7 Customer Support KPIs Ecommerce Founders Should Actually Track

Most support dashboards are built to make someone feel good. Tickets closed, chats answered, a satisfaction score sitting comfortably above 90%. All up and to the right, none of it telling you whether support is helping or quietly bleeding money.

The customer support KPIs worth your attention do one of two things: they change a decision, or they expose a problem you'd otherwise miss. Everything else is decoration.

Here are seven that earn their place, with the formula for each, a realistic range to aim at, what it's actually good for, and the part nobody writes down, precisely how each one gets gamed.

The seven at a glance

These ranges are practical rules of thumb for judging your own trend, not published research benchmarks. Ticket mix varies enormously between categories.

1. First response time (FRT)

Definition: How long a customer waits for the first meaningful human or AI reply.

Formula: Σ (timestamp of first reply minus timestamp of ticket created) ÷ number of tickets. Use the median, not the mean. One ticket left over a bank holiday weekend will wreck an average.

Good for: Predicting cart abandonment and satisfaction better than almost any other single metric. Pre-sale questions are time-sensitive, and slow replies carry a real cost.

How it gets gamed: Auto-acknowledgements ("Thanks, we've got your message!") counted as a first response. That's a receipt, not a reply. Also watch for teams sending a one-line holding message to stop the SLA clock. Exclude automated acknowledgements from the calculation entirely.

2. Resolution rate (also called automation rate)

Definition: The share of conversations closed out without a human needing to intervene.

Formula: Conversations resolved without escalation ÷ total conversations × 100.

Good for: Capacity planning. It tells you how much of your incoming volume is genuinely repetitive, which is the input to every staffing and automation decision you'll make.

How it gets gamed: Two ways, both common. First, counting deflection as resolution. The customer gave up and left, which the system records as "no escalation". Second, shrinking the denominator by excluding conversations tagged out of scope. Always pair this metric with CSAT and with your reopened-ticket rate.

3. Average resolution time (ART)

Definition: Total elapsed time from a ticket opening to it being genuinely closed.

Formula: Σ (resolved timestamp minus created timestamp) ÷ resolved tickets. Report the median and the 90th percentile together, the 90th percentile is where your angry customers live.

Good for: Spotting broken processes. If returns take four days to resolve and shipping questions take four minutes, your problem is a workflow, not a staffing level.

How it gets gamed: Closing tickets prematurely and letting the customer reopen as a new ticket, which resets the clock and inflates your volume. Merging related tickets to close several at once. Track reopen rate alongside ART or the number means nothing.

4. CSAT (customer satisfaction)

Definition: The share of surveyed customers who rate their support experience positively.

Formula: Positive ratings (4 to 5 on a 5-point scale) ÷ total ratings × 100.

Good for: Acting as the guardrail on every efficiency metric above. Cost per ticket falling while CSAT holds is progress. Cost per ticket falling while CSAT slides is cost-cutting with a delay on the invoice.

How it gets gamed: Selective surveying, only sending the survey after tickets that closed well. Perfect-moment timing, straight after a refund is approved. And the structural problem nobody mentions: response rates are usually low and skewed toward the delighted and the furious, so a small sample swings hard. Report your response rate next to the score, always.

5. Ticket volume per 1,000 orders

Definition: Your contact rate, normalised against business size.

Formula: Total tickets ÷ total orders × 1,000, over the same period.

Good for: This is the most underrated metric on the list. Raw ticket volume rises when you grow, which tells you nothing. Contact rate tells you whether your product, pages and post-purchase experience are getting better or worse. When it falls, something upstream got fixed. When it climbs, something broke. A courier, a size chart, a delivery estimate.

How it gets gamed: Bundling several customer questions into one ticket. Counting a follow-up as part of the original thread. Also note that a growing store's contact rate falls naturally as repeat customers need less help, decompose by new versus returning customers before you claim credit.

Segment it by type. WISMO ("where is my order?") is usually the single largest bucket in ecommerce, and it's the most reducible. Our guide to cutting WISMO tickets covers the fixes.

6. Cost per ticket

Definition: The fully loaded cost of resolving one customer conversation.

Formula: (Fully loaded salaries + tooling + management overhead + training) ÷ tickets resolved. The full breakdown lives in our guide to what a support ticket actually costs.

Good for: Turning support from a felt expense into a modelled one. It's also the input that makes an AI support ROI model credible rather than aspirational.

How it gets gamed: Almost always by omission. Excluding management overhead. Excluding the founder's six hours a week in the inbox because it isn't payroll. Dividing by tickets received rather than tickets resolved, which rewards you for abandoned conversations.

7. Support-influenced revenue

Definition: Revenue from customers who had a support conversation within a defined window before purchasing.

Formula: Revenue from orders placed within X days of a support conversation ÷ total revenue × 100. Set X deliberately, 7 days is defensible for pre-sale, 30 is a stretch.

Good for: Making the case that support is not purely a cost centre. Pre-sale questions, sizing help and product recommendations all convert; this is how you show it.

How it gets gamed: Long attribution windows that sweep up customers who'd have bought anyway, and treating correlation as causation. The honest version needs a holdout or a clean before/after, not just a window.

Vanity metrics to ignore

Build your scorecard this week

1. Pick one metric per category: speed (FRT), quality (CSAT), efficiency (cost per ticket), volume (per 1,000 orders), revenue (support-influenced).
2. Define each formula in writing, including what's excluded. Get agreement before you report a single number.
3. Set a baseline over 90 days, not one week.
4. Report medians and 90th percentiles for anything time-based.
5. Review monthly, as a set. Any single metric moving alone is usually a measurement artefact.

A word on honesty: no metric will tell you when AI gave a technically correct answer to an upset customer who needed a human. Read a sample of transcripts every week, especially the escalated ones and the low CSAT scores. Emotionally charged complaints, unusual edge cases and anything requiring a policy exception are where automation should hand over, and your dashboard won't flag those on its own.

The bottom line

Seven metrics, reviewed monthly as a set, will tell you more than a dashboard with thirty tiles. Speed, quality, efficiency, volume and revenue. One number each, with the formula written down and the exclusions agreed.

Then apply the discipline that makes them useful: every efficiency metric gets paired with a quality guardrail. Cost per ticket with CSAT. Resolution rate with reopen rate. Handle time with satisfaction. Improvement on one side and deterioration on the other isn't a win, it's a trade you made without noticing.

If you run a Shopify store and want the speed and cost numbers to move without the quality ones slipping, [AskZoye](https://askzoye.com) handles the repetitive front line at around 5.7 cents per conversation, replies in about three seconds in 50+ languages, and hands off to your team when a conversation genuinely needs a person.