- AHT (average handling time)
- The average time an agent spends on a ticket, including the call or writing, hold time and after-work such as notes. Used mainly to plan staffing. As a target it is risky: agents pushed on AHT tend to close tickets too quickly, which shows up as a lower resolution rate and more repeat contacts.
- AI Act
- The EU regulation on artificial intelligence, Regulation (EU) 2024/1689. It classifies AI systems by risk and sets requirements accordingly, from outright bans to documentation and oversight duties. For customer service the transparency rules matter most: people talking to a chatbot must be told it is an AI. The rules start applying in stages from 2025.
- AI agent
- An AI that finds out for itself what it needs to solve a task, for example by searching a knowledge base and fetching details from other systems, and then answers or acts. In customer service it usually means the AI answers the customer's question directly from the knowledge base and the customer's details, and hands over to a person when the material is not enough.
- AI draft (draft answer)
- An answer the AI produces from the ticket and the knowledge base, for an agent to review, edit and approve before it is sent. The difference from an automatic answer is that a person stands behind every answer that goes out.
- AI form
- A contact form that shows relevant answers from the knowledge base while the customer writes their question, before the ticket is sent. The customer decides whether the answer was enough or whether the ticket should go on to an agent. The form is a route to help, not a barrier in front of the team.
- Automatic answer
- An answer sent to the customer without an agent reviewing it. Should only be used in limited flows where the material is sufficient and the controls are decided in advance, for example for questions with an unambiguous answer in the knowledge base.
- Backlog
- The open tickets waiting to be handled at a given moment. A growing backlog means more tickets are arriving than the team can resolve. The number says most together with age: a hundred tickets from today is something else than a hundred tickets that have waited a week. See also ticket volume and SLA.
- CES (Customer Effort Score)
- A measure of how easy the customer found it to get help, usually a question after the ticket answered on a scale, for example from one to seven. It rests on the idea that low effort goes together with loyalty. Useful for finding awkward flows, such as when the customer has to repeat themselves or switch channel.
- Channel
- A way for the customer to get help: help centre, form, chat, email, phone, customer portal. Channels should be alternatives, not steps every customer must pass through, and they should give the same answer whichever the customer chooses.
- Chatbot
- A program that answers questions in a chat. Older chatbots follow predefined flows and buttons, or match the question against a list of answers. Newer ones are built on language models and can phrase answers freely, ideally grounded in the company's knowledge base. A chatbot should always be able to hand over to a person. See also AI agent and handover.
- CSAT (customer satisfaction)
- Customer Satisfaction Score. The share of customers who answer positively when asked how satisfied they were with the help, usually measured right after a resolved ticket on a scale from one to five. Measures the experience of a single ticket, not loyalty over time.
- Customer portal
- A signed-in page where the customer sees their own tickets, their status and the conversation with customer service, and can reply in the ticket. Because the customer is identified, personal details can be shown here, unlike in the public help centre.
- Customer service software
- An umbrella term for the software a support team works in: ticket handling, channels such as email, chat and phone, a knowledge base, reports and increasingly AI support. Used roughly interchangeably with helpdesk. When choosing a system it makes sense to start from ticket volume, the channels you need and which other systems it has to connect to.
- Data processing agreement (DPA)
- The agreement the GDPR requires when a supplier processes personal data on a customer's behalf. The customer is the data controller and the supplier the processor. The agreement governs, among other things, what may be done with the data, where it is stored and for how long.
- Deflection
- The share of questions answered in self-service, for example in the help centre or an AI form, that therefore never become a ticket with an agent. A measure of how much self-service relieves the team, to be read together with customer satisfaction so that customers are not simply kept from contact.
- Erlang C
- A formula from teletraffic theory used to calculate how many agents are needed to reach a given service level. The inputs are contacts per time unit, average handling time and the target waiting time. The formula assumes no customer gives up while waiting, so it often overestimates the need slightly. Best suited to phone and chat.
- Escalation
- A ticket being passed from one level to another: from self-service to an agent, or from first line to a specialist. A good escalation brings the context along, so the customer does not have to start over.
- Escalation matrix
- An agreed table of where a ticket goes next, when and how. It states, for example, which types of fault go to second line, which tickets go to a manager and within what time. The matrix makes escalation predictable for both agents and customers and reduces the risk of tickets getting stuck between teams. See also escalation.
- Fine-tuning
- Training an existing language model further on your own examples so that it changes its behaviour, for example tone or format. Fine-tuning is rarely the right way to give a model facts about a company: the knowledge becomes hard to update and cannot be traced to a source. For that, RAG with a reviewed knowledge base is more common.
- First response time
- The time from a ticket arriving until the customer gets a first reply from a person or a system. Often measured as a median, since a few slow tickets would otherwise pull up the average. Differs from resolution time, which measures until the ticket is resolved.
- GDPR
- The EU General Data Protection Regulation, which governs how personal data may be processed. Customer service handles a lot of personal data, so the regulation affects how long tickets are kept, who may see them and which suppliers may process them. Customers have the right to access and in some cases erase their data. See also data processing agreement.
- Grounded answers (grounding)
- An AI's answers being based on stated sources, for example the company's knowledge base, rather than on what the model has learned in general. A grounded answer can be traced to an article, and the model can say it does not know when the sources lack the answer.
- Hallucination
- When a language model writes something that sounds credible but is not true, for example a returns rule that does not exist. In customer service it is countered by letting the model answer only from a reviewed knowledge base and by having an agent review answers before they are sent.
- Handover
- When an AI or chatbot passes a conversation to an agent, or when a ticket moves from one person to another. A good handover brings the whole conversation and what is already known, so the customer does not have to repeat themselves. It should be clear to the customer when a person takes over and roughly how long it will take.
- Help centre
- A public website with searchable help articles, often organised in categories, where the customer finds the answer themselves. The help centre shows the public part of the knowledge base and can be indexed by search engines and read by AI assistants.
- Helpdesk
- A system or function for receiving and resolving questions and problems, used for both internal IT support and external customer support. As a system, a helpdesk is in practice the same as a ticketing system: tickets with owner, status, priority and history, often complemented by a knowledge base and reports.
- Human-in-the-loop
- A way of working where a person reviews and approves what the AI proposes before it takes effect, for example before an answer is sent to a customer. Used to build trust in the AI support and to catch errors before they reach the customer.
- Intent
- What the customer wants to achieve with their question, for example returning an item or resetting a password. Recognising intent is the basis for routing, categorisation and automated flows. Older chatbots require every intent to be defined and trained in advance. Language models interpret intent more flexibly but still need clear rules for what happens next.
- Internal note
- A comment on a ticket that only the team sees, not the customer. Used to pass on context, ask a colleague or record a decision. Good internal notes mean the next person on the ticket does not have to ask the customer the same questions again, something customers often experience as needless hassle.
- Knowledge base
- The company's reviewed answers to customers' questions, written once and used in every channel. Each article has an owner, a version and a decision about who may see it: everyone, only the team or only signed-in customers. The difference from an FAQ is the structure around the content, not the number of questions.
- Knowledge gap
- A question that keeps recurring without an approved answer in the knowledge base, or with an answer that does not work. Found in searches with no hits, in the chatbot's unanswered questions and in tickets that keep arriving even though an article exists.
- Language model (LLM)
- An AI model trained on large amounts of text to understand and write text, a large language model or LLM. Language models are what power AI chats, suggested answers and summaries. A language model knows nothing about a particular company on its own, so in customer service it has to be connected to a knowledge base, for example with RAG.
- Live chat
- A chat where the customer talks to an agent in real time. Often combined with an AI chat that answers first on what the knowledge base covers and hands over to the agent when needed, with the whole conversation kept.
- Macro (canned response)
- A saved template that fills in an answer, and sometimes also sets status, labels or owner, in one click. Saves time on recurring questions and makes answers more consistent. The risk is that macros live a life of their own beside the knowledge base and go out of date, so they need an owner just like help articles.
- NPS (Net Promoter Score)
- The customer is asked how likely they are to recommend the company, on a scale from zero to ten. The share answering nine or ten minus the share answering zero to six gives a value between minus 100 and 100. NPS measures the relationship with the company as a whole and is affected by much more than customer service. See also CSAT.
- Occupancy
- The share of logged-in working time that agents actually spend handling tickets, compared with time when they are available but waiting for the next contact. High occupancy looks efficient, but close to a hundred per cent over long periods usually brings longer queues and more tired agents, because there is no margin for peaks.
- Omnichannel
- A way of working where the customer's contacts in every channel, such as email, chat, phone and social media, are gathered in the same history and handled in the same system. The customer can switch channel without starting over, and the team sees the whole picture. Differs from multichannel, where the channels exist but are handled separately.
- Opening hours (for chat)
- The times when an agent is available in live chat. Outside opening hours the chat should show the contact options, for example sending the question as a ticket, instead of a queue nobody is working.
- Priority
- How urgent a ticket is compared with others, often in levels such as low, normal, high and urgent. Priority sets the order in the inbox and often which SLA applies. It should be set by clear rules, for example outages or contract level, so that it is not whoever sounds angriest who comes first.
- Prompt
- The instruction or text sent to a language model to get an answer. In customer service systems the prompt usually consists of fixed instructions on tone and rules, excerpts from the knowledge base and the customer's question. How the prompt is written affects the answer a great deal, but it does not replace correct material: a good prompt cannot fix a wrong article.
- RAG (retrieval-augmented generation)
- A technique where the language model first retrieves relevant passages from a knowledge source and then formulates its answer from them. It is how an AI chat can answer from the company's own articles instead of its general training material, and show the source.
- Repeat contact
- The customer getting in touch again about the same matter, after the ticket was considered resolved or in another channel. A high share of repeat contacts suggests that answers do not solve the problem, that information was unclear or that tickets are closed too early. The measure is the mirror image of resolution rate and often only shows when tickets are linked per customer.
- Resolution rate (first contact resolution)
- The share of tickets resolved in the first contact, without the customer having to come back. One of the measures that best reflects the customer's experience, and one that reveals when answers go out quickly but do not solve the problem.
- Resolution time
- The time from a ticket arriving until it is resolved, unlike first response time which only measures until the first reply. Tickets waiting for the customer should be paused in the measurement, otherwise you measure the customer's response time and not your own.
- Response time
- An umbrella term for how long the customer waits for an answer. First response time measures the time to the first reply, next response time the time from the customer's latest message to the team's reply, and resolution time the time until the ticket is resolved. It is easy to cut first response time with a quick but empty reply, so the measures should be read together.
- Routing
- Automatically sending a ticket to the right group or agent based on rules, for example channel, language, category, the customer's contract or who has capacity. Good routing reduces the time a ticket sits without an owner and the number of times it is passed on. It can rely on simple rules or on AI interpreting the ticket's content. See also triage.
- Self-service
- The customer resolving their ticket themselves, for example in the help centre, in an AI form or in the customer portal, without an agent stepping in. Only works when the answer exists, is correct and is easy to find.
- Sentiment
- The feeling expressed in a message, often roughly divided into positive, neutral and negative. Sentiment analysis is done with language models or other text models and can be used to prioritise upset customers or follow the mood over time. It is an aid, not a verdict: irony, short replies and jargon are often misread.
- Service level
- In phone and chat, the share of contacts answered within a set time, for example 80 per cent within 20 seconds. Used to plan staffing, often together with Erlang C. Should not be confused with an SLA, which is a commitment to the customer, even though the terms are often used interchangeably.
- Shared inbox
- An inbox several agents work in together, for example for a common support address. Unlike an ordinary mailbox, it shows who owns each message and what status it has, so that nothing is answered twice or forgotten. Often the first step from plain email towards a ticketing system.
- SLA (service level agreement)
- A set level for, for example, response time or resolution time, often per channel or priority. The inbox shows how much time remains per ticket, and the follow-up shows the share of tickets kept within the level.
- Suggested answer
- See AI draft. An answer proposed by the AI and reviewed by an agent before it is sent.
- Ticket
- A customer's question or problem, registered with owner, status, labels and history so that it can be followed from start to resolution whatever the channel. A new email about the same matter should land in the existing ticket, not become a new one.
- Ticket categorisation (ticket type)
- Labelling every ticket with what it is about, for example invoice, delivery or login. Categories make it possible to see which questions take the most time, where the knowledge base has gaps and what can be automated. A manageable number of categories used consistently gives better data than a long list everyone interprets differently.
- Ticket handling
- The way of working and the system for receiving, assigning, answering and following up tickets. A shared inbox with owner, status and SLA is the simplest form; a helpdesk or customer service system is the same thing with more features.
- Ticket volume
- The number of tickets arriving in a period, often per week or month and per channel. The basis for staffing, for pricing customer service systems and for seeing which question types account for the largest share of the work.
- Ticketing system
- Software for receiving, assigning and following up tickets from several channels. Every question becomes a ticket with owner, status, priority and history, and the system keeps track of SLAs and statistics. Also called a helpdesk or ticket system. See also ticket handling, which describes the way of working.
- Triage
- The first assessment of an incoming ticket: what it is about, how urgent it is and who should take it. The term comes from healthcare. In customer service triage is done by an agent, by rules or by AI, and leads to category, priority and routing. The earlier it happens, the less time urgent tickets spend in the wrong queue.
- Vector search (embeddings)
- A search method where texts are converted into series of numbers, called embeddings, that capture meaning rather than exact words. The search then finds articles about the same thing even when the customer uses different words from the article. Vector search is a common retrieval step in RAG, often combined with ordinary keyword search.
- Zero data retention
- An operating mode with an AI provider where the content sent to generate an answer is not kept by the provider afterwards and is not used to train models. The only storage is then in your own system, with its retention periods.
Rickard Collander