Threll.ai
AI customer service

AI customer service by phone: what it solves, what it hands over, and how to measure the difference

AI customer service by phone takes the calls that come in again and again – order status, returns, opening hours, rebooking – and puts the rest through to a person with a summary. Here is how it works, where it gets its answers, and which numbers show whether it is helping.

ByKjell Rune Monsø6 min read
A young man with fair hair in a teal hoodie and navy jacket stands in a queue at a parcel pickup counter with a phone to his right ear and a small cardboard box under his left arm, two other people waiting behind him in front of shelves of parcels.

It is 09:12, and fourteen people are in the queue. Almost half are calling about the same thing: where is my order? The answer is in the order system. It takes a team member thirty seconds to look up. But the caller has already waited nine minutes.

That is where AI customer service by phone makes the biggest difference. Not in the hard calls, but in the many simple ones that stand in their way.

How it differs from a chatbot

Many already have a chatbot on their website and wonder why the phone should be any different. It is, for three reasons:

  • Callers have often given up on the website already. The call starts with little patience, and the agent has to get to the point straight away.
  • Speech is harder than text. Order numbers, names and addresses have to come through correctly over a phone line that throws away part of the voice, sometimes with other voices in the background.
  • The call happens in real time. Callers won't wait for an answer that arrives in a minute. The agent has to keep talking while it looks things up, and know when to ask the caller to hold for a moment.

What it solves

The calls that suit it best have three things in common: they come up often, the answer is in a system or a document, and there is a clear right answer.

  • Order status and delivery. The agent looks up the order and says where it is.
  • Returns and complaints. It explains the process, takes the case and sends the instructions by text or email.
  • Bookings and rebooking. It finds a free slot and books it in the calendar.
  • Questions about opening hours, prices and terms. Answers that live in your own sources.
  • Identification. It looks the number up in your customer records to know who is calling and what you have talked about before.

How many of your customer service calls are the same question again?

Fifteen minutes going through your most common matters, with an honest view of which AI customer service can solve all the way.

Book Meeting (opens in a new window)+47 24 12 66 30

15 minutes · no obligation · pick your own time

What it hands over

Just as important is what the agent should leave alone. Upset customers who want to be heard by a person, matters that need a decision outside the rules, and questions the agent has no material for. Those it should put through – with a summary, so the customer doesn't have to repeat themselves.

How that handover looks shapes much of how customers experience the whole thing. We cover it in designing escalation from voice agent to human, and where a call can go wrong in one customer call, beat by beat.

Where it gets its answers

AI customer service shouldn't make things up. It answers from what you have given it: your knowledge base, your terms, the order system, the calendar. In practice that means two things.

First, the answers are never better than the material. If the returns policy on your website is out of date, the agent will give out-of-date answers.

Second, when the agent doesn't know, it should say so. The questions it couldn't answer are one of the most useful lists you get from it – they show what is missing from the material.

Measure what matters

Number of calls answered is a poor yardstick. The agent answers all of them, so that number tells you nothing. Look instead at:

  1. First-call resolution. How many matters the agent handled all the way without anyone having to call back.
  2. Share handed over, and why. Group the reasons. Some are because the matter genuinely needed a person. Others are because the material was missing, and those you can fix.
  3. Repeat calls. If the same customer calls again about the same thing within a couple of days, the first answer wasn't good enough.
  4. Queue time for calls that go to people. This is where the effect shows most clearly: when the simple calls leave the queue, the wait gets shorter for everyone else.

Read a dozen transcripts a week for the first few weeks. The numbers show where things grate; the transcripts show why.

Start without changing everything

Let the agent start with the overflow: calls that would otherwise have gone to the queue, or that come in outside opening hours. Nothing changes for the calls your team already takes. How it connects to your existing phone system is described in an AI switchboard.

Read more on the AI customer service page.

Frequently asked questions

Yes, if it is connected to your order system or e-commerce platform. Threll has ready-made integrations with Shopify, Magento and Intercom. Other systems can be connected through the API, or we build the integration on request.

It looks the number up in your customer records or CRM at the start of the call. If the matter needs secure identification, that should happen before the agent shares any details, so nothing reaches the wrong person. Ask us which methods are supported for you.

Calls are processed and stored in the Nordics, with European providers, under GDPR. You get a data processing agreement, decide yourselves how long recordings and transcripts are kept, and can delete them at any time.

The setup itself takes an afternoon. Getting it really good at your particular calls takes a few weeks of reading transcripts, filling in material where it didn't know the answer, and adjusting when it should put calls through.

Related Articles

Continue reading more articles

See all articles