At Elsevier, we run a global customer service operation with around 300 agents supporting academic researchers, doctors, research leaders, and R&D pharma companies. They rely on our products every day to do serious, meaningful work. 

So when something goes wrong with access, or when they can't find the information they need, the stakes feel real.

For years, we had a problem I suspect a lot of you will recognize: our agents knew exactly what customers were contacting us about. That knowledge was sitting right there in the room. But we couldn't translate it into data. We couldn't take it upstairs. We couldn't use it to drive change across the business. It was anecdotal, and anecdotal only gets you so far.

We've been on a deliberate journey to change that. We introduced a set of analytical tools designed to extract genuine insight from customer conversations and journeys, then convert those insights into concrete initiatives. 

Here's how we did it, what tools we used, and what we actually built as a result.

The problem with support agent tagging

Before getting into the tools themselves, it's worth pausing on why we needed them in the first place.

Like most support operations, we used to rely on our support agents tagging tickets to understand contact reasons. In theory, that makes sense. In practice, it rarely works. Agents want to move on to the next ticket – they're not thinking about data integrity in the middle of a queue.

What we found, consistently, was that the first option on the dropdown was the most frequently selected. Not because it was the most accurate, but because it was the easiest.

That's not insight, that’s just noise. And when you're trying to build a business case for investment or change, noise doesn't help you.

What we needed was a way to understand what customers were actually saying, in their own words, without relying on agents to interpret and tag in real time.

Interaction analytics: Letting the conversations speak

The first tool we introduced was interaction analytics. 

The mechanics are fairly straightforward: it takes every verbatim interaction from phone calls, transcribes them into text, then applies analytical tools alongside chat and email data to extract patterns and themes. In the past, that meant keyword identification. Increasingly, it means large language models doing the heavy lifting.

The output tells you:

  • How many times a particular issue came up
  • How customers felt during those interactions, based on the language they used
  • How those contact reasons shift over time

You can visualize all of this in dashboards, spot peaks and troughs, and catch a specific issue starting to spike before it becomes a crisis.

The real value, though, is in the depth of understanding you get. Because you're working from customers' actual words, you can go much further than a tag ever allowed. You can understand what the problem really was, what the customer was feeling, and what your team did to resolve it. That's a fundamentally different quality of information.

Adobe Analytics: Understanding the digital journey

Alongside interaction analytics, we deployed Adobe Analytics to track how customers behaved on our support hubs. Other tools do similar things, but the principle is the same: you're watching what customers actually do when they try to help themselves online.

How many clicks did it take them to find an answer? Did they eventually give up and click on a contact option? That last action is particularly telling. When a customer moves from browsing your support hub to initiating a chat or picking up the phone, that's a signal – they didn't find what they needed, and your self-service content failed them in that moment.

Tracking that journey lets you understand how effective your support hub genuinely is, not based on assumptions, but based on observed behavior.