Customer Service Challenges: Which Ones Do You Actually Have?

Monday morning. The queue is growing faster than the team can clear it, the same question is sitting in the inbox for the third time, and someone in chat is asking why the colleague on the phone just said something completely different than the one in the chat. You do not need a list of challenges to know that customer service is hard.

The more interesting question is a different one: which of these challenges do you actually have? Almost every team recognises all ten – but each team only has two or three that genuinely cost money and nerves. And you do not spot those by instinct. You spot them in numbers that are already sitting in your reporting.

In this article we go through the ten most common challenges and set out, for each, how to recognise it, what usually causes it and what helps first. We also look honestly at where AI makes a difference and where it does not. Knowing that saves you the projects that miss the real problem.

The quick test: which challenge do you have?

Before we go deeper, three questions. They take less than a minute and decide which part of this article matters most to you.

Question 1: How often does someone come back a second time about the same thing? If your repeat contact rate is above 20 per cent, that is your most expensive problem. Go to Ten symptoms and how to spot them.

Question 2: Are enquiries reaching you that the help page could have handled? If so, the gap sits before the ticket. Go to Why self-service is the part that jams.

Question 3: Can you say off the top of your head what your first contact resolution was last month? If not, start at Five numbers that give it all away. Everything else comes after that.

Ten symptoms and how to spot them

This table is the heart of the article. On the left, what you feel. In the middle, how you measure it. On the right, what usually sits underneath.

Challenge	How to spot it	What usually sits underneath
Rising expectations	CSAT falls although handling time has not changed	Response times that used to be fine no longer are
Waiting times	First response time climbs, queue abandonment rises	Volume peaks meet a fixed roster
Repeat contacts	Repeat contact rate above 20 per cent	Answers do not fully resolve the issue
Channel breaks	The same person appears in two channels within hours	The context does not travel
Understaffing	The backlog grows faster than it is cleared	Volume grows faster than the team
Repetitive routine work	A few topics account for most of the volume	Recurring questions are answered by hand
Scattered knowledge	Different answers to the same question	Content is maintained per channel
Inconsistent quality	CSAT varies sharply by shift or by person	There is no shared basis for answers
Fragmented systems	Agents switch windows several times per case	Missing integrations between systems
Lack of measurement	Improvement discussions stay anecdotal	Metrics are not reviewed regularly

If you can only check a single number, check the repeat contact rate. It is the only metric that measures two things at once: how annoyed your customers are, and how much work you are making for yourself. When it rises, an answer somewhere is incomplete, unclear or not traceable.

That this matters so much is not a new insight, by the way. In “The Effortless Experience” (2013), Matthew Dixon, Nick Toman and Rick DeLisi showed, on the basis of extensive CEB research, that low customer effort predicts loyalty more reliably than any attempt to delight customers. A second call about the same thing is exactly that effort.

You now know which number to look at first. If you would like a second opinion while you interpret it, we are happy to go through it with you.

Why self-service is the part that jams

This is where it gets concrete, and it is the area where most automation already happens. According to the Bitkom survey of 603 companies with 20 or more employees, 72 per cent of AI-using companies in Germany apply the technology in customer contact, for instance in handling enquiries. By some distance the most common use case.

Which makes the next figure more uncomfortable. A Gartner survey of 5,728 customers found that 73 per cent use self-service at some point in their service journey, yet only 14 per cent fully resolve their issue there. Even for issues customers themselves describe as very simple, it is only 36 per cent.

So roughly 59 percentage points sit between usage and resolution. And here is the part that explains the gap: 43 per cent could not find relevant content, and 45 per cent said the company did not understand their issue.

Neither is a technology problem. The first means the entry is missing. The second means the entry exists but the question was not recognised. No model on earth repairs an entry that does not exist.

On top of that, you usually get one attempt. In a further Gartner survey of 3,566 B2B and B2C customers, only 27 per cent said they would try a chatbot again after a poor experience. The counter-check from the same survey is interesting: 49 per cent would have used a chatbot had one been available – only 7 per cent actually did at their last service contact. The willingness is there. It just does not last long.

If both reasons sit in the knowledge base, it is clear where to start. Here is what one source feeding the help page, chat, email and ticketing at the same time looks like:

What is changing inside the teams

This part is missing from most overviews, even though it is where most movement is happening right now.

A Gartner survey of 321 customer service and support leaders paints a more nuanced picture than the headlines: 85 per cent are expanding the responsibilities of their agents, and 75 per cent are shifting people into entirely new roles within the service organisation. 80 per cent report pressure to change their workforce structure, 63 per cent are reducing headcount gradually through attrition, and 31 per cent have implemented or are planning layoffs.

What does that mean day to day? The bottleneck does not disappear, it moves. Automate routine enquiries and you will not need fewer people afterwards – you will need different work for them. Planned, not improvised.

In Germany there is a second building site. According to Bitkom, 66 per cent of companies rate their employees’ AI skills as low. On the positive side, 70 per cent now provide training. And 42 per cent of AI-using companies name their own employees’ lack of acceptance as an obstacle – a point that almost never appears in project plans and derails projects anyway.

The costs sit somewhere else

Anyone planning automation usually budgets for licence costs. The Bitkom figures say something different.

AI users name infrastructure (51 per cent), data preparation (50 per cent) and integration into existing systems (41 per cent) as major cost items. Subscriptions and licences for the software are named by only 21 per cent, and token consumption barely registers at 8 per cent.

Bitkom President Ralf Wintergerst puts it in one sentence:

Wer nur eine Lizenz kauft, hat KI eingekauft, aber noch nichts verändert.

Translated: buying a licence means you have bought AI, not that you have changed anything. For customer service that means the effort sits in preparing the content and connecting the systems. It is the unglamorous work that makes no presentation – and it decides whether anything works in the end.

Since 2 August 2026, Article 50 of the EU AI Act has also applied: people must be able to tell that they are interacting with an AI system, unless that is already obvious from the circumstances.

Where AI helps and where it does not

Now the part you might not expect from an AI vendor. AI does not solve all ten challenges. It clearly helps with four, barely with three, and not at all with one.

Challenge	Effect of AI
Waiting times	High – standard enquiries are answered instantly, at night too
Repetitive routine work	High – recurring questions leave the queue
Channel breaks	High, provided every channel answers from the same source
Inconsistent quality	High – the same answer regardless of shift and experience
Self-service gap	Medium – only as good as the content behind it
Repeat contacts	Medium – falls when answers become more complete
Rising expectations	Medium – speed yes, understanding only partly
Understaffing on complex cases	Low – those are exactly the cases that stay with people
Fragmented systems	Low – AI does not connect systems that are not connected
Lack of measurement	None – measuring remains an organisational job

One finding belongs here even though it is inconvenient: 87 per cent of customers consider access to a human agent essential once a company uses generative AI in service.

Another figure from the same Gartner release points the same way: in a separate survey of 5,801 customers in the US, 54 per cent said they trust human agents more than AI for product or service recommendations, against 32 per cent who trust AI more.

None of that argues against automation. It gives you two design rules. First: the route to the service team stays visible. Second: the knowledge base gets built before the bot, not after it.

And because a table like that deserves a raised eyebrow when it comes from an AI vendor, the honest move is to try it rather than take our word for it.

Five numbers that give it all away

You do not need a new dashboard. Five numbers are enough, and all of them can be built from data you already hold.

Metric	What it answers
First contact resolution	Are enquiries settled conclusively at the first contact?
Repeat contact rate	How much work arises because the first answer was not enough?
Average handling time	How much team time does a case consume?
Contact volume per channel	Is effort moving rather than disappearing?
Zero-result searches	Which content is missing from the knowledge base?

That last row is the most actionable instruction there is: every search term returning nothing is a missing entry. No further analysis required.

One warning to close the table: on automated channels, deflection and CSAT belong in the same report. Falling ticket counts alongside falling satisfaction is not a saving, it is displacement into channels nobody measures.

The first 90 days

Day 1 to 30: measure and sort

Collect the five metrics and pull the twenty most frequent enquiries from the ticketing system. Nothing more. By the end of this month you know which of the ten challenges you actually have – and which you only thought you had.

Day 31 to 60: one source instead of many

Bring existing lists, canned responses and notes together, and name a person who approves entries. In parallel, rewrite the five most common answers so that they include the next step. That costs an afternoon and lowers the repeat contact rate measurably.

Day 61 to 90: automate one channel

The help page is usually the right place to start, because that is where search behaviour becomes visible. Then work through the zero-result searches and close the gaps before the next channel follows.

After three months you have a before-and-after comparison solid enough to carry an investment decision. One built on your own numbers, not on numbers from a vendor deck.

The terms that matter

These metrics come up in every conversation about customer service – and get mixed up surprisingly often.

• First contact resolution (FCR): share of enquiries settled conclusively at the first contact.
• Repeat contact rate: share of customers who come back about the same issue.
• Ticket deflection rate: share of enquiries resolved before a ticket is created.
• Containment rate: share of conversations completed entirely in the automated channel, without handover.
• Cost per contact: average cost of a service interaction, covering staff, licence and variable costs.
• Channel break: a move between channels where the context does not travel with the customer.
• Intent recognition: the method an AI uses to determine the purpose behind a free-text message rather than matching keywords.

Conclusion

Customer service challenges look similar across industries. What differs is their weighting – and you do not read that off a list, you read it off your own numbers.

The order matters more than the pace. Automate before the knowledge sits in one place and you automate the gaps. The Gartner data says exactly that: the main reasons self-service fails are missing content and misunderstood enquiries. Both are addressed by a maintained knowledge base, not by a change of model.

And one point holds regardless of the technology: the route to a human has to stay visible. Without it, satisfaction falls even while the dashboard looks better at first.

How that knowledge then actually reaches customers – through the help page, chat, email or phone – is a topic of its own. Our article on customer self-service is a good place to start.

Frequently asked questions (FAQ)

What are the biggest challenges in customer service?

How do you work out which challenge you have?

Which challenge should you tackle first?

Does AI solve customer service challenges?

Why do customers use self-service and still not resolve their issue?

Where do the costs of customer service automation actually sit?

How do you deal with understaffing in customer service?

What does the EU AI Act mean for customer service?

Sources

• Gartner: Only 14% of Customer Service Issues Are Fully Resolved in Self-Service, survey of 5,728 customers
• Gartner: Only 27% of Customers Would Try a Chatbot Again After a Negative Experience, survey of 3,566 B2B and B2C customers, February and March 2026
• Gartner: 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent
• Gartner: 85% of Service and Support Leaders are Expanding Human Agent Responsibilities, survey of 321 service and support leaders
• Bitkom: Erstmals nutzt die Mehrheit der Unternehmen KI, press release of 14 September 2026, representative survey of 603 companies with 20 or more employees by Bitkom Research
• Matthew Dixon, Nick Toman, Rick DeLisi: The Effortless Experience. Conquering the New Battleground for Customer Loyalty, 2013