Your customer service data is currency. Here's how to spend it
Intro
My name is Kateryna. Six years in customer service operations, linear career from a support agent then data analyst & implementation expert. I'm proud of doing complex setup and implementations long before Agentic AI existed.
In the last year my focus shifted to customer service systems audit. And now I am also a small content maker on LinkedIn who talks about customer service. One of my best content series so far is: if customer service tools were… [anything], where I proved my implementation experience through satirical comments and cultural comparisons. For some reason, people love it.
And when I was preparing for Dixa Connect I recalled what I wrote about Dixa in the series, and the main perception of Dixa was always as of something niche, sophisticated, not for everyone (in a good sense) while other customer service tools really suffered from my critical judgments.
But jokes aside, what I really want to achieve with my content is to validate the importance of customer service for those outside, talk about the talent of people who represent customer service, talk about software innovation in customer service, and scream about how many insights about customers, only we can generate. And this is my topic today: “Your customer service data is currency. Here's how to spend it”.
You already had this idea in mind. A support queue with 10,000 monthly requests contains more competitive intelligence than most companies get from their entire market research budget. Customers tell you exactly what's broken, what competitors offer that you don't, and what would make them stay or leave. Every ticket is a market research delivered directly to the one team nobody invites to the strategy meeting.
Outliers
“Your greatest assets are almost always hidden as outliers. I was taught to treat outliers as emerging trends, and it always allowed me to see something more.”
And of course, the single outlier can be a behavioral noise, but the reiteration is the finding. One person contacting us twenty times touching different channels can be a noise. But it also can be one significant problem generating 20 data points about the insight we need to transfer upstream.
And here I also recommend measuring interaction depth per original request. When you retrieve ticket data, there is often a split available between customer and agent interactions. Open your dataset, sort by customer interaction count per request descending, and take the top 5%. Five percent is the standard statistical cut for extreme values, and it leaves enough volume in the set for repeating patterns to show up.
Do ticket sampling (preferably manually, so that you see interactions associated with a particular case). Do these with the goal of identifying insights towards product, finance, retention. This is a great exercise. And this is the framework that already exists. It was founded by John Seddon, almost 40 years ago, and it’s called failure demand. Failure demand is contact that only happened because an organization failed to do something, or didn't do it right, the first time. The customer comes back because they have to, not because they want to. And in some support centres Seddon studied, that was up to 80% of the total volume.
Speaking the Language of Money
Dixa covers QA for you up to 100%, so half of the work is done, do the outliers' job, and do the interpretation job.
Most teams are dealing with retrospective data, you work with a moment, real time data, true customer language.
The customer who contacts you about a billing question and uses the word "understand" three times in the same request like "I don't understand why I was charged" — is telling you something about how your pricing is communicated, not just about the charge itself.
In customer service, we have learned to perceive “Customer is upset” as a call to action. Leadership, product, finance teams probably won’t. The formula is to speak the language of money and risk only. “This driver is this many tickets a month, it costs this much to handle, and here's the churn it's putting at risk.” That’s it.
“Sizing-related returns run 3,000 tickets a month at $15,000 in handle cost. 65% of them cite the size chart on the product page specifically, which means a one-time copy and image fix retires $9,750 a month in support cost permanently.
Promo code tickets run 1,200 a month at $6,000 in handle cost. 70% are paid-traffic customers whose code didn't apply at checkout, which is $4,200 a month in support spend on traffic we already paid acquisition cost to land.”
If you are not confident in your data literacy skills, fine. I highly recommend you to obtain those, considering the future of our industry. But for now, you can invent your own process that turns the customer story into a priced thing that other teams can act on, reject on the record, or accept the cost of inaction.
Hesitations
And here you might experience a couple of hesitations.
We already track ticket volume split by contact reason, why is your suggested approach better?
There is a difference between real billing questions and “your invoice is impossible to read” though both land in the billing bucket and look the same. And when we read this bucket in an aggregated way, we often think that we need more agents or more deflection. But what if we change what the measurement is done for? What if we allow ourselves more power and more influence?
Not everything is a failure, of course. Some refunds are people who just changed their mind. Some wrong addresses are honest typos. Most people happily use the thing you built and sell. The idea is to notice what even good dashboards can't show you.
We really reported global issues to product, finance or leadership, and nothing happened.
I think most of you in this room have experienced this. Support teams learned not to bother. You spotted something, you sent it up, and nothing moved. So again, the language of the insights is important. But practically, you need a backlog where these insights and the priced items will live. This is your foundation, your data repository. A Slack or Team channel, where the conversation happens, cross-team (including dedicated sponsors from product, engineering, ops, marketing etc). A meeting cadence where the knowledge transfer actually happens. A leadership report page, where the company sees what it's costing. Derived from the sources above. One-two pages. This will give your efforts political weight.
This is actually what we often recommended as a result of our audit projects. Because it had huge potential in breaking corporate silos. And of course, some teams accepted this recommendation and implemented it (i saw a couple of Jira epics). Some teams did not. And that’s normal. Sometimes you do all of that and still nothing happens. Sometimes leadership looks at the cost of fixing the global issue, compares it to the cost of just living with it, and decides to live with it. You don’t need to force the fix. But you can make it visible. At least for the next time, Finance won’t ask why your numbers are high.
Other teams (Product, Engineering, Marketing etc) will perceive this as an attack.
Sales has pipeline, marketing has lifetime value, product has feature adoption, every revenue function has a number that proves it earns its keep. Finance tracks us with cost per ticket, when our actual job is keeping the customer who was about to leave. So we already earned the right to talk.
And actually, the most dangerous meeting you'll ever be in is the one where everyone agrees. I don't know if you know this. When an investment committee is moving toward an absolute yes, some of them automatically pause the deal. Until someone is assigned to destroy it. It's called a pre-mortem. They don't wait for the startup to die and then do an autopsy. They assume it's already dead, and they write the cause of death before the check is signed. Failure demand is the pre-mortem your company is already running every day in your support queue. Read it before the market does.
Closing
And the last, I know what some of you are thinking. You're already doing the work of three people, and the last thing you want is one more thing on the list. So let me be clear about what I'm actually asking, because I'm not adding anything to your week.
We all have this predictable work that AI is built to take, and it will take it. Be honest with yourself. The value moves. For years the question was how well we closed the conversation. That question is going to the machines now, and sometimes they'll answer it cheaper than we can. AI can close a conversation. But it cannot tell you why that conversation should never have happened. AI can resolve a ticket about a stuck order. It cannot walk into a product meeting and say this one problem cost us four hundred tickets last month and here is who needs to fix it. The output of support is getting cheap. The input, the data behind it and what it means, that's the part worth having.
And this is something that should make you feel less worried, not more. That same input data is what the AI runs on. Garbage in, garbage out. So the person who understands the data is the person who makes the machine work and the person who can read what it means.
This is going to be the story of the next few years. Everyone will focus on the input side eventually. I'd rather you got there while it still counts as being early, instead of catching up after it's obvious.
And you already know how to do this. It's the whole thing we just spent twenty minutes on. Reading the outlier. Asking whether the contact should exist. Doing the split between a real request and a failure you were hired to absorb. That’s judgement, and judgment cannot be automated.
And if you want to check the power of your judgement, I really want you to write something on LinkedIn about what will inspire you at this event and what it made you rethink about customer service. Because there will be a lot of insights and fresh perspectives today. The whole Dixa Connect event is about the New CX Playbook. So share your thoughts publicly, and feel the power of our community.