How metrics are calculated

What each number means, what it counts and what it leaves out.

Rules every metric shares

Time. Every day and hour is Asia/Yerevan (UTC+4). The current month is always month to date.
People. Georgi, Davit, Temirlan, Elena and Alexandr. Agent KPIs count only what the agent controls.
Human reply. A message by a teammate who is not one of the two bots. A bot message is never a reply.
Which month. An event counts in the month or week it happened, so one conversation can appear in two months.
Median. The lower-middle value when the count is even, as in the October report.
System emails. Missed-call alerts, login codes and HR mail are removed from the KPIs, using the sender list in Settings.
Colours. Red below threshold, amber at threshold, green at target, blue at stretch. Levels live in Settings.
Month by month. Rates compare directly. First replies compare as a per-day average, so a part month is fair against a full one.
Freshness. Data refreshes once a day at 06:00 Yerevan. Ended months are frozen and served as stored.

Chat KPIs

The core team and agent numbers. They are pinned by a test to the October report, so a change to any of them starts with the spec.

First response within 5 min (FRT)

first replies within 300 s of assignment / all first replies

Clock: Starts at the latest assignment to that agent before their first reply. Stops at the agent's first reply in the conversation.
Counted in: The month the reply was written. A reply with no earlier assignment to that agent is left out.
Team figure: All agents' first replies pooled. A conversation answered by two agents counts twice, once per agent.
Levels: Threshold 75%, target 85%, stretch 90%. Bonus KPI.
Shown on: Overview, Agents, Agent page, Bonus

First reply, median

median of (first reply - latest assignment), in minutes

Scope: The same replies as FRT, shown as a time instead of a share.
Shown on: Agents

First replies

count of the replies FRT measures

Use: The sample size behind FRT. The bonus needs at least 30.
Shown on: Overview month by month, Agents

First contact resolution (FCR)

resolved and not reopened / assigned and not escalated

Base: Conversations assigned to the agent in the period, minus those the agent escalated.
Resolved: The agent closed it in the period and the conversation carries the issue_resolved tag.
Not reopened: No real customer message within 24 h of the agent's last close in the period. A message that is only thanks or ok (RU, EN, UA, AM word list and common emoji) does not count, unless a person replied to it within 24 h. A new question does count as a reopen.
Team figure: The agents' counts added together.
Levels: Threshold 28%, target 30%, stretch 32%. Bonus KPI, and only pays when close tagging passes its gate.
Shown on: Overview, Agents, Agent page, Bonus

What happens to assigned conversations

the FCR base, split into six groups

Groups: Resolved. Resolved, reopened within 24 h. Closed as no answer. Closed with no outcome tag. Closed by a colleague. Still open.
Reading it: Only the first group counts for FCR. 'Closed by a colleague' is what remains after the other five.
Shown on: Agents

Close tagging

conversations with an outcome tag in time / conversations the agent closed

Unit: Each conversation the agent closed in the period, judged at their last close in it. Later closes are ignored, so a past month stays the same.
In time: An outcome tag from anyone, applied no later than 5 min after that close.
Outcome tags: 2ndline_relevant, api_transferred, bug_created, issue_resolved, no_answer, rework_created, sales_reassign, transfer_sales, transfer_to_accounting. Editable in Settings.
Gate: FCR pays only above 58%. Exactly 58.0% does not pass.
Shown on: Overview, Agents, Agent page, Bonus

Escalation rate

escalated / assigned

Transfer: In the period, the agent assigned it to someone outside tech support's in-house set.
Technical: The agent applied bug_created, rework_created or api_transferred.
Reading it: Lower is better. Either one makes it escalated, and the Transfer / tech column shows each share. No target is set. Never a bonus input.
Shown on: Overview, Agents, Agent page

Customer waiting time

median of (agent's first reply - earliest customer message since the last human reply)

Includes: Bot and routing time, so this is the wait the customer felt, not only the agent's part.
Counted when: The customer's message came at or before the assignment to that agent.
Reading it: Lower is better. No target is set. Never a bonus input.
Shown on: Overview, Agents

CSAT

ratings of 4-5 / all ratings

Status: Calculated when ratings exist, but the Overview tile shows Not collected, planned from December. Not in the bonus.
Shown on: Overview tile

Attendance

The bonus uses shifts, not hours, because agents switch to Away to stop new chats while they keep working.

Attendance factor

shifts with presence / (scheduled shifts - excused shifts)

Presence: The agent was Active at any point in the shift, or did any work in it: a reply, note, assignment, close or call.
Which shifts: Shifts that already ended, and only up to the end of the downloaded activity log.
Excused: Days in the absences table (sick, annual leave, other) leave the calculation.
No result: With no shifts left to count, the factor is empty and the bonus pays nothing.
Shown on: Agents, Agent page, Bonus, Shifts roster (red outline on a missed shift)

Active hours and activity checks

Active time inside scheduled shifts, compared with work done per 30 min

Purpose: Diagnostics only, never in the factor: active hours, auto-away hours, 'Active but no activity' for 2 h or more, and 'Away but working'.
Shown on: Agent page

Bonus

Option A, per agent and per month. Values below come from Settings, and each change creates a new config version.

Payout

cap x score x attendance factor

Cap: Per agent, with its own currency. Until amounts are set, payouts read as % of cap.
Maximum: 120% of the cap, at stretch on every KPI and full attendance.
Example: FRT at target pays 100%. FCR fails the tagging gate and pays 0%. Score = 50% x 100% + 50% x 0% = 50%. With 90% attendance and a cap of 100, the payout is 45.
Shown on: Bonus

Score

sum of (KPI weight x tier share)

Weights: FRT 50%, FCR 50%.
Tier share: Below threshold 0%. Threshold up to target 80%. Target up to stretch 100%. Stretch and above 120%.
Small sample: A KPI with under 30 in its sample loses its weight, and the weight moves to the other KPI. FRT's sample is first replies, FCR's is its base.
Rounding: Tiers and the gate use the one-decimal value on screen, so a value that rounds up to a level counts as reaching it.
Shown on: Bonus

Close and reopen

a stored snapshot of inputs, config version and result

Rule: A month can be closed only after it ends. A closed month never changes unless it is reopened.
Estimate: Until then, the bonus is a month-to-date estimate that the daily refresh recalculates.
Shown on: Bonus

Option B (simulator only)

pool of caps x agent's share of (score x routed x attendance)

Use: For comparison, never for payment. The simulator changes nothing that is saved.
Shown on: Bonus simulator

Shifts

How the page compares the roster with expected chats. Comfortable 7, limit 10 and minimum cover 2 come from Settings.

Expected chats

average real chats for that weekday, spread over hours by when chats arrived

Basis: The last finished month before the one shown. Customer-started chats, including Instagram and Facebook, but not email.
Daily total: One chat counts once per agent per day when the agent replied, divided by how many of that weekday the month had.
Hour shape: From the hour of the customer's first message, for chats that reached a person. Chats written while support is closed go to 08:00, 09:00 and 10:00 as 50%, 30% and 20%.
Shown on: Shifts roster and day view

Capacity, chats

net hours of everyone on shift x 7

Net hours: Shift hours minus breaks. These are the 'paid hours' on the page.
Speed: Settings can give an agent a speed factor, which multiplies their hours.
Per hour: The day chart uses time on shift in that hour x 7, with a second line at x 10.
Shown on: Shifts roster and day view

Load, %

expected chats / capacity

Colours: Green up to 95%. Amber up to 143% (10 / 7). Red above that, or when chats are expected and nobody is on shift.
Shown on: Shifts roster

Days over the limit

count of red load days in the month

Equivalent to: Expected chats above 10 per paid hour.
Shown on: Shifts tile

On shift / need, hours below requirement

hours in support time where on shift < need

Need: One agent per 7 expected chats, rounded up, and at least 2 before 20:00 and 1 after.
On shift: An agent counts when on shift for at least half of the hour.
Support time: Weekdays 08:00-21:00, weekends 08:00-20:00.
Colours: Red below need, amber at need, green above.
Shown on: Shifts tile and day view

Hours with one agent

support hours before 20:00 with exactly one agent on shift

Note: Hours with nobody on shift are not in this count. They show up under hours below requirement.
Shown on: Shifts tile

Busiest vs quietest

highest chats per paid hour / lowest chats per paid hour

Inputs: The estimated figures from Workload balance. 1x is an even split.
Shown on: Shifts tile

Workload balance

agent's estimated chats / agent's net hours

Chats, estimate: Each hour's expected chats split among the agents on shift, by their time in the hour x speed.
Real: Chats replied in / net hours, on work days before today when the agent replied.
Shown on: Shifts

Workload and calls

Volume figures. Averages per weekday and hour use only the days of the month that have passed.

Incoming chats

customer-started conversations that reached tech support / number of that weekday

Reached: Sitting in a tech queue, or assigned to a tech queue or an in-scope agent at any point. Placed by creation hour.
Shown on: Workload

Conversations handled per day

conversations with at least one reply that day

Agent and team: A conversation counts once per agent per day, and once per day for the team. So the agents can add up to more than the team.
Shown on: Overview, Workload, Agent page

Self-assigned vs routed

assignments to the agent, split by who made them

Self-assigned: The agent assigned it to themselves.
Routed: Anyone else did, including bots and the assigner.
Shown on: Workload

Inbound calls and missed

all inbound calls on the Support RU line; missed = not answered

Answered: Unitalk state ANSWER. Every call counts, so one caller can count several times.
Shown on: Calls

Call backs and connected

outbound calls; connected = answered

By agent: Matched through the agent's Unitalk line.
Shown on: Calls

Talk time

sum of talk seconds per agent

Scope: Answered inbound calls plus the agent's outbound calls.
Shown on: Calls

Customer journeys

Follows what happened to each customer-started conversation and each caller who began in the month. Recent customers are still inside their follow-up window, so repeat figures read low.

Path

who replied

Reached an agent: Any person replied, not only tech support.
Bot or routing only: Closed with no human reply.
No human reply: Still open with no human reply.
Shown on: Customer journeys

Outcome

the first rule that matches, in this order

Order: Assigned to second line. Sent to another department, by a person's assignment or tag. Tagged bug_created, rework_created or api_transferred. Tagged resolved. Tagged no answer. Otherwise still open, or closed without an outcome tag.
Reading it: The bot's EscalatedTechnical tag is routing, so it is not an escalation here.
Shown on: Customer journeys

Summary tiles

each one is a share of a stated base

Reached a person: Reached an agent / all conversations.
Resolved, sent to another department: Those outcomes / conversations that reached a person.
Closed with no human reply: Closed by the bot or without a reply / all conversations.
Came back within 7 days: Same customer started a new conversation within 7 days / all conversations.
Callers missed on the first call: Callers whose first call was unanswered / all callers.
Shown on: Customer journeys

Times

medians, plus p25, p75 and p90 where shown

Wait for the first human reply: Conversation start to first human reply.
First reply to close: First human reply to final close.
Time to close: Conversation start to final close.
First reply to handoff: First human reply to the assignment outside tech support.
Shown on: Customer journeys

Effort and repeats

counts per conversation or customer

People: Distinct colleagues who replied.
Customer messages: Customer message parts, plus 1 for the opening message.
Reopened within 24 h: A real customer message within 24 h of the first close.
Started out of hours: Started outside 08:00-18:30 Yerevan.
Shown on: Customer journeys

Callers

each number once, from their first call in the month, followed for 3 days

Called again within 3 days: Another inbound call within 3 days.
Missed callers called back within a day: Missed callers we phoned within 24 h / missed callers.
Also wrote to tech support: Their Intercom contact has a conversation within 3 days either side of the call.
Median length of a missed call: Full duration of the missed first call.
Shown on: Customer journeys

Daily page

Month to date by day for the team or one agent. Every number is counted on the day of the event that ends it, in Yerevan time.

Bonus KPIs, month to date

the same FRT, FCR and close tagging as the Bonus page, for the 1st of the month up to each day

Pace: rate needed on the rest = (next level x (so far + rest) - value x so far) / rest, from counts, not averaged percents.
Rest of the month: Conversations per worked shift so far, times the shifts still on the schedule from tomorrow, absences taken out.
FCR by day: A day's FCR groups conversations by the day they were assigned and follows their outcome up to now, so the last 2 days are not final.
Close tagging: Higher is better. Above the gate there is no next level; below it, the page shows the rate that passes the gate.
Focus: For one agent: the tagging gate first when it is not met, then the bonus KPI with the most weight that can still reach its next level. Out of reach: how to hold the current level.
Shown on: Daily

Conversations handled

Conversations the agent replied in that day. A conversation counts once per agent per day.

Shown on: Daily

Handled per paid hour

Conversations handled divided by the scheduled net shift hours that day (Shifts page, breaks excluded).

Shown on: Daily

Open at day end

Conversations whose last customer message is newer than their last close at 24:00, counted for the agent they were last assigned to.

Shown on: Daily

Inbound calls

Inbound calls to the Support RU line that day (Unitalk). Team only; answered calls are on the Calls page.

Shown on: Daily

Customer waiting time

Median time from the customer's first message to the first agent reply, for first replies sent that day. Below it, the month split: working hours on the working-hours clock, with their share of first messages, and out of hours as customers lived it.

Shown on: Daily

Next reply time

How fast the agent answers once the dialog is under way: for every reply after the first, the time from the customer's first unanswered message to the agent's reply, counting only time when someone was on shift (Shifts page, absences out). The value is the median, the typical wait; below it, the share answered within 5 and 10 minutes, and the average wait of messages written when nobody was on shift. The hover shows the average, which a few stalled dialogs pull up.

Shown on: Daily

Stalled dialogs

Next replies that took over an hour of working time, counted for the agent who finally replied. Below it, the month total and what happened while the customer waited: reassigned (including Away handovers), an internal note, or an Intercom workflow; one stall can have several.

Shown on: Daily

Longest wait in a conversation

For each conversation, the longest wait for any agent reply, on the working-hours clock. The value is the 90th percentile: 1 conversation in 10 had a longer worst wait.

Shown on: Daily

Time to first assignment

Median time from the customer opening the conversation to its first assignment to an agent. It shows routing and coverage, not the agent's speed.

Shown on: Daily

Time to close

Median time from the conversation being created to the agent's close, for closes that day.

Shown on: Daily

Replies per closed conversation

Agent replies in a conversation up to the agent's close, averaged over that day's closes.

Shown on: Daily

Reopened within 24 h

Share of the agent's closes after which the customer wrote again within 24 hours; a thanks-only message does not count. The last day is not final.

Shown on: Daily

Escalation rate

Share of the conversations assigned that day that the agent transferred outside the in-house team or tagged for a technical escalation.

Shown on: Daily

Reassignment rate

Share of the agent's assignments that were later moved to a different agent.

Shown on: Daily

Data through 03 Nov 2026 06:06 (Yerevan time).