Contact center metrics: the 15 that matter (and 5 small teams should track first)

Contact center metrics and call center metrics mean close to the same thing, and most teams use the terms interchangeably. The harder question is which ones matter, and that depends on team size, channel mix, and how much of the business runs through support. A five-person team answering ecommerce questions over email and chat needs a different starting point than a 200-agent voice operation running a workforce management platform.

This page covers 15 contact center metrics worth knowing: five customer experience metrics that describe how people felt, and ten operational metrics that describe how the team performed. It also covers the five a small team should track first, before adding a WFM tool or a dashboard nobody has time to read. Ecommerce examples run throughout, since a lot of what drives contact volume, from “where is my order” questions to return windows, is specific to how retail support works.

Quick reference: the 15 metrics

Formulas and benchmarks below are sourced wherever a credible source exists. A commonly repeated “industry average” with no real methodology behind it gets flagged as such, right alongside the number.

Customer experience metrics

These five ask a version of the same underlying question: did the customer have a good experience? Each one asks it a different way, at a different moment, and pairing more than one gives a fuller picture than any single score can.

Net Promoter Score (NPS)

NPS asks one question, “how likely are you to recommend [company] to a friend or colleague,” on a 0-to-10 scale. Subtract the percentage of detractors (0–6) from the percentage of promoters (9–10) and the result is a single number from −100 to +100. Passives (7–8) aren't counted directly, but they drag down the promoter share.

Bain's own framework treats any score above 0 as good, above 20 as favorable, above 50 as excellent, and above 80 as world-class. The NPS glossary page has the full industry breakdown.

NPS tracks the overall relationship, and a quarterly or post-renewal cadence captures that better than surveying after every contact.

Customer satisfaction score (CSAT)

CSAT asks how satisfied a customer was with one specific interaction, usually on a 1-to-5 or 1-to-10 scale, and reports the share who gave a top-two-box score as a percentage: (satisfied responses ÷ total responses) × 100.

There's no single good CSAT. Retently's 2026 benchmark study puts ecommerce and retail at 77, close to the US national ACSI average, which has held around 77 for most of the last decade. More detail is on the CSAT glossary page.

For ecommerce specifically, CSAT works best tied to delivery. Sending the survey right after an order arrives catches a fulfillment problem that a clean support interaction can hide, since the conversation itself may have gone fine while the package didn't.

Customer effort score (CES)

CES asks how much effort a task took, on a 1-to-7 scale, and is reported as the average score, sum of scores ÷ number of responses, or the share of top-two-box “easy” responses.

Gartner's research is why CES gets taken seriously: 94% of customers with low-effort interactions intend to repurchase, against 4% of those with high-effort ones, and a low-effort interaction costs 37% less to deliver. A score of 5.5 or higher, or a 70%+ top-box rate, is generally considered healthy. See the CES glossary page for the full picture.

For a returns flow or a self-service order-tracking page, CES is often the more useful signal than CSAT, since it measures exactly the friction ecommerce brands are trying to design out.

Interaction quality score (IQS)

IQS is less standardized than the other four. Most teams build it as a weighted composite scored against an internal rubric: did the agent, or the AI, follow policy, address the actual question, and communicate clearly? Because every team's rubric differs, there's no industry benchmark to cite, and comparing your IQS against someone else's is comparing two different tests.

Where IQS earns its place is catching what CSAT and FCR miss on their own: a technically resolved, technically satisfying interaction that skipped a compliance step or gave a customer information they haven't discovered is wrong yet.

First contact resolution (FCR)

FCR measures the share of issues resolved on the first contact, with no follow-up needed: (resolved on first contact ÷ total contacts) × 100.

SQM Group's benchmark puts the industry range at 70 to 79%, with 80% or higher considered world-class and reached consistently by only about 5% of contact centers. The FCR glossary page covers the full methodology.

FCR is the metric most tightly connected to cost. SQM's research shows roughly a 1-for-1 relationship between FCR and both CSAT and operating cost, so a 1-point FCR improvement tends to move both. For ecommerce, the biggest FCR drag is usually a return or exchange that needs a second contact because the first agent couldn't see the order status or the return window.

Operational metrics

These ten describe how the operation performed. The five above described how the customer felt about it. Read on their own, several of them can be improved in ways that make the team look better while making the experience worse, which is why each one below comes with what it hides as well as what it measures.

Average handle time (AHT)

AHT is the average time an agent spends on one contact: talk time plus hold time plus after-call work, divided by contacts handled. Full formula and worked example on the AHT glossary page.

Published figures range from Call Centre Helper's 6 minutes 3 seconds to SQM Group's roughly 10 minutes, a 65% gap between two credible sources, so the better comparison is your own AHT by contact reason over time.

For ecommerce, WISMO (“where is my order”) contacts should have one of the shortest AHTs on the board, since the answer is a lookup, not a conversation. If WISMO AHT is climbing, that's usually a tooling problem, not an agent problem.

Average speed of answer (ASA)

ASA is the average wait before a customer reaches a person, measured from queue entry to pickup: total wait time for answered contacts ÷ contacts answered. See the ASA glossary page for the full breakdown.

ContactBabel measured 116 seconds across UK contact centers in 2024 and ACXPA's mystery-shopping research measured 109 seconds in Australia in 2025, both well past the 20-to-30-second targets most teams plan around.

Always read ASA next to abandonment rate. A queue that lets its worst waits abandon looks faster than it is. For ecommerce brands, ASA is also the metric that spikes hardest around a flash sale or a shipping-delay announcement, which is exactly when a customer already annoyed about an order is least willing to wait.

After-call work (ACW)

ACW is the wrap-up time an agent spends after a contact ends: notes, disposition coding, and record updates, before they're free for the next one. More on the ACW glossary page.

There's no credible industry benchmark. Published figures range from under a minute to 20 to 40 minutes depending on the source, with no named study behind most of them. Segment ACW by disposition code and watch your own trend; a number from a vendor blog won't tell you much.

This is the component AI tools move most directly. A generated call summary an agent reviews and corrects is usually faster than writing one from scratch.

Abandonment rate

Abandonment rate is the share of contacts that hang up before an agent answers: (abandoned contacts ÷ total contacts offered) × 100. The abandonment rate glossary page has the full formula and grace-period notes.

ACXPA's 2024 mystery-shopping research put the national voice-channel average at 8%, with practitioners commonly citing under 5% as a good working target. Exclude abandons under a short grace period, commonly 10 seconds; someone who misdials and hangs up in two seconds isn't telling you anything about your queue.

For ecommerce brands, abandonment climbs predictably during BFCM and other peak-volume windows. Planning extra coverage, or offering a callback, for that specific week does more than a general staffing increase spread across the quarter.

Service level

Service level is the share of contacts answered within a set time threshold, written as a pair like 80/20, meaning 80% of contacts answered within 20 seconds: (contacts answered within threshold ÷ total contacts offered) × 100. Full detail on the service level glossary page.

80/20 is the industry default, though operations researchers Koole and Mandelbaum noted in a 2002 paper that nobody has ever justified the specific numbers, a position that hasn't moved since.

Service level and ASA answer different questions from the same data. ASA can look fine while the slowest 20% of customers wait far longer than the average suggests, and service level is what catches that.

Agent utilization

Agent utilization is the share of a team member's total paid time spent on customer work, counting breaks, training, and meetings in the denominator: productive time ÷ total paid or scheduled time × 100. See the agent utilization glossary page for why the formula itself is contested.

The commonly cited 70-to-85% range has no identifiable source and looks like an occupancy figure that migrated onto the wrong label, and with no single agreed formula for utilization, a published benchmark means little without knowing how it was calculated.

This metric belongs in capacity planning, never in an individual's performance review. A week of required training will tank it for reasons that have nothing to do with how someone is doing their job.

Occupancy

Occupancy is the share of logged-in, available time an agent spends handling contacts, excluding breaks and training entirely: handle time ÷ (handle time + available time) × 100. The occupancy glossary page covers the healthy range in more depth.

Most centers target 75 to 85%, though Call Centre Helper's widely cited 83.3% is a planning input rather than a measured result, and ICMI puts the burnout threshold at 88 to 92% sustained over several consecutive half-hours.

Schedule adherence

Schedule adherence measures how closely a team member followed their assigned schedule: start times, break lengths, and shift end all count. Formula: (scheduled time minus time out of adherence) ÷ scheduled time × 100. Full detail on the schedule adherence glossary page.

There's no universally recognized benchmark. Call Centre Helper offers 85 to 95% as generally acceptable, and ICMI's 2007 survey of 211 centers found roughly two-thirds setting targets at 90% or above. This metric only applies once a team is big enough to run a formal schedule; a five-person team working a shared inbox doesn't need it yet.

Cost per contact

Cost per contact is the fully loaded cost of handling one contact: total support costs divided by contacts handled. The cost per contact glossary page breaks down what should and shouldn't go into the numerator.

Gartner's February 2024 research put the median at $13.50 for assisted channels and $1.84 for self-service. ContactBabel, surveying 189 US contact centers, put the cost of an inbound voice call at $6.91 in 2023. Both are directional; what counts as a “cost” and what counts as a “contact” varies enough between organizations that comparing your figure against a published average tells you less than comparing it against your own history.

The trap: cost per contact can fall while total cost rises, if cheaper, faster contacts are generating repeat business a slower, more thorough one wouldn't have. Read it next to FCR, never alone. For a retail brand deciding whether to add phone support alongside chat and email, cost per contact by channel is the number that answers the question, not the blended average across all three.

Contacts per order

Contacts per order is an ecommerce-specific metric: total support contacts divided by total orders placed in the same period. The contacts per order glossary page has the full formula and root-cause breakdown.

There's no universal benchmark; the direction that matters is down, toward 1 and ideally lower. A brand handling 1,000 contacts against 250 orders has a ratio of 4, which is high, and usually traces back to unclear order tracking, sizing mismatches, or a returns process that requires a phone call instead of a form.

This is one of the clearest volume-side signals available to a retail or ecommerce team, since it flags fulfillment and product-information problems before they show up anywhere else on the dashboard.

Which metrics matter for a small team

A two-to-ten-person support team doesn't need all 15 metrics on a dashboard, and building a WFM-grade measurement program before there's a WFM-grade team to run it usually means nobody looks at any of it. Five are worth tracking from day one.

• CSAT. Cheapest signal to set up, and the fastest way to catch a bad pattern before it shows up in reviews.

• First contact resolution. For a small team, most contacts get resolved the first time by default, since there's nobody to hand off to. Once that stops being true, it usually means volume has outgrown what one round of triage can handle.

• Abandonment rate. The simplest queue-health check available without a WFM platform. If it's climbing, either volume grew or coverage didn't, and there's no third explanation worth chasing first.

• Cost per contact. Even a rough version, support salaries and tools divided by monthly contact volume, is enough to know whether the next dollar is better spent on headcount or on automation.

• Contacts per order. For ecommerce-adjacent SMBs specifically, this is often more useful than any of the four above, since it points straight at the product or fulfillment problem generating the volume, going a step past simply counting it.

The ones that can wait: service level, agent utilization, occupancy, and schedule adherence all assume a formal schedule and enough headcount to staff against a forecast. AHT and ASA are worth watching loosely, but setting a formal target on either before there's a queue big enough to need one solves a problem that doesn't exist yet. NPS is worth adding once the customer base is large enough for a quarterly number to mean something; on 40 responses a quarter, a small swing is noise, not a trend.

How the metrics connect

CX metrics and operational metrics pull in opposite directions if only one side gets watched. Push AHT down hard enough and FCR falls with it, since a rushed contact creates the callback that shows up as a second contact next week. Push service level up by adding headcount and cost per contact rises to match.

Gladly's position on this trade-off is efficiency and devotion to the customer, together. The operational metrics say whether the team can afford to keep doing this. The CX metrics say whether it's still worth doing. Reading one column without the other produces decisions that look good on one dashboard and cost the company on the other.

For ecommerce specifically, the connecting metric is contacts per order. When it's falling because fulfillment improved, cost per contact should be falling too, and CSAT should be flat or rising alongside it. When contacts per order falls because self-service replaced a real conversation and CSAT drops with it, the ratio improved for the wrong reason, and the dashboard alone won't tell you which one happened. Only reading both sides will.