpretation, and theirs is the version people remember.

The results this needs to be judged against

The third problem is the scoreboard. Centrica's retail division made £346 million in the first half, up from £338 million, while the customer base slipped from 7.5 million to 7.45 million. The company is also paying out £112 million in compensation over force-fitted prepayment meters. So the context for "customers prefer digital" is a business holding profits up while slowly losing customers, in a market where its most effective competitor is famous for answering the phone.

That competitor comparison is the uncomfortable one. Octopus Energy deployed AI across its customer service operation years ago. Its tool drafts over a third of customer emails and, by CEO Greg Jackson's account, was doing the work of around 250 people within months. Octopus made no redundancies, moved agents onto phones and complex cases, and kept growing its customer base, much of it taken from the incumbents. Same technology, same market, opposite deployment logic: Octopus used AI to make humans more available for the contacts where a human is worth something. Centrica is using it, whatever the official framing, to need fewer humans.

Neither approach is automatically right. Cost-led automation is a legitimate strategy, and Centrica's contact volumes genuinely are falling. But we've already seen the endpoint of automating customer contact past what customers will accept. Klarna replaced 700 agents with AI, celebrated the savings publicly, and fourteen months later its CEO admitted they'd gone too far and started rehiring, repositioning human support as a premium differentiator. The test Centrica has now set itself is simple: if customers truly prefer the digital channels, satisfaction and retention should hold or improve as the contact centres shrink. If the customer base keeps eroding towards competitors who answer the phone, the revealed preference will have spoken, and it won't be the one in the press release.

What I'd take from this

If you're planning a customer-facing AI programme, three things fall out of this week's news.

Don't launder a cost decision through a customer preference claim. If the driver is cost, say cost, and connect it to something the customer gets back. The claim has to survive contact with your own customers' lived experience, and "you prefer the chatbot" doesn't.

Decide what your displaced capacity is for before you announce anything. IKEA studied what its chatbot couldn't do and retrained 8,500 agents into a design advisory service now feeding a EUR 1.3 billion channel. Octopus moved agents to complex cases. Both had an answer to "what happens to the people" built into the story. Centrica's answer arrived via the GMB instead.

And treat the announcement as part of the operating model, not an afterthought. The communications around workforce-affecting AI decisions shape adoption, trust and resistance for years. Section 11 of the playbook covers how to design that messaging deliberately. Centrica this week is a live demonstration of what it costs when you don't.

The irony is the underlying operational story here is reasonably strong: falling contact volumes, high self-service adoption, continued investment in engineers and 500 apprentices. Told straight, it's a defensible transformation. Told as "customers prefer chatbots", it became a story about a profitable company cutting 1,300 jobs and blaming its customers' preferences. When you deploy AI into customer service, the technology is the easy half. The story you tell about it is the half that ends up in the headlines.

Sources

• Customers prefer AI chatbots, says British Gas owner as 1,300 call centre jobs axed (The Guardian, July 2026)
• Centrica cuts 1,300 jobs as British Gas accelerates digital overhaul (London Loves Business, July 2026)
• Centrica to axe 1,300 jobs in next two years (Personnel Today, July 2026)
• British Gas owner confirms 1,300 jobs are being cut (GB News, July 2026)
• Report: UK Customers Still Prefer Human Support Over AI (Business Wire / 8x8, October 2025)