AT&T announced in late October that it will serve all of its home internet, phone and postpaid wireless customers through care staff based in the United States by the end of 2027. The company did not say how many roles move. CX Dive reported the plan on 4 November 2026, alongside a proposed FCC rule requiring a share of customer service calls to be handled from US call centres.

The claim is “jobs come home”. The cost comes home too.

The story is being read as a win for American jobs. I read it as a cost decision first, because that is what it is for anyone running a service operation.

King White of Site Selection Group puts US outsourcer rates at US$35 to US$40 an hour against US$12 to US$14 in the Philippines. Michael Replogle says labour is 70% of a BPO’s profit and loss. Move the same volume of calls from the cheap location to the expensive one and every conversation costs roughly three times as much. Nothing about the call changed, apart from the price of the person answering it.

The location changed. The volume of routine enquiries did not. Unless something else takes those enquiries away, reshoring is a bigger bill for the same work.

Reshoring only makes sense after AI has emptied the queue

The article’s experts point at what makes the sums work. Replogle says the first wave of AI is taking the issues that can be handled easily and raising the complexity of what is left. Maribel Lopez of Lopez Research describes fewer agents onshore working with AI, at the same cost as a larger offshore team.

My view is that a competent AI customer service automation can already handle 80 to 90% of enquiries. Where is my order, can I change my delivery date, what is your return policy, is this in stock. Those are transactional questions. They do not need a person, even when they run across several steps and systems. AeroChat, which I build, is my test case: for one bakery merchant it handles 87 to 94% of conversations with no person, through festive peaks.

Once that layer is in place, the people you keep answer the 10 to 20% the automation handed over: the angry customer, the big order, the complaint that could become a public review, the enquiry that could become a long-term account. Those conversations are worth far more than US$35 an hour. Paying offshore rates for them was always the wrong trade. Paying US rates for “where is my order” is the wrong trade in the other direction.

Reshoring, then, is the second step. The first step is making sure the routine work has already left the building. Do it the other way round and it is a cost decision you will regret within a year.

What this looks like for a business much smaller than AT&T

Most readers of this post run a clinic, a shop, a service business or a small marketing team, not a contact centre. The same logic applies at every size.

Take a clinic with two front-desk staff who spend most of the day on booking, rebooking, directions, opening hours and whether a treatment needs a referral. If the AI handles them in the channel the patient already uses, usually WhatsApp, the two people stop being a queue. They become the ones who call the patient who has not returned in a year, who sit with the anxious first-timer, who handle the complaint before it turns into a one-star review. The front desk stays the same size and gets pointed at the work that sets the clinic apart.

A shop works the same way. The AI takes order status, exchanges, stock questions and product recommendations at 11pm. The owner and the one staff member take the corporate order for 200 gift sets and the customer who wants to talk through a S$2,000 purchase.

The posture I hold is that customer service teams in their current form will cease to exist. The function moves toward customer success, winning and keeping customers, and toward service that is deliberately human where a human is the differentiator. AT&T’s decision fits that: people come back for the work that remains once the routine is gone.

What to actually do

  1. List your enquiries by type for one month and count them. Most will be a handful of transactional questions.
  2. Put AI on that transactional surface first, in the channels your customers already use, with the limits on what it can commit to, such as refunds, discounts and dates, set in the system rather than left to the prompt.
  3. Only then decide where your people go. Pick the moments where a human conversation changes the outcome: big-ticket purchases, retention, complaints, first visits.
  4. Measure what the humans now do by revenue kept or won, not by calls answered.
  5. If you outsource, re-read your BPO contract. You are probably paying for volume that should never reach a person at all.

Source: CX Dive: AT&T, reshoring and what’s next for US contact center operations