What AI Automates in Customer Support, and How to Offshore the Rest

In January 2024 a customer went looking for a missing parcel on DPD’s website and came away with a chatbot that had sworn at him and written a haiku calling the company useless. DPD switched the AI part of its bot off straight away and put it down to a system update.

People pass that story round as a joke. I think it’s more useful than that, because it shows exactly where the line sits. AI can do a lot of support work now. It can also embarrass you in front of a customer at 2am with nobody watching.

So this isn’t a “will AI replace your support team” piece. It’s a practical look at where AI automates customer support well, where it doesn’t, and how UK firms are handing the human part to a non-voice customer service agent or live chat agent working from India.

What does AI automate in customer support today?

Start with your own inbox. Pull last month’s tickets and count them. You’ll usually find the same handful of questions over and over: where’s my order, how do I send this back, can I change the address, why was I charged twice.

That’s the bit AI handles well now. In practice it can:

• answer your top twenty questions at any hour, using your help centre content
• read a new email, tag it as billing, delivery or a fault, and drop it in the right queue before anyone logs on
• write a first-draft reply that an agent checks, edits and sends
• summarise a long chat when it’s handed to a person, so the customer doesn’t start again
• do simple lookups like order tracking or rebooking, if your systems let it

Big firms have pushed this a long way. In 2024 Klarna said its AI assistant was doing the work of 700 full-time agents, and the company’s headcount fell from about 7,000 to 3,000. Gartner goes further, predicting that agentic AI will sort out 80% of common service issues on its own by 2029.

Look at the word “common” in that Gartner forecast, though. And look at what Klarna did next.

Where does AI still fall short?

In May 2025 Klarna started hiring human agents again. Its chief executive, Sebastian Siemiatkowski, has since said the company leaned too hard on cost and that anyone who wants a human should be able to get one. That’s not a small firm guessing. That’s the business everyone held up as proof AI support works, changing its mind.

Customers had been saying it for a while. Gartner asked 5,728 of them, and 64% said they’d rather companies didn’t use AI for service. Just over half (53%) said they’d think about going to a competitor. Their biggest worry wasn’t the tech. It was that reaching an actual person would get harder.

There’s a legal angle too. In February 2024 a Canadian tribunal told Air Canada to pay a customer CA$812.02 after its website chatbot gave him the wrong rules on bereavement fares. The airline argued it couldn’t be held responsible for what its chatbot said. The tribunal disagreed. If your bot invents a returns policy, that’s your returns policy now.

Put all that together and the weak spots are fairly obvious. AI struggles when:

1. 1the customer is upset and needs to feel heard, not processed
2. 2the answer is a judgement call, like a goodwill refund for someone who’s spent £4,000 with you
3. 3the problem runs across three systems: your shop, the courier’s portal and the payment provider
4. 4it doesn’t know the answer and guesses anyway

Why offshore the human side of support?

Once the bot has the repeat questions, what’s left is written, fiddly work that needs a bit of judgement. Emails, chats, notes in the CRM. None of it needs someone in your office, and that’s why it offshores so well.

Getting it wrong isn’t cheap either. The Institute of Customer Service puts the cost of sorting out service problems at £7.3bn a month across UK businesses, counted in staff time. That works out at around four days a month per employee. Quicker, better human follow-up is where a lot of that time comes back.

I’m not suggesting you replace your UK team. The idea is to have the queue cleared every morning, so they can spend their day on escalations and the customers you really can’t afford to lose.

Which customer support roles should you offshore?

Four roles cover most of it.

One thing I like about non-voice support: it’s all written down. You can read every reply your new agent sends in their first week. Try doing that with phone calls.

How do you offshore customer support, step by step?

Here’s the order I’d do it in.

1. 1Sort a month of tickets into three piles. The bot can do it; the bot drafts and a person approves; person only. That’s your playbook.
2. 2Write your escalation rules down. Be specific: hand over if a refund is over £100, the word “complaint” appears, or the customer has replied to the bot twice. Vague rules are how people get stuck in bot loops.
3. 3Hire one person first. Give them a slice of the queue and compare their numbers with your in-house team for a month.
4. 4Put them in your systems, with their own login to your help desk, CRM and chat tool. Not a third-party platform you can’t see into.
5. 5Train them on real examples rather than a script. Your tone of voice guide, your top macros, a few replies you’re proud of and a few you’re not.
6. 6Sort security before day one. UK GDPR treats overseas access to customer data as a restricted transfer, so you’ll want the ICO’s IDTA or UK Addendum in place. Aspire Offshore works from ISO 27001-certified facilities.
7. 7Read conversations every week at first. Track first response time, resolution time, CSAT and how often things get escalated, then drop to monthly once it’s steady.

What does a good AI-plus-offshore setup look like?

Take an online homeware shop handling around 1,500 support conversations a month. (This is an illustration, not a client case study.)

The chatbot covers tracking and returns questions overnight. By 9am UK time there’s a pile of tagged tickets it couldn’t close. The offshore agent works through them, checking each AI draft before it goes out. A cracked lamp, a sofa missing a leg, a regular who’s fed up with a late delivery: all handled by a person before lunch.

Meanwhile the UK manager isn’t firefighting. She’s on the phone to the courier asking why so much is arriving damaged, which is the bit that actually cuts next month’s tickets.

Summary

AI automates customer support well wherever speed matters more than judgement, so use it there. But DPD, Air Canada and Klarna have each shown, in different ways, what happens when you push it past that point, and customers have been clear they want a person when it counts. Let the bot shrink the queue and give what’s left to someone who knows your customers. If you’re weighing up your first offshore support hire, see how our offshore staffing works or book a consultation and we’ll talk through which role fits.

“Let AI handle the repeat questions. Tell us the support roles you need, and we’ll recruit and set them up in India, working your hours.Book a consultation →