The acronyms fly fast in customer support — WISMO, FCR, CSAT, deflection, autonomous resolution. Here is what they actually mean, in plain English, with no fluff. Every term is defined the way we use it at BearScope, and many link to a deeper guide.
A handful of patterns make up most of the inbox. Name them, and you can fix them.
The customer already placed an order and now wants to know where it is. WISMO is the single largest ticket category in most retail and ecommerce queues, and the vast majority of it is a tracking lookup the customer could not finish on their own. It is high-volume, repetitive, and the perfect work for an AI agent that can read live order and carrier status — with a person stepping in the moment something is genuinely wrong.
Related: Stop WISMO tickets
The post-purchase sibling of WISMO: the customer returned something, or is owed money back, and wants to know the status. It is high-volume and high-anxiety — "where's my refund" follow-ups pile up fast and erode trust if the timeline is unclear. A good answer checks the real return and refund status and explains the timeline in plain language, and only a person authorizes the money moving.
Related: Automate return & refund status
Resolving a customer's question before it ever lands as a ticket a person has to handle. Done well, deflection means the customer actually got their answer — from a help article, a self-serve flow, or an AI agent. Done badly, it just hides the contact button and frustrates people into giving up, which is not resolution and shows up later as worse satisfaction.
The share of incoming contacts resolved without a person touching them. The honest version counts only the contacts where the customer was actually helped — not the ones who rage-quit the chatbot or opened a second ticket an hour later. Pair it with reopen rate and CSAT so a high number means real resolution, not avoidance.
Related: Measure deflection rate honestly
A chatbot answers. An AI agent reads, reasons, and acts — within limits. Here is the spectrum.
Software that can read a whole conversation, reason over your real data, and take an action on the customer's behalf — not just return a canned answer. The difference from a chatbot is action: an AI agent can look up an order, draft and send a reply, or start a refund, each one checked before it runs. On BearScope, AI agents work the same board as your people and are scored on the same rubric.
Related: AI agents for support, explained
AI that helps a human rep instead of replacing them. While a person handles the case, the copilot drafts the reply grounded in the thread and your data, surfaces the customer's history, and suggests the next step — and the rep edits and sends. Faster replies, fewer false promises, and the person stays in control. Note the language: the human is your rep or teammate, never the "agent."
Related: A copilot that drafts replies
When an AI agent resolves the customer's issue end to end, with no person in the loop. Real autonomy needs three things: the AI must be grounded in your data, every action must be checked against guardrails before it runs, and the whole thing must leave a receipt you can audit. Without those, "autonomous" just means "unsupervised," which is how customers get hurt.
Related: What autonomous resolution requires
Passing a conversation from an AI agent to a person. Nothing burns a customer faster than re-explaining everything after a bot gives up, so a good handoff carries the full context across: the thread, what the AI already tried, the customer's history, and why it escalated. Your rep picks up mid-sentence instead of starting over.
Related: Hand off with context
Quality is measurable. These are the terms behind scoring conversations and acting on what you find.
Scoring customer conversations against a rubric so you can measure quality and coach from it. A rubric breaks "good" into categories — did the rep follow process, was the tone right, was the issue actually resolved — and each conversation gets graded on each one. The point is not the number; it is the specific miss you can fix, on a rep or in an AI agent's guardrails.
Related: How BearScope scores conversations
Traditional QA samples one or two percent of conversations by hand and argues about the rest. 100% scoring grades every conversation — people and AI agents alike — and pulls the ones that fell short to the top. You stop guessing what the other 98% looked like, and you catch the bad pattern while it is still small instead of in next quarter's CSAT.
Related: 100% scoring vs. sampling
Getting your reviewers — and the AI scorer — to grade the same conversation the same way. Without calibration, a score depends on who happened to review it, and reps stop trusting it. Regular calibration sessions align everyone on the rubric, tighten the edge cases, and keep the AI's scoring honest against human judgment over time.
Related: Running QA calibration sessions
The share of tickets marked resolved that get reopened later. It is one of the most honest quality signals you have: a low first-contact-resolution number with a high reopen rate means agents are closing tickets that were never actually fixed. Watch it alongside FCR and CSAT to see whether "resolved" really means resolved.
Related: Reopen rate as a quality metric
Satisfaction, effort, speed, and cost. Each one answers a different question — and they are easy to confuse.
A short survey sent right after a contact, asking how satisfied the customer was — usually "rate your experience 1 to 5." CSAT is transactional: it measures how that one interaction felt, while it is fresh. It is the most common support metric, but it is biased toward people who bother to respond, so read it alongside the conversation scores, not instead of them.
Related: CSAT vs. NPS vs. CES
A relationship metric that asks "how likely are you to recommend us?" on a 0–10 scale. Promoters (9–10) minus detractors (0–6) gives a score from -100 to +100. NPS measures loyalty to the whole brand, not a single support contact — so it moves slowly and is a poor way to grade an individual conversation. Useful at the company level, blunt at the ticket level.
Related: CSAT vs. NPS vs. CES
How hard the customer had to work to get their problem solved, usually phrased as "the company made it easy to handle my issue — agree or disagree." Effort is the metric that best predicts loyalty: customers rarely reward you for going above and beyond, but they punish you for making things hard. Lowering effort — fewer repeats, fewer handoffs, no re-explaining — is often the highest-leverage thing a support team can do.
Related: Reduce customer effort
The share of issues solved in a single interaction, with no callback, no follow-up email, no second ticket. FCR is one of the strongest drivers of satisfaction — customers hate repeating themselves — and a good proxy for whether your team has the data and authority to actually fix things on the first try. Measure it honestly against reopen rate so a closed ticket is not counted as resolved.
Related: First contact resolution, explained
The average time to handle one contact end to end — talk or chat time plus after-contact wrap-up. AHT is an efficiency metric, not a quality one, and chasing it alone backfires: rushing customers off the line tanks resolution and satisfaction. The right move is to cut the time the work actually takes — better context, fewer lookups, AI drafting the reply — not to rush the customer.
Related: The tradeoffs of average handle time
A promise to respond to, or resolve, a contact within a set time — for example, "first reply within one hour" or "resolution within 24 hours." SLAs set the customer's expectation and the team's target, and are often tiered by channel or plan. Set them where you can actually hit them; a missed SLA is worse than an honest, slightly longer one.
Related: Setting achievable SLAs
The fully-loaded cost of resolving one conversation: agent time, tooling, overhead, and now AI usage, divided across the conversations you handle. It is the metric that turns support quality into a business case — when an AI agent resolves the routine surge, cost to serve drops without quality slipping. Track it per conversation and per channel so you can see where the money actually goes.
Related: Cost to serve per conversation
This is the part most "AI support" skips. These four terms are why you can let an AI act at all.
A complete, auditable record of an AI action. Every time an AI agent does something — sends a reply, looks up an order, starts a refund — BearScope writes a receipt: what it did, why, on whose authority, what data it used, and what changed. A receipt is the difference between trusting AI and hoping it behaved. You can read any one of them, after the fact, and know exactly what happened.
Related: Reading an AI receipt
The rules that define what an AI agent may and may not do, enforced before any action runs — not as a polite suggestion in a prompt. Guardrails set the limits: which actions are allowed, what needs a person to confirm, what is always off-limits, and what triggers an escalation. Good guardrails are specific and checked deterministically, so the same input always gets the same safe decision.
Related: Guardrails checklist
Tying an AI's answer to your real data instead of letting it guess from training. A grounded AI agent states facts it can point to — this order shipped on this date, this policy says this — rather than inventing a plausible-sounding answer (a "hallucination"). Grounding is the practical defense against an AI confidently telling a customer something that is simply not true.
Related: Grounding AI support in real data
A safety default: if anything is uncertain — missing data, an ambiguous request, an action outside the guardrails — the AI agent stops and hands off to a person rather than acting anyway. The opposite, "fail-open," is what produces double refunds and wrong promises. Fail-closed means the worst case is a human gets involved, not that a customer gets hurt.
Related: Why AI support should fail closed
Running customer support as one system where people and their AI agents work, are scored, and improve together — rather than a help desk over here, a QA spreadsheet over there, and a bolt-on bot somewhere else. CX operations treats support like a real operation: one board for every conversation, every one scored, every AI action checked and recorded. It is the category BearScope is built for.
Related: What is CX operations?
A glossary tells you what the words mean. The rest of the site shows you what they do. The product page walks one conversation top to bottom — sensed, reasoned, acted — and use cases map each term to a real job your team does every day.
When you are ready to compare, pricing shows what it costs to run a real shift, and security covers how every AI action stays safe and auditable.
A few that pair naturally with the terms above:
A 30-minute walkthrough on a real shift of your conversations, white-labeled to your brand. You will see WISMO and returns resolved, every conversation scored on your rubric, and a receipt for each action an AI agent takes.