Picture the monthly business review. The support lead puts up first response time, CSAT and ticket volume. Finance asks what support did for revenue this month, and nobody in the room can answer from that screen.
Here is how we build a CX report finance will read, with the money questions at the top and the service metrics that explain them underneath.
Start from the P&L and work back to the helpdesk
Most CX reports start from whatever the helpdesk exports. We start from the other end. Support touches the P&L in four places, and each one gets its own metric:
- Pre-sale conversations that end in an order.
- Customers who stay after something went wrong with their order.
- The cost of every answer a person types that software could have sent.
- Refunds, where support hears the reason before anyone else does.
Speed and satisfaction scores still belong on the report, lower down, to explain why a money metric moved.
Does support help people buy?
Pre-sale chat conversion. Of the shoppers who asked something before buying (sizing, stock, delivery dates, materials), how many placed an order within a set window afterwards? Agree that window with finance once and leave it alone. If it changes, month-on-month comparisons stop meaning anything.
Shopping Assistant return. If you run Gorgias Shopping Assistant or another pre-purchase AI, divide the revenue attributed to it by what it costs you, counting the hours someone spends tuning it. Track it every month. We don't publish a target for this one, because it swings with your average order value and with how many of your visitors ask before they buy.
Share of revenue with a CX touch. Revenue from orders where the customer spoke to support first, as a share of total revenue. Read alone, it flatters support, since some of those shoppers would have bought anyway. It earns its place through the trend line and through the cohort comparison in the next section.
Does support keep customers?
This is the slowest feedback loop on the report. For most brands it's also the one with the most money riding on it.
Repeat purchase rate, with and without a support contact. Take the customers who contacted support about an order and compare how often they buy again with customers who never got in touch. If the contacted group comes back less, support is where you lose people. If it comes back as often or more, problems are being recovered instead of passed along.
Repeat purchase by refund history. Group customers by how many refunds they've had (none, one, two or more) and compare how many order again over the following quarter. The useful comparison is between none and one. When a customer whose single refund was handled well keeps buying, your returns flow works. Customers with repeated refunds are the churn risk, and they often point at a sizing or product fault nobody has fixed yet.
What does a resolved question cost?
Volume finally matters here, in the opposite direction to most dashboards. A team answering fewer tickets because the repetitive ones never reach a person is improving.
AI Agent coverage. Tickets the AI Agent resolves end to end, with no human touch, as a share of incoming tickets. On the accounts our team works with, Gorgias AI Agent handles 50-60% of repeat tickets (median, on properly configured accounts). Report it against repeat tickets and against total volume, clearly labelled, because the two get confused in meetings.
AI Agent quality. High coverage with wrong answers is a liability. Each week, have someone review a sample of AI-resolved tickets against a short rubric. Was the answer correct? Did it sound like the brand? Did any action it took (a refund, an address change, a cancellation) stay inside the rules you set? Falling accuracy is usually the first sign that the knowledge base or the rules have drifted. We go deeper on this in our honest take on Gorgias AI Agent.
Help Center deflection. Repeat questions that end on a Help Center article instead of becoming a ticket. A well-built knowledge base deflects 25-40% of repeat questions. When yours sits far below that, the articles usually exist but are written in the brand's words, not the words customers type into search.
Macro use. The share of human-handled tickets that started from a macro. When it's low, agents are writing the same reply from scratch over and over, which costs time on every ticket and makes answers impossible to audit.
Cost per resolution. Fully loaded support cost (people, software, the AI Agent) divided by resolved tickets.
Where is money leaking out?
Refund rate by reason. Refunds split by the reason the customer gives: wrong size, faulty item, late delivery, changed mind. This one says more about the business than about support. A rising wrong-size line means the size guide or the product photos need work, and fixing them can protect more margin than hiring another agent would.
Service metrics: keep them and read them as causes
These are the numbers most dashboards open with. Keep them, below the money questions, and use them to explain what moved above.
First response time. Time to the first reply from an agent or the AI Agent. Report email and chat separately. Customers expect very different speeds on each, and a blended average hides both.
Resolution time by category. A sizing question and a damaged-parcel claim shouldn't share an average. Report the median per category, starting with the categories that bring in the most tickets.
Ticket volume by category. The most useful operational metric here. Tag last quarter's tickets, rank the categories, and the top of that ranking is your automation plan.
First contact resolution. Tickets solved with no follow-up from the customer. We watch it more closely than CSAT, because a customer who has to write twice rarely rates the experience well, however friendly the agent was.
CSAT and customer effort. Keep CSAT, and remember who answers surveys: often the customers who feel strongly in one direction or the other. For transactional contacts (order status, address changes) we prefer an effort question, "how easy was it to sort this out?", because it rates the contact instead of the brand.
Why this guide has no benchmark table
Plenty of KPI guides end with a grid of targets. We left ours out. A response time that's fine for a furniture brand is slow during a sneaker drop, and chat conversion depends on price point and on how many visitors ask before buying. A universal target would be a number we made up.
The two figures we do quote above come from our team's accounts and keep their conditions. For everything else, take your own last quarter as the baseline and set targets against it.
Where each number comes from
| Metric | What it answers | Data source |
|---|---|---|
| First response and resolution time | Is the basic service working? | Gorgias statistics |
| Ticket volume by category | What do we automate next? | Gorgias tags |
| CSAT | How did the contact feel? | Gorgias satisfaction survey |
| AI Agent coverage and quality | Is automation resolving tickets, and correctly? | Gorgias AI Agent statistics plus the weekly review |
| Help Center deflection | Do customers find answers on their own? | Gorgias Help Center statistics |
| Pre-sale chat conversion | Does support help people buy? | Gorgias tickets joined to Shopify orders through Klaviyo or GA4 |
| Share of revenue with a CX touch | How much revenue passed through support? | The same join, totalled by month |
| Repeat purchase by contact or refund history | Does support keep customers? | Cohorts in a BI tool or data warehouse |
| Refund rate by reason | Where is margin leaking? | Shopify returns, with the reason recorded |
The first five rows come straight from Gorgias once tags and surveys are set up. If you can't see them today, the problem is configuration. The two revenue rows need tickets and orders joined in your attribution tool, which is a job for an analyst and only has to be built once. The cohort rows need a BI tool; if you don't have one, decide whether the answer is worth the setup before buying anything.
The metric we leave off
Tickets closed per agent per day. It rewards closing fast, and agents learn quickly what the report rewards. If you need a productivity measure, pair macro use with the weekly quality score. One shows agents aren't rewriting answers that already exist, and the other shows those answers are right.
Send us your current report
Send us a screenshot of the CX dashboard you use today. On a 30-minute call we'll go through which money questions it answers, which it skips and what we'd connect first.
Our team was named Best European Gorgias Partner 2024.
Book a demo. Half an hour on your numbers, without slides.
