Self-service: why your help centre goes unused

Many companies have a help centre, and almost as many wonder why customers still email about what it already says. The answer is rarely that customers are lazy. They try, fail and then contact you, a little more irritated than if they had emailed straight away. This article covers why self-service goes unused, where the help centre needs to be to get used, how search should work, which questions do not belong there and how to measure whether it takes tickets.

How many customers try on their own, and how many succeed?

Almost everyone tries, few succeed. According to a study published by Harvard Business Review, 81 per cent of customers attempt to resolve their issue themselves before contacting a person. The will is there. What is missing is the result: according to a survey of 5,728 customers that Gartner presented in 2024, 73 per cent of customers use self-service at some point in their journey, but only 14 per cent of issues are fully resolved there. Even for issues customers themselves describe as very simple, the share is only 36 per cent.

The numbers look better at companies that have invested. Salesforce states that self-service solves an estimated 54 per cent of issues at organisations that use it, and that 61 per cent of customers would rather use self-service for routine matters. The gap between 14 and 54 per cent is not a difference in customers. It is a difference in how the help centre is built, placed and maintained.

Why does self-service fail?

It fails for four reasons, and they have different solutions. In Gartner's survey the most common cause was missing content: in 43 per cent of cases the customer found no article about their own problem, and 45 per cent of those who started in self-service felt the company had not understood what they were trying to do.

The fourth reason is the one most often missed. An article explaining "our returns policy" without saying how many days, where the label is and when the money arrives is read and abandoned. The customer then emails the same question. How to write an article that actually resolves the ticket is covered in the article on writing help articles, and how to find the questions with no article in the article on knowledge gaps.

Where does the help centre need to be to get used?

Where the question arises, not where it is easy to put a link. A footer link is used by people who already know the help centre exists. Everyone else needs to meet the answer at the moment they wonder.

• At the checkout and on the product page. Shipping, delivery time, returns and payment methods are questions that arise before purchase. A line saying "Questions about delivery and returns" linking to the right article catches them before they become chats.
• In the order confirmation and the delivery notification. That is where the customer is when "where is my order?" arises. Link to the article on delivery status, not to the help centre's front page.
• In the contact form, before it is sent. As the customer types their question, the form should show the matching articles. Some questions are resolved there, and those sent anyway arrive with better context.
• In the chat. A chatbot that answers from the help centre is the help centre in conversational form. It is the placement that reaches most people, because the customer has already asked in their own words.
• In replies from customer service. Every email reply to a common question should link to the article. Next time the customer goes there directly.

Supportifier's public help centre is built on the same reviewed knowledge base as the chat, the form and the inbox, so an article written once exists in all those places at the same time. That is how self-service ends up where the question arises without anyone maintaining five copies.

How should search work?

Search should find the article when the customer types in their own words, and it should tell you what it did not find. Those are two requirements, and the second is what makes the help centre better over time.

Customers do not search for your headings. They type "parcel gone", "money back", "change size", "cancel". An article called "Delayed or lost delivery" has to be found on all those words. That is solved with synonyms in the article, the customer's phrasing in the heading and, today, search that understands meaning instead of exact words. AI search across the knowledge base does that, provided it only answers with what the articles say and says so when nothing fits.

The log of searches with no results is the cheapest list of knowledge gaps you can get. Read it every week. The ten most common questions with no results are the next ten articles.

Which questions should not be solved with self-service?

Questions that require a judgement, a decision or a conversation with an upset customer. That includes complaints where something has to be decided, disputes about invoices, cancellations with conditions and GDPR requests. An article can explain how the process works, but the ticket goes to a person, and the way there should be clear from the first page.

Self-service should not be the only route either. Internetstiftelsen's report Svenskarna och internet 2025 shows that nearly a quarter of Swedes have skipped booking a doctor's appointment because the e-service was too complicated, and that pensioners more often struggle with new digital contact channels. The customer who does not find the answer must be able to reach you in one click, and the help centre must work with keyboard and screen reader. A help centre that hides the contact route to "steer towards self-service" gets fewer tickets in the statistics and more dissatisfied customers in reality.

How do you measure self-service?

Measure resolved questions, not visits. A help centre with many visits and just as many tickets resolves nothing; it is an extra step before the email. Four measures are enough.

1. Self-service rate. The share of questions resolved without a ticket. Most simply counted as article views not followed by a ticket from the same customer within a day, or as the share of chats closed without handover. How the measure is defined and what is reasonable is covered in the article on customer service metrics.
2. Tickets after an article view. Which articles are read and still followed by a ticket? Those articles do not answer the question. Rewrite them first.
3. Searches with no results. The count and the most common phrasings. Those are the knowledge gaps.
4. Tickets per hundred orders or customers. The only measure that shows whether self-service reduces volume as the business grows.

Gartner's distinction between "uses self-service" and "resolves the issue there" is the one to keep an eye on. Visits say customers are trying. Only the last measure says it pays off.

What to do

1. Pull the twenty most common questions from the inbox over the past month. For each, check whether there is an article and whether it answers with concrete timings, steps and amounts.
2. Write or rewrite the articles that are missing or vague. Answer in the first sentence, the customer's words in the heading.
3. Place the links where the questions arise: checkout, order confirmation, notifications, contact form and reply templates. The footer is not enough.
4. Let the contact form suggest articles as the customer types, and the chat answer from the same knowledge base.
5. Turn on the log of searches with no results and read it every week.
6. Keep the contact route visible on every page of the help centre, and check that it works with keyboard and screen reader.
7. Measure self-service rate and tickets per hundred orders from now on, and follow them month by month. In Supportifier the help centre, chat and inbox are parts of the same ticketing system, so the measures can be followed in one place and the missing article is written where the ticket landed.

Common questions

What is a good self-service rate?

It depends on what you sell and how you count, so compare with your own trend rather than other people's numbers. Gartner's survey shows that few companies fully resolve more than a fraction of issues in self-service, while Salesforce reports around half at those that have invested. The goal is a share that rises every month while tickets per hundred orders fall.

Should we hide the contact details to get more people to use the help centre?

No. It cuts tickets on paper and raises frustration in reality, and the customer finds a way to reach you anyway, often through a channel you do not staff. Show the contact route on every page. What should steer towards self-service is that the answer exists and is easy to find, not that the alternative is missing.

Is an FAQ page enough?

Not if you have more than a dozen questions. An FAQ page has no search, no owners and no log of what customers did not find, and it grows into a list nobody reads. A help centre has one article per question, search, update dates and measurement. That is the difference between publishing answers and knowing whether they are used.

How often should the help centre be updated?

Whenever something changes, and otherwise whenever the log of searches with no results or the list of tickets after article views points at an article. A weekly review of those two lists is enough for most. Put an update date on every article; customers and search engines trust a dated text more.

Sources

• Kick-Ass Customer Service — Harvard Business Review, 2017
• Gartner Survey Finds Only 14% of Customer Service Issues Are Fully Resolved in Self-Service — Gartner, 2024
• What Is Customer Self-Service? A Complete Guide — Salesforce, read 2026
• Svåra e-tjänster, inblick i svenskarnas AI-promptar och en riktig Google-utmanare — Internetstiftelsen, Svenskarna och internet 2025