Reference
The customer support glossary
The words the job runs on — every metric, method, and model, defined in plain English. Each entry gives you the formula, an honest note on what it hides, and a sourced benchmark where one exists.
Metrics & Measurement
Abandonment Rate
The share of customers who leave the queue before reaching an agent.
Roughly 0-5% is considered good and 5-8% acceptable; sustained rates above 8% signal a problem.AHT (Average Handle Time)
The average time an agent spends actively working a contact, including after-call work.
CES (Customer Effort Score)
A measure of how much work the customer had to do to get their problem solved.
Contact Rate
Support contacts normalised by customers, users, or transactions — demand per unit of business.
Cost to Serve
The total, fully-loaded cost of supporting a specific customer, account, or segment over a period — a superset of cost per contact.
There is no single target — the point of the metric is that cost to serve varies so widely across customers that the most profitable 20% typically generate 150-300% of total profits while the least profitable 10-20% lose 50-200% of them; the Pareto 80/20 rule understates how concentrated it is.CSAT (Customer Satisfaction Score)
The share of surveyed customers who rate a specific interaction as satisfactory.
Roughly 70-90% is considered good; over 90% is excellent, with most industries clustering in a 65-80% band.CSAT Response Rate
The share of surveyed customers who actually complete a CSAT survey — responses received divided by surveys sent.
No universal standard; feedback programmes most commonly land around 15%, with 8-18% typical and single-digit rates (4-8%) still normal, with some exceeding 40%.DSAT (Dissatisfaction Score)
The share of surveyed responses that fall in the bottom box — the dissatisfied mirror image of CSAT.
FCR (First-Contact Resolution)
The share of issues fully resolved in a single interaction, with no follow-up needed.
The cross-industry average sits around 70%; 75%+ is good and top teams reach 80-85%.FRT (First Response Time)
How long a customer waits from opening a ticket to the first human (or meaningful) reply.
Resolution Time
The total elapsed time from a ticket opening to it being resolved.
Ticket Volume
The count of tickets created in a period — the raw demand signal for a support team.
Transfer Rate
The share of contacts an agent hands to another agent, team, or department rather than resolving in place.
Benchmarks vary by industry from roughly 5% to 15%, with most teams aiming to keep it under 10%.QA & Quality
Calibration
The practice of aligning multiple QA reviewers so they score the same interaction consistently.
QA Score / IQS (Internal Quality Score)
A reviewer's rating of a handled interaction against a defined rubric of quality behaviours.
Reopen Rate
The share of resolved tickets that get reopened because the issue wasn't actually fixed.
Root-Cause Analysis (RCA)
A structured method for tracing a recurring class of tickets back to the underlying defect that causes them, so the fix removes future contacts instead of resolving each one by hand.
Tone of Voice
The emotional register and word choice of a support reply — how something is said, layered on top of the literal answer.
People & Careers
Agent Experience (AX)
The quality of a support agent's day-to-day working conditions — the tools, workload, autonomy and support that shape how well they can actually do the job.
Attrition
The rate at which support staff leave and must be replaced over a period.
Rates vary sharply by size — one industry survey reports ~17% for small centers, ~37% for medium, and ~44% for large operations.Coaching
The recurring, usually one-on-one conversation where a lead reviews real interactions with an agent and agrees a specific behaviour to improve.
Coaching has no healthy "rate," but the payoff is real: Gallup finds 80% of employees who received meaningful feedback in the past week are fully engaged.Schedule Adherence
How closely agents follow their scheduled work and break times.
Span of Control
The number of people who report directly to one manager — in support, agents per team lead or supervisor.
Service-desk agent-to-supervisor ratios average about 8.6:1, ranging from roughly 3 to 19 depending on process complexity and how much non-people work supervisors carry.Swarming
A collaborative model where specialists join to solve a case together instead of passing it up a tier ladder.
Time to Proficiency
The elapsed calendar time from a new support agent's start date to the point they consistently meet production targets and handle the large majority of contacts without help — a measure of ramp, not of training length.
Contact centers commonly report 4–8 weeks of formal ramp (classroom plus nesting) before an agent hits proficiency; complex or highly regulated products stretch it to 6–9 months.Support Ops
ACW (After-Call Work)
The wrap-up time an agent spends finishing a contact — notes, tagging, CRM updates — after the customer is gone but before taking the next one.
Averages run roughly 45-90 seconds per contact by industry, and wrap consumes an estimated 6-12% of an agent's shift; most teams target 30-60 seconds and treat consistently over 90 as a process or tooling problem.ASA (Average Speed of Answer)
The average time contacts wait in queue before an agent answers.
Backlog
The set of open, unresolved tickets waiting for action at a given moment.
Callback (Virtual Queue)
A virtual-queue feature that holds a caller's place in line and rings them back when an agent is free, instead of making them wait on hold.
In Nextiva's 2025 Customer Patience Benchmark survey, 75% of U.S. adults said they would rather get a callback than wait on hold.Concurrency (Chat Concurrency)
The number of live conversations a single agent handles at the same time, mostly used for chat and messaging channels.
Live-chat concurrency typically runs 2-6 simultaneous chats per agent; 2-3 is common where quality is protected, with higher numbers reserved for simple, repetitive queries.Cost per Contact
The fully-loaded cost of handling one customer interaction.
Deflection
Preventing a contact from reaching an agent, usually via self-service or automation.
Erlang C
The century-old queueing formula workforce planners use to turn call volume and handle time into the number of agents needed to hit a target service level.
Erlang C is typically sized to a service level of 80% of calls answered within 20 seconds — the long-standing "80/20" industry-standard target, though it was adopted for operational convenience rather than proven optimal.Escalation
Moving a case to a higher tier, specialist, or manager when the current owner can't resolve it.
Forecasting
Predicting future contact volume and workload so staffing matches demand.
Occupancy
The share of an agent's logged-in, available time that is spent actively handling contacts.
Around 75-85% is considered a healthy balance for inbound teams; consistently higher risks burnout.Queue Time (Wait Time)
The time a contact spends waiting in line after it has been routed to a queue but before an agent picks it up — reported on voice as Average Speed of Answer.
The most common target is the 80/20 service level — 80% of contacts answered within 20 seconds. The 20-second threshold came from balancing staffing cost against caller patience in early call-center operations rather than from customer research, so treat it as an operational convention, not a hard rule.Service Level
The percentage of contacts answered within a target time, expressed as a pair like 80/20.
80/20 — answering 80% of contacts within 20 seconds — is the widely cited industry-standard target.Shrinkage
The portion of paid agent time that is unavailable for handling contacts.
Global average shrinkage runs roughly 30-35% for most inbound operations, varying by sector.SLA (Service Level Agreement)
A committed target — often on response or resolution time — that support promises to meet for a set share of tickets.
Ticket
The unit of tracked work — a single logged customer request and everything attached to it.
Ticket Aging
How long currently-open tickets have been sitting unresolved, usually read as a distribution across age bands rather than a single average.
HDI benchmarking treats any open ticket older than 30 days as unacceptable and sets the target for that age band at zero; a healthy end-of-day backlog runs around 5% of daily ticket volume.Tiered Support
A model that routes issues through escalating levels (T1, T2, T3) by complexity or specialisation.
WFM (Workforce Management)
The planning discipline of matching the number of agents on shift to the volume of work arriving, interval by interval.
Schedule adherence targets typically run 85-95%; teams handling longer or more complex interactions aim for 85-90%, since a 100% target is neither achievable nor healthy.Knowledge & Self-Service
KCS (Knowledge-Centered Service)
A methodology where solving a customer issue and capturing the knowledge are the same act.
Knowledge Base (KB)
A structured, searchable repository of help content — articles, FAQs, how-tos — that customers or agents use to resolve issues without a live conversation.
Even though roughly 73% of customers use self-service at some point, only about 14% of customer service issues are fully resolved there — and the top failure is customers not finding content relevant to their issue, not a shortage of articles.Knowledge Gap
The distance between the questions people actually ask and the answers your knowledge base can actually give.
No universal healthy range, but as proxies: keep zero-result search rate under roughly 15% early on and 5-8% at maturity, and cover 90%+ of recurring ticket topics with an article.Knowledge Review Interval
The maximum time an article can go without an accountable owner confirming that its answer, screenshots, links, and scope are still current.
There is no universal healthy interval. Zendesk's verification rules support risk-based frequencies such as 2 weeks, 6 months, or 1 year and illustrate a 3-month review cadence.Self-Service
Channels that let customers resolve issues on their own — help centres, FAQs, portals, communities.
Ticket Deflection
The specific measure of self-service resolving a would-be ticket before it is created.
Tooling & Stack
API / Integration
A documented interface, and the live connections built on it, that let a help desk exchange data and trigger actions across the rest of your software stack.
Help Desk vs Service Desk
A help desk is reactive, incident-focused support (and the ticketing tool behind it); a service desk is the broader ITIL practice that also owns service requests and ties into change, problem and release management.
Helpdesk Seat Price
The published subscription charge for one paid support-user seat, before usage, add-ons, implementation, integration, and internal administration costs.
Public July 2026 list prices span from $19 to $139 per full agent per month across the selected Zendesk and Intercom tiers, with billing terms and included capabilities varying materially.Macro (Canned Response)
A saved, reusable reply template agents insert to answer common questions consistently and fast.
Omnichannel Support
A support model where a customer moves between channels and their conversation history and context follow them, so they never have to start over.
Customer expectation is the real bar: in Zendesk's CX Trends 2026, 81% of consumers want a conversation to continue across channels without backtracking, and 74% are frustrated when they have to repeat information.Support AI List Price
The public price of a support AI product per seat, outcome, automated interaction, or another vendor-defined billing unit.
July 2026 public pricing includes outcome, interaction, and seat models: Intercom Fin lists $0.99 per outcome; Intercom Copilot lists $29 per agent/month annually; Zendesk Copilot lists $50 per agent/month annually; Gorgias lists an AI allocation plus per-interaction overage.TCO (Total Cost of Ownership)
The full multi-year cost of a support tool, counting implementation, integration, admin, and training on top of the subscription line.
For help desk and customer service software, acquisition is only about 30% of five-year TCO; the remaining ~70% is configuration, integration, administration, upgrades, and support.CX & Trust
Churn Rate
The share of customers (or recurring revenue) you lose over a set period — the inverse of retention.
For subscription businesses, monthly churn averages around 3.3% overall — roughly 3.8% for B2B and 6.5% for B2C — with voluntary churn (~2.4%) outweighing involuntary, payment-driven churn (~0.9%).CLV (Customer Lifetime Value)
The total profit a business expects to earn from a single customer across the whole relationship — the number that reframes retention and support as revenue rather than cost.
No universal healthy CLV exists — it is entirely business-specific. The common proxy target is an LTV:CAC ratio above 3:1, with the strongest companies reaching 3–8x.Customer Health Score
A composite index that blends usage, support, sentiment, and engagement signals into one number meant to predict whether an account will renew or churn.
Customer Journey Mapping
A visualization of the steps a customer takes toward a goal — their actions, questions, and emotions at each stage — drawn from the customer's side rather than the company's org chart.
Moments of Truth
Any interaction where a customer forms a lasting impression of a company — the high-stakes touchpoints, like a complaint or an outage, where loyalty is won or lost.
NPS (Net Promoter Score)
A relationship metric based on how likely customers are to recommend you, scored from -100 to +100.
Above 30 is generally considered good and above 70 excellent; cross-industry averages run roughly 26-68 depending on sector.NRR (Net Revenue Retention)
The share of recurring revenue kept from your existing customer base over a period — after expansion, contraction, and churn, with new customers excluded.
Above 100% is the health line; median B2B SaaS NRR ran about 101% in 2025, with best-in-class teams at 120%+. Gross retention, which strips out expansion, sat near 88%.Proactive Support
Reaching customers before they reach you — heading off a predictable problem, warning about a known issue, or answering an anticipated question ahead of the ticket.
Service Recovery
The work of repairing a customer relationship after a service failure — the acknowledgement, fix, and make-good that decides whether they stay or leave.
Roughly 55-70% of customers whose complaint you resolve will do business with you again, climbing toward ~95% when they feel it was handled quickly.VoC (Voice of Customer)
The practice of systematically capturing what customers say about their experience — across surveys, reviews, and the support conversations you already own — and turning it into decisions.
Fewer than a third of consumers give feedback directly to a company after a good or bad experience, and that share fell roughly 7–8 points between 2021 and 2025 — so any survey-based VoC program is sampling a shrinking, self-selected minority.AI in Support
Agent Assist
AI that supports a human agent in real time — suggesting replies, surfacing knowledge, drafting summaries — without acting on its own.
Agentic AI (AI Agents)
AI that doesn't just answer a question but plans and takes multi-step actions in real systems to resolve a request on its own.
Gartner projects agentic AI will autonomously resolve 80% of common customer service issues by 2029, cutting operational costs about 30% — a vendor-quoted target for the future, not a rate any team hits today.AI Agent Task Success
The share of eligible support tasks an AI agent completes correctly end to end, including the answer, policy decision, tool actions, and resulting system state.
Do not use a generic target. In the ICLR 2025 tau-bench study, a leading function-calling model completed fewer than half of realistic retail and airline tasks, while retail pass^8 reliability was below 25%.Autonomous Resolution
AI that resolves a customer's issue end-to-end with no human in the loop.
Chatbot
A program that holds a text or voice conversation with a customer through a chat interface — spanning everything from a scripted decision tree to an LLM-powered agent.
Over 98 million people — roughly 37% of the US population — used a bank chatbot in 2022, a figure projected to exceed 110 million by 2026.Containment
The share of contacts an automated channel handles end-to-end without escalating to a human.
Context Window
The fixed budget of tokens a language model can hold in view at once — the system prompt, conversation history, retrieved knowledge, and drafted reply all have to fit inside it.
Conversational AI
The umbrella term for software that holds a natural, two-way conversation with a customer — in text or voice — using language understanding, dialogue management, and increasingly LLMs, rather than a fixed menu.
Deflection vs. Resolution
The crucial distinction between a contact being avoided and a customer's problem actually being solved.
Fine-Tuning
Continuing to train a pretrained model on a smaller, labeled dataset so it adopts a specific behavior — a house voice, a fixed output format, a task — rather than new facts.
Guardrails
The layered automated checks that keep a support AI on-policy — screening its inputs, validating its outputs, and forcing a human handoff when it strays outside safe bounds.
Hallucination
When an AI model states false information — an invented policy, step, or citation — with the same fluent confidence it uses for correct answers.
On grounded document summarization, the best current models hallucinate roughly 1-2% of the time while the weakest exceed 20%; open-ended and citation-heavy tasks run far higher.Human-in-the-Loop (HITL)
An AI architecture that keeps a person embedded at defined decision points to review, approve, or correct the model's output before it affects the customer.
Intent Detection
The NLU step that maps a customer's message to one of a predefined set of intents, so a bot or router can act on it.
On the HINT3 benchmark of real chatbot queries, four leading commercial intent services averaged roughly 58-72% in-scope accuracy, with the best system near 74% — far below the 90%+ typically quoted on curated test sets.LLM (Large Language Model)
A neural network trained on vast text to predict the next token, which is what powers modern support features like drafted replies, summaries, and autonomous resolution.
On grounded summarization, the best-performing models hallucinate on roughly 1–2% of documents; general-purpose models and harder, open-ended questions push error rates much higher (17–33% even with retrieval in one legal-tool study).Prompt Injection
An attack that smuggles instructions into the text an AI model reads, hijacking it to ignore its real orders and do the attacker's bidding instead.
Not a healthy-range figure but a risk one: in the StakeBench agent benchmark, direct prompt-injection attacks succeeded more than 79% of the time and indirect attacks between roughly 42% and 68%, with current defences only partly effective.RAG (Retrieval-Augmented Generation)
An AI pattern that retrieves relevant documents first, then generates an answer grounded in them.
Sentiment Analysis
Automated classification of the emotional tone of a customer message, usually positive, neutral, or negative.
The Support Brief
One considered issue a month — delivered on the first calendar day.
Monthly. No ads. Unsubscribe anytime. Privacy