The live chat support metrics worth tracking tell you what to fix in the store: contact rate per 100 orders, pre-purchase share, chat-assisted orders, true resolution rate, and first response time as a floor, with the set growing as volume does. Agent speed scorecards matter far less than dashboards suggest.
A chat that ends fast is not the same as a chat that ended well. The most useful thing live chat produces is not a response time. It is a weekly list of the questions your store failed to answer before the customer had to ask.
Open the reporting tab of almost any chat app and you will find a dozen numbers waiting: first response time, average handle time, chats per agent, concurrent chats, missed chats, wait time, CSAT, transfers, tags. A Shopify brand with one support person and a part-time founder reading the inbox at night has a different problem. It needs to know why customers are asking, and whether the answers are turning into orders.
That gap is where most chat reporting goes wrong. The dashboard grades the people answering, while the real information sits in what customers keep asking. If 30 people a week ask whether a candle ships in recyclable packaging, the fix is a line on the product page, not a faster typist.
This guide is for merchants from the first few hundred orders a month up to roughly $1M a month. If you are just starting, focus on the transcript habit and one or two numbers. If you are scaling with a team or an AI agent, the resolution checks later in the piece are where I would spend your attention first.
Most live chat dashboards measure the support team’s speed and workload because that is what the software was first built to manage, and those numbers rarely tell a small Shopify brand what to change. Response time, handle time, and chats per agent are staffing metrics. They answer “are we keeping up,” which matters, but they say nothing about why the conversations exist in the first place.
A store’s chat volume is mostly a mirror of the store itself. Questions about sizing come from a size chart that does not answer the real question. Questions about delivery dates come from a shipping policy written in business days when customers think in calendar dates. “Where is my order” messages spike when tracking emails are slow or vague. None of that shows up as a staffing problem, so a dashboard built around agents will report a busy week as a performance issue when it is really a content issue.
Anthony Morgan, founder and CEO of the CRO agency Enavi, made this point when he joined me on episode 440 of the podcast. He was talking about how to find the cause of cart abandonment before testing anything, and support conversations were his second research source after reviews: “The second thing can be like customer support ticket logs. What are the questions that they’re asking? Because those are friction points. Can we better address those on the site, on the product page, maybe even in the cart?”
That framing changes what a chat metric is for. Instead of asking whether your team answered quickly, you ask what the conversation reveals about the page the customer was on. The rest of this piece builds the short set of measures that serve the second question, and the weekly routine that turns them into fixes.
Five numbers cover what a Shopify brand needs from live chat each week: contact rate per 100 orders, true resolution rate, chat-assisted orders, the share of chats that happen before purchase, and first response time. Together they tell you whether the store is generating avoidable questions, whether those questions get answered for good, and whether chat is helping people buy.
Contact rate is the anchor because it controls for growth. Chat volume rising 40% in a month you doubled orders is good news; chat volume rising 40% in a flat month means something broke. Divide by orders and the number stops lying to you. One caution: a paid campaign or influencer spike sends pre-purchase chats from people who have not bought yet, so check sessions in the same week before reading a jump as a broken page.
The chat-assisted orders figure depends on how your tool attributes sales, and the definitions vary. Shopify Inbox, for example, reports “assisted orders” for its AI agent as orders placed by customers who interacted with the agent, alongside assisted sessions, a satisfaction rate, and average response time, according to Shopify’s Inbox conversations documentation. That counts any order from a customer who chatted, whether or not the chat changed anything, so treat it as a direction signal. A customer who was going to buy anyway still counts.
Notice what is missing. Average handle time, chats per agent, and concurrent chats are staffing numbers. If you have one person answering, you already know how busy they are. Add them back only when you are scheduling shifts across several people. CSAT stays as a sanity check you glance at, not a goal, because in my experience the customers who rate chats skew toward the delighted and the furious.
If you need a monthly version of these figures for a partner, an investor, or your own records, this walkthrough of what belongs in a customer service report covers the wider set to include.
First response time is fast enough once a customer who opens chat during your stated hours reliably gets a real reply within a couple of minutes, and gains past that point are worth far less than getting the answer right the first time.
The research supports pairing speed with accuracy rather than ranking speed first. Zendesk’s 2026 CX Trends research, fielded in June 2025 across 22 countries with 6,182 consumers and 5,115 business respondents, found that “74% are frustrated when they have to repeat information,” and that “86% of consumers say responsiveness and accurate resolution highly influence their purchase decisions,” as summarized in Zendesk’s November 2025 announcement of the report. Responsiveness and accuracy travel together in that finding. A two second reply that asks for the order number the customer already gave is fast and still a failure.
Two practical rules keep this number honest. First, measure the median, not the average. One chat that sat for three hours overnight will drag an average into nonsense, while the median tells you what a typical customer experienced. Second, only count hours when the widget says someone is available. If your chat shows as online at midnight and nobody answers until 9am, that is not a slow response time; it is a broken promise, and the fix is changing the widget’s hours or switching it to an email capture form after hours.
Where speed really does matter is pre-purchase chat. A shopper asking whether a jacket runs small is deciding now, in a browser tab that will close. Post-purchase questions like return windows or address changes can tolerate a slower reply. If your tool lets you route by page or by whether the customer has an order, put pre-purchase chats at the front of the queue.
Chat transcripts tell you why customers hesitate, in their own words, at the exact page where they hesitated, which is something session analytics and conversion reports can only hint at. A heatmap shows that people stall on the size chart. A transcript shows them typing “I’m between a medium and large, which should I get for a loose fit?”
Once a week, read 20 to 30 transcripts, weighted toward pre-purchase chats, and sort each question into one of four buckets: product information missing from the page, policy confusion (shipping, returns, subscriptions), something broken, or a request only a person can handle. The first three buckets are fixes. Only the fourth is real support work.
The “something broken” bucket deserves special attention because customers are often the first to find site errors. Dan Wardle, VP of Sales at Noibu, the checkout error monitoring company, described on episode 322 how the team spotted the problem while talking to DTC CEOs: “the way they’re finding errors is their daughters, their nieces, their cousins.” For a lot of brands, the support inbox plays that role. A run of chats that say “the discount code isn’t working” or “I can’t select a shipping option” is a bug report that arrived through the wrong door, and it should go to whoever owns the theme or checkout the same day, not into a weekly summary.
Most chat tools now tag conversations automatically, which helps with sorting. Shopify Inbox, for example, automatically labels a conversation with a topic when it is confident of the match, and shows up to two topic labels per conversation. Use those tags to count, but read the words yourself before acting. A tag that says “shipping” cannot tell you whether customers want faster delivery, cheaper delivery, or just a date they can trust, and those three call for different fixes.
When the question is about how something is worded rather than what is missing, the reply itself may be the problem. This piece on when positive scripting helps and when it reads as spin walks through rewriting the replies your team sends most.
When an AI agent answers your chats, its resolution rate is a vendor-defined figure, often tied to billing, so check exactly what counts as resolved before you treat it as a measure of customers helped. The definitions are written down, and they are more generous than most merchants assume.
Gorgias, for example, counts an interaction as automated when the customer “does not require the involvement of a human agent within 72 hours of the interaction,” per its billing documentation. Intercom’s Fin counts a resolution when the customer confirms the answer was satisfactory or “exits the conversation without requesting further assistance,” which Intercom calls an assumed resolution, and prices a resolution at $0.99 in its help article on Fin outcomes. A customer who gives up and closes the tab can fit both definitions.
That does not make these tools bad. It means the vendor’s resolution rate measures something narrower than “the customer got what they needed,” and you need your own check alongside it. Three are worth running. Sample 20 AI-resolved chats a week and read them. Track how many AI-resolved customers contact you again about the same issue within seven days through any channel, including email. And compare chat-assisted orders for AI-handled pre-purchase chats against human-handled ones over the same period, keeping in mind that routing sends each side different questions, so compare like topics where your tool lets you.
On episode 463, where I walked through where AI voice agents are already working for Shopify brands, my advice for scaling merchants was to run a 90 day pilot and “literally measure the cost per resolve ticket, you know, and your CSAT score” before deciding whether to scale it or kill it. The same discipline applies to chat. Cost per resolved conversation and satisfaction, measured with your own resolution check rather than the vendor’s, are the two numbers that decide whether an AI agent is earning its place.
Run on one composite store, the five numbers point to three page fixes and an email rewrite rather than a second support hire. What follows is a labelled composite built from patterns I have seen across merchant conversations, not a real brand, and every figure is illustrative. Assume a home fragrance brand doing about $80K a month on Shopify, around 1,300 orders a month, with one part-time support person and Shopify Inbox on the site.
In a typical week the store gets about 300 orders and 96 chats, a contact rate of 32 per 100 orders. The founder’s instinct is to hire a second support person because the inbox feels busy. First response time is a median of four minutes during staffed hours, which is fine. Here is what I would look at, in order.
First, the pre-purchase share. Sixty of the 96 chats come from people without an order, so most of the volume is happening before the sale, not after. Second, the transcript pass. Reading 30 of those pre-purchase chats, three questions account for almost half: whether the candles are safe around pets, how long a candle burns, and whether a gift note can be added. All three are product page and checkout content problems. Third, the post-purchase side: 20 of the remaining 36 chats are “where is my order,” clustered two to four days after purchase, which points at a shipping confirmation email that only says “shipped” without a delivery estimate.
My call is not to hire. I would add a short ingredients and pets line and a burn time to every product page, turn on a gift note field at checkout, and rewrite the shipping confirmation email to carry an estimated delivery window. Then I would watch contact rate for three weeks. In this illustration, if those fixes removed even half of the repeat questions, contact rate would fall toward the low twenties per 100 orders, and the part-time hours would cover the volume with room left for the chats that need a person.
You could disagree with that call. If the pre-purchase chats were converting at a high rate, you might argue the conversations themselves are selling and should be encouraged, not removed. That is why chat-assisted orders sits in the weekly five: it tells you whether a question is friction to remove or a sales conversation to keep.
Drop any chat metric you cannot connect to a decision you would make this month, and let the set grow only as your volume and team do. A store answering 15 chats a week needs a reading habit, not a scorecard. A store with five agents and an AI layer needs the full five plus staffing numbers.
If you are under roughly $30K a month, skip the dashboard. Read every chat, keep a running note of repeated questions, and fix the page each time a question shows up three times. On episode 463 I suggested that newer merchants focus on “improving your inbox and the speed at which you’re replying to customers” before adding an AI voice agent, and the same logic applies to chat automation. A free chat tool and an honest set of business hours are enough.
Between about $30K and $500K a month, add contact rate per 100 orders, the weekly transcript pass, pre-purchase share, chat-assisted orders, and a first response time floor. This is the stage where content fixes pay off most, because volume is high enough to show patterns and low enough that one person can still read a meaningful sample. If you are choosing or replacing a tool at this point, this comparison of live chat apps for Shopify by store type is a useful starting shortlist.
Above $500K a month, with several people or an AI agent, add true resolution rate, cost per resolved conversation, and the staffing metrics you need for scheduling. This is also where I would cut hardest. Teams at this stage often track concurrent chats, transfers, tags per chat, and agent leaderboards, and most of those numbers drive behavior nobody wanted, like agents closing chats early to protect a handle time target. If a metric would change how an agent behaves without changing what the store fixes, take it off the screen.
Whatever your stage, the test is the same one I would apply to any report: when this number moves, do you know what you would do next? If not, it is decoration.
The most important live chat metrics for a Shopify store are contact rate per 100 orders, true resolution rate, chat-assisted orders, the share of chats that happen before purchase, and first response time. Contact rate shows whether the store is creating avoidable questions as it grows. True resolution rate checks whether issues stay solved for at least seven days. Chat-assisted orders and pre-purchase share show whether chat is helping people buy. First response time is a floor to hold rather than a number to keep pushing down. Agent workload metrics like handle time only matter once you schedule several people.
A good first response time for live chat is a median of a couple of minutes during the hours your widget shows as available. Use the median rather than the average, because a single chat left overnight distorts an average badly. Slower than that, replies cost you sales, especially on pre-purchase questions where the shopper is deciding in real time. Past that threshold, shaving more seconds matters much less than giving an accurate answer the first time, without asking the customer to repeat information. If nobody can answer at night, change the widget hours or switch to an email form after hours.
Calculate contact rate by dividing the number of support conversations in a period by the number of orders in the same period, then multiplying by 100. For example, 96 chats against 300 orders in a week gives a contact rate of 32 per 100 orders. Use the same channels each time, chat alone or chat plus email, so the trend stays comparable. Contact rate matters because it adjusts for growth: rising chat volume during a sales spike can be healthy, while rising contact rate in a flat month usually means a page, policy, or email is generating confusion that a content fix could remove.
You know an AI chatbot is resolving issues by checking its resolved chats yourself rather than relying on the vendor’s resolution rate. Vendors define resolution narrowly: some count a chat as resolved when the customer leaves without asking for more help, or when no human gets involved within a set window. Each week, read a sample of about 20 AI-resolved chats, track how many of those customers contact you again about the same issue within seven days through any channel, and compare orders from AI-handled and human-handled pre-purchase chats. Those three checks show whether the agent is helping or just closing conversations.
Each week, read 20 to 30 live chat transcripts, weighted toward pre-purchase chats, and sort every question into four buckets: missing product information, policy confusion, something broken, and requests only a person can handle. The first three buckets are fixes for your product pages, policies, emails, or checkout, so assign each one an owner and a date. Anything in the broken bucket, like a discount code or shipping option failing, should go to whoever manages the theme or checkout the same day. Only the fourth bucket is true support work, and it should shrink as the fixes land.
