Customer Service KPIs for Australian Online Stores

Customer service KPIs show how quickly your online store answers shoppers, whether it solves their problems and how much work each order creates. Start with six measures: first response time, full resolution time, first contact resolution, customer satisfaction, tickets per 100 orders and backlog.

Keep AI-answered and staff-answered conversations separate. When an AI agent takes routine questions, the remaining staff tickets can take longer even while customers get help sooner.

The KPIs That Matter for Online Stores

Check the six core customer service metrics weekly, using the same definitions and reporting window each time. Review them by channel and contact reason before assigning an action.

KPI	Calculation or definition	What to investigate
First response time	Time from first message to first useful reply	Queues and coverage gaps
Full resolution time	Time from first message to final resolution	Cases waiting on staff or a carrier
First contact resolution	Cases solved in one contact without a repeat / eligible cases, multiplied by 100	Missing answers or failed actions
Customer satisfaction score	Positive survey responses / valid responses, multiplied by 100	Poor outcomes and confusing replies
Tickets per 100 orders	Inbound support tickets / orders, multiplied by 100	Recurring friction in shopping or delivery
Backlog	Unresolved cases at a fixed reporting time	Old cases and work arriving faster than it closes

Look monthly at cost per ticket and average handle time by contact reason. Review quality scores and survey response rates monthly too, with weekly attention to serious errors.

For customer service representatives, set targets within their control. Score the accuracy of their answer and the completed action. Compare workloads within the same queue, so a refund specialist isn’t judged against someone answering sizing questions.

First Response and Resolution Time

First response time measures the wait between a shopper’s first message and the first reply that addresses it. A receipt saying that the message arrived is a separate acknowledgement.

Zendesk records first reply time from ticket creation to the first public agent response, with calendar-hour and business-hour versions. Gorgias’ chat report measures the first human reply separately from AI messages. Read the helpdesk definition before comparing its number with another system.

Use median response time to describe the typical wait. Add the 90th percentile, the wait that 90% of measured tickets fall within, to expose slow cases. Show unanswered tickets beside both figures.

The following are suggested starting targets for a staffed store, not industry averages. Confirm that your roster can meet them before promising them publicly.

Channel	Suggested first useful reply target	Coverage rule
Email	Within one business day	State the store’s working hours and time zone
Live chat	Within one minute while marked available	Offer another contact route when staff are unavailable
Social messages	Within one business hour during staffed hours	Separate public comments from private support messages

Service response time is the first wait shoppers experience. Once AI answers first, report time to its useful answer and time to a human reply separately. Also measure the wait from handover request to human reply.

Full resolution time runs from the first message to the final solved or closed event used by your system. A quick greeting leaves a shopper waiting if their parcel problem remains open. Record both the total elapsed wait and the business-hours measure.

For a routine policy or order-status question, aim to resolve it in the first conversation. For a refund you can approve immediately, aim to complete the store’s action within one business day. Payment settlement is a separate wait.

For a delivery delay requiring carrier investigation, set an update target of one business day and give the shopper a case-specific next step. Keep the case visible until resolved. Don’t report an update as a completed delivery investigation.

First Contact Resolution

First contact resolution means the customer’s issue is solved during the first contact and they don’t need to return about it. One reply and a closed ticket alone don’t establish that result.

Choose and document a repeat-contact window of seven days, for example. Match reopened tickets and new contacts using the customer, order and reason. Exclude duplicates and spam under a written rule.

In a hypothetical week, 100 eligible cases have completed that observation window. Seventy are solved in the first contact, but five of those customers return about the same issue. First contact resolution is 65%.

Report resolution without a person alongside handover rate for AI-answered chats. Count verified resolved chats with no staff involvement, divided by eligible AI-started chats. For handover rate, divide chats transferred to staff by the same eligible total.

Keep abandoned, unresolved and unclassifiable chats as separate outcomes. An unanswered chat isn’t an automated resolution. Use transcripts and action records to confirm that an answer or action met the customer’s request.

If AI resolves simple delivery-policy questions first, staff inherit more exceptions. Their first contact resolution rate can fall because the queue changes. Compare staff rates within each contact reason and report the combined customer outcome separately.

Average Handle Time and Cost Per Ticket

Average handle time measures active work on a contact, including the reply and after-contact administration. Add the tracked work minutes and divide by handled contacts. Time spent waiting for a carrier belongs in resolution time.

Call-centre average handle time often includes talk time and hold time as well as after-call work. That definition doesn’t transfer directly to asynchronous email or several simultaneous chats. There’s no useful single industry standard across those workflows.

Once AI handles easy questions, staff average handle time can rise because harder cases remain. To reduce staff handling time, pass the transcript and order details into the handover. Check that staff avoid repeating questions and still complete the required action.

Cost per ticket equals allocated support cost divided by handled tickets. Include staff hours at the fully loaded hourly cost, software fees and other support costs for the same period. Use one tax basis consistently and document the allocation.

Split AI-only and staff-handled costs without counting a transferred chat twice. Assign its AI usage to the staff-handled group if you want mutually exclusive groups. Allocate shared software costs using the same rule each period.

Compare cost per verified resolution alongside cost per ticket. Cheap abandoned contacts lower the ticket average without solving a shopper’s problem.

Volume and Workload KPIs

Tickets per 100 orders puts support volume beside store activity. For a hypothetical 2,000 orders and 300 inbound tickets in one period, the ratio is 15.

Some tickets concern orders from an earlier period, so this is a workload ratio, not the percentage of buyers who contacted you. Define whether your count includes pre-purchase questions and keep that choice consistent.

Tag contact reasons, such as order status, returns, sizing and delivery delays. A rising delivery share suggests a different action from rising pre-purchase questions. Read the guide to reducing avoidable support tickets when repeat questions drive the workload.

Tickets per agent per day equals handled tickets divided by staffed agent-days. Read it beside complexity and quality. Count backlog at the same point each week, then split it by age and the party responsible for the next action.

Quality KPIs

Customer satisfaction score (CSAT) records how customers rate the interaction. Reopen rate exposes cases marked solved before the problem stayed solved. Quality review scores check whether the conversation followed your standards.

Calculate reopen rate as resolved cases reopened for the same issue within your chosen window, divided by eligible resolved cases. Wait until each case has completed that window before comparing periods.

Read all three measures together. A polite answer can receive a good rating while misquoting a returns policy. A correct refusal can receive a poor rating even when the staff member followed the policy.

CSAT and Customer Effort Score on Agent-Answered Chats

CSAT measures satisfaction with a specific chat through one rating question. Send the survey after the customer receives the answer or the requested action finishes, including chats resolved by AI.

A short customer satisfaction feedback form can ask:

1. How satisfied are you with the help you received? Choose 1 for very dissatisfied through 5 for very satisfied.
2. What would have made this chat more helpful? This free-text answer is optional.

Count ratings of 4 or 5 as positive. If 40 of 50 valid ratings are positive, CSAT is 80%. Show the 50 responses and the survey response rate beside the percentage.

Low response rates make the result less representative. Dissatisfied shoppers can be more motivated to reply, while silent customers remain unmeasured. Send the same short survey consistently, make it easy to answer on mobile and avoid surveying only successful chats.

Customer effort score (CES) measures how easy it was to get help. For a quick order-status chat, ask, How easy was it to get the answer you needed? Use a scale from 1, very difficult, to 7, very easy.

For that scale, a higher average means less reported effort. Add all ratings and divide by valid responses. Keep the scale direction fixed, and show the count plus the share of low ratings.

CSAT asks how satisfied the shopper felt. CES can reveal repeated questions or a difficult handover even when the answer is correct. Qualtrics describes CES as a single-item measure of effort in a customer interaction.

Link each survey response privately to its conversation. Put low ratings beside the quality assurance scorecard so a reviewer can open the matching transcript in the agent’s console. Use both AI-only and handed-over samples.

A QA Scorecard for Conversations

A quality assurance (QA) scorecard gives reviewers the same criteria for each conversation. Use this five-part example and score each criterion from 0 to 2.

Criterion	0 points	1 point	2 points
Accuracy	Wrong product, order or factual answer	Incomplete answer needing correction	Correct answer supported by the available records
Policy compliance	Action or advice breaches store policy	Policy correct but required check missed	Correct policy and required checks applied
Resolution	Request left unresolved without a next step	Useful progress with a clear next step	Answer or action completes the request
Tone and clarity	Confusing or dismissive reply	Understandable with avoidable jargon	Clear and respectful language
Handover	Needed person unavailable or context lost	Transfer made with missing context	Correct escalation with usable context, or no escalation needed

Add the points for a score out of 10. Treat a serious accuracy or policy failure as a failed conversation regardless of its total. Calibrate reviewers by scoring the same transcripts and discussing differences.

For software that monitors live chat quality, look for transcript sampling and scoring against your own scorecard. It must let reviewers open the source conversation and compare trends by agent or contact reason. Permissions, exports and links between ratings and transcripts make follow-up practical.

Sample random conversations as well as low-rated ones. Reviewing only complaints hides how often the wider queue meets your standard.

Building a Weekly Support Dashboard

Build a one-page customer service dashboard with six tiles and a contact-reason breakdown below them. Keep filters for channel and whether AI or staff resolved the case.

The following layout uses hypothetical data for one reporting week. It is a worksheet you can reproduce in your reporting tool.

Dashboard tile	Sample result	Question it answers
First response time	Median human email reply: 4 business hours	How long does a shopper wait for a person?
Full resolution time	Median: 18 elapsed hours	How long does the issue remain open?
First contact resolution	65% across 100 eligible cases	Does the first contact solve the problem?
CSAT	80%, from 50 responses	How do responding customers rate the help?
Tickets per 100 orders	15, from 300 tickets and 2,000 orders	Is support demand rising relative to sales?
Backlog	40 open cases, including 8 older than two business days	Which cases need attention now?

Use mature cohorts for repeat-contact measures and record the backlog snapshot time. Label each tile’s units and calculation window. Add the previous comparable period so a change is visible.

Turn the dashboard into a short monthly customer service report for the owner. State the largest change, its likely cause and one action with an owner. For example, a rising delivery-delay share prompts the operations lead to improve dispatch updates.

Keep the figures supporting that decision in the report. Include the response count when a satisfaction score changes, and compare the same channel mix before claiming service improved.

Setting Targets Without Gaming Them

Pair each speed target with a measure that confirms the problem was solved. A lone closure target rewards staff for closing cases early, then makes the next contact look like someone else’s work.

Speed or volume target	Paired quality measure	Review trigger
First response time	Accuracy score	Fast replies contain wrong information
Resolution time	Reopen rate	Closed cases return with the same issue
Average handle time	QA resolution score	Staff rush customers through unfinished actions
Tickets handled per day	First contact resolution	Staff split one problem into several tickets
AI resolution rate	CSAT and unresolved-chat share	Chats end without a useful answer

Start with a baseline for each contact reason, then choose one improvement your team can control. Review the slowest cases and a random quality sample together. Change the workflow that caused the wait before tightening its target.