AI customer service: what to automate and when to hand over to a person

What the studies say

On 16 October, Twilio presented its study Is AI Adding Value to the Customer Experience or Taking It Away?, reported by MuyCanal. It was carried out between April and May 2026 with 7,652 consumers and 660 executives across 18 markets, and the European data come from Spain, France, Germany, Italy and the UK, as detailed by AEERC. The headline figures for Europe:

• AI already handles around 52% of customer service interactions, and companies expect to reach 60% in 2027.
• 84% of companies believe their customers are satisfied with AI-handled service. Only 48% of users are.
• 81% say bots have saved them time and 58% that they've improved over the past year.
• Half of consumers rank the option to speak with a person among their top three priorities when contacting a brand.
• 53% wouldn't trust AI to start a loan application and 51% to make a purchase or payment on their behalf.

Two days earlier, on 14 October, the European Commission published a Eurobarometer on AI in SMEs and mid-caps, according to The Brussels Times. 42% already use AI or algorithmic management, mostly in a limited and experimental way, and among those that do, customer service is the top use (47%). The barriers: cost (50%), legal uncertainty (49%) and lack of expertise (48%).

Read together, the message is clear: customer service is how AI gets into a business, and also where it's easiest to get it wrong.

Why customers aren't as happy as companies think

That gap between 84% and 48% isn't explained by the AI giving bad answers. It's almost always down to design:

• The bot doesn't know when to stop. It keeps answering something it can't resolve instead of passing it on.
• There's no way out to a person, or it's hidden behind five menus.
• When it hands over, the context is lost and the customer has to explain everything again.
• The bot answers but doesn't resolve: it gives general information instead of checking the specific order, appointment or case.

According to the same study, what customers value most is getting the problem solved on the spot and speed (40%), followed by accurate information and 24/7 availability (38%). An AI agent properly connected to your systems can deliver all four. One that only chats delivers none.

The golden rule: knowing when to hand over to a person

The most important part of the design isn't what the AI answers, but when it stops. These are the situations where the case should go to a person:

And the handover has to be done well: the person receives the full conversation and a summary of what the customer wants, their details and what's already been tried. The customer doesn't repeat anything.

What the law requires (even if it doesn't apply to you)

Spain's Law 10/2025 on customer service becomes fully enforceable on 28 December 2026, when the one-year adaptation period ends. It applies to basic service providers (water, gas, electricity, passenger transport, postal services, communications and financial services) and to companies selling to consumers with at least 250 employees, more than €50M turnover or more than €43M balance sheet. Most small businesses are out of scope, but its rules are a good bar for any AI customer service:

• It bans automated answering systems and similar means as the only channel.
• Bots must let customers request personalised service at any time, from a person who identifies themselves at the start.
• 95% of those requests must be answered, on average, in under three minutes.
• Complaints and claims must be resolved within 15 working days at most.

On top of that, the EU AI Act does apply to everyone: since 2 August 2026 you have to tell customers they are talking to an AI. We explain it in the AI Act after the Omnibus.

Where a small business should start

1. Look at which questions repeat. Go through a month of emails, calls and messages and group them. There are usually five or six reasons that make up most of the volume.
2. Automate only what has a clear answer and lives in your systems: order or case status, appointments, pending documents, opening hours, FAQs.
3. Connect the AI to your data. Without access to the CRM, the calendar or your management software, the bot can only give generic answers.
4. Define the handover before launch: when it happens, to whom, through which channel and with what summary.
5. Measure what matters: cases resolved without intervention, time to first response, escalated cases and complaints about the bot itself.

If first contacts arrive by WhatsApp or email, how to automate lead handling walks through the flow step by step.

How we do it at Amud

The AI agents we build for customer service work connected to the company's systems, so they answer with real data rather than stock phrases. They say they're an AI from the first message, have it in writing what they can resolve on their own and what they can't, and hand over to someone on the team when the customer asks or the case requires it, with the conversation and a summary ready. The AI takes the repetitive work and your team keeps what needs judgement.