It is the monthly business review. The support lead presents first reply time at 2.1 hours. Five minutes later, the operations analyst shares a slide from a spreadsheet export. Same metric, same month: 6.4 hours.
The next twenty minutes are not about customers. They are about whose number is right. Someone asks for a reconciliation by Friday. The actual question, whether customers are waiting too long, gets pushed to next month.
This is one of the most common situations in support reporting, and it is rarely anyone’s fault. Both people did their job. They just calculated the same KPI in two different ways.
Why the same KPI produces different numbers
A KPI name is a label, not a definition. “First reply time” sounds precise until you ask a few questions about it.
Does the clock run 24 hours a day, or only during business hours? Does it start when the ticket is created, or when it reaches the right group? Do tickets solved without any public reply count? What about tickets created by an agent on a customer’s behalf, or spam that was later deleted? Is the month based on when tickets were created or when they were solved?
Each answer is a reasonable choice. The trouble starts when two people make different choices without writing them down. A calendar-hours figure and a business-hours figure for the same tickets can easily differ by a factor of two or three, because the business-hours clock stops overnight and at weekends.
There is also drift over time. Someone changes a report filter in March to exclude a test group. Nobody records it. By September, nobody remembers why the trend line has a step in it.
The cost is bigger than one awkward meeting. When people stop trusting the numbers, they stop acting on them. Every review becomes a debate about data instead of a decision about service.
The fix: one KPI card per metric
A KPI card is a short, written definition that lives in one shared place. It answers every question that could make two people calculate the number differently. Once it exists, any report that does not match the card is wrong by definition, and the argument ends.
A good card fits on half a page and has these fields:
| Field | What it answers | Example: first reply time |
|---|---|---|
| Meaning | What does this tell us, in one sentence? | How long customers wait for a human response |
| Formula | How is it calculated? | Median minutes from ticket created to first public agent reply |
| Start and stop | When does the clock start and end? | Starts at ticket creation, stops at first public agent comment |
| Hours basis | Calendar or business hours? | Business hours, using the team’s published schedule |
| Included | Which tickets count? | Email, web form and chat tickets from customers |
| Excluded | Which tickets do not count? | Agent-created tickets, merged tickets, deleted spam |
| Date basis | Which date puts a ticket in a period? | Ticket solved date |
| Source | Where does the official number come from? | One named dashboard |
| Owner | Who answers questions and approves changes? | Support operations lead |
| Change log | What changed, when and why? | Dated list of every edit |
The example column is only an example. Your choices may be different, and that is fine. What matters is that the choice is written down and everyone uses the same one.
How to write your first KPI cards this week
You do not need a project for this. One focused hour is enough for the first few cards.
- List the KPIs that appear in leadership reviews. Usually five to eight: first reply time, full resolution time, CSAT, SLA achievement, backlog, reopen rate, tickets solved. Ignore the rest for now.
- Start with the one that causes the most arguments. If you have ever had two numbers for it in the same meeting, that is your first card.
- Find every place the number is produced. Dashboards, spreadsheet exports, the monthly slide deck. Write down how each one calculates it today. This step often shows the cause of the disagreement within minutes.
- Check the metric settings in your reporting tool. In Zendesk Explore, time metrics such as first reply time and full resolution time come in calendar-hours and business-hours versions, and you also choose how the metric is aggregated, for example average or median. Note which version each report uses. Also check the date attribute. A report filtered on ticket created date and one filtered on ticket solved date will count different tickets for the same month.
- Fill in every field on the card. Where there are two reasonable options, pick one and write one line on why. If you cannot decide alone, agree it with the people who present the number.
- Name one source. Pick the dashboard that matches the card and mark it as the official source. Other reports can exist for exploring, but only the official one goes on review slides.
- Run the two-person test. Give the card to two people who did not write it. Ask each of them to calculate last month’s number on their own. If they get the same answer, the card works. If not, the gap shows you which field is still vague.
- Put the card where the number lives. Link it from the dashboard, or add the one-line meaning and the hours basis in small text under the chart. A definition nobody can find will not stop the next argument.
After the first card, add one per week. Within two months, every number in your review has a written definition and an owner.
Where this can go wrong
Changing a definition breaks your trend. If you move from calendar hours to business hours, last quarter’s numbers are no longer comparable with this quarter’s. When you change a card, either recalculate the history using the new definition or mark the change clearly on the chart. Record it in the change log either way.
The card can become a cover for a convenient number. Excluding a group of tickets is sometimes right and sometimes a way to make the result look better. Every exclusion should have a reason a customer would accept. If you would be uncomfortable explaining it to a customer, keep those tickets in.
Too much detail makes cards unread. A card with three pages of edge cases will be skipped. Keep the main fields short, and put rare edge cases in a short notes line at the bottom.
Owners move on. If the owner leaves and nobody replaces them, the card slowly goes out of date while everyone still trusts it. Review the full set of cards once a quarter, and reassign ownership as part of any role change.
Tool updates can change behaviour. Reporting tools add, rename and adjust metrics over time. When a number moves suddenly with no change in the team or volume, check whether the metric or the report changed before looking for an operational cause.
The honest tradeoff: cards take time to write and they will sometimes make a number look worse, because the agreed definition is stricter than the one someone was using. That short-term discomfort is the price of a number people will actually act on.
The takeaway
Most KPI arguments are not about performance. They are about definitions nobody wrote down. A one-page card for each metric, with one formula, one hours basis, one source and one owner, turns a debate about data back into a discussion about customers.
This week: pick the support KPI that caused the most debate in your last review, find every report that produces it, and write its card. Then give the card to two colleagues and ask them to calculate last month’s number separately. If their answers match, publish the card next to the dashboard.
I also share a short daily lesson on KPIs and processes on Instagram at @vijitlabs. Comment GUIDE on any reel for the free 2-page guide that goes with it.
Not sure what your Zendesk is hiding?
Book a free 30-minute Zendesk review. We look at your SLAs, ticket flow and reporting together, and you leave with at least one thing worth fixing.
Book a free review