Managers Evaluation (agent performance)

The Managers Evaluation section helps you track support performance across the team and per individual agent based on your configured criteria. It gives you a clear, measurable view of quality and how it changes over time — so you can identify top performers, spot coaching opportunities, and monitor consistency.

How to use it

• Use Select Criteria to choose what you want to analyze (e.g., total score or a specific criterion).

• Filter by Date Range (quick presets like 1w / 2w / 1m / 6m / 1y or custom dates) to see trends for a specific period.

• Compare agents side by side to understand who is consistently meeting standards and where specific gaps appear.

💡Useful tips: Use this view to guide 1:1 coaching and team calibration: focus on agents with low or declining scores, drill into the criteria that drive the drop, and review conversation examples to pinpoint what to change. Over time, you can track whether coaching improves scores, validate that criteria are applied consistently, and spot recurring issues that may require updated playbooks, training, or knowledge base improvements.

Agent table colors

Colored cells compare the value with the configured target, or with the rest of the report when no target is set. Use significance test is off by default for new and existing columns: green means better, red means worse, and light grey means equal. The good direction (or category meaning) determines which is better. Hover the value to see the comparison and sample size.

Enable Use significance test below a supported column to require a statistically significant difference (two-sided p < 0.05). Green and red then mean significantly better or worse; light grey means the test ran and found no significant difference. Dark grey means there was not enough information to run the comparison: for example, only one observation, a missing or too-small comparison group, or no variation for a mean test. Hover for the specific reason. A zero-percent progress bar also marks its number in dark grey so this state remains visible.

The switch is available for category percentages, weighted scores, criterion/total scores, and Average aggregations of numerical, duration and clock attributes. It also applies to the right-hand metric in Pivot stacked bars. Counts, distributions, sums, medians, minima, maxima and predefined work-time metrics do not support it. Each column saves its own switch; changing only the aggregation preserves it. A new source starts with the switch off. A neutral direction or category meaning leaves comparison coloring off regardless of the switch.

Shares use Fisher’s exact test against the rest of the report or a binomial test against a fixed target. Means use Welch’s t-test against the rest or a one-sample t-test against a target. These tests use sample size and variation; a larger displayed value alone does not guarantee significance. Heatmap and the Number visualization retain their own rendering; comparison explanations remain available in tooltips.

Agent table totals and heatmap

In the report editor, Total row chooses Hidden, Top, or Bottom. Existing and new agent tables default to Bottom. The total stays in place when you sort agents; an empty report has no total row.

Totals use the current filters and accessible agents. Shared conversations count once. Numerical sums, averages, medians, minima and maxima are recalculated from their effective values; categorical counts, shares, distributions and weighted scores use unique conversations too. For total and criterion scores, available evaluation cards of the selected agents are averaged within each conversation first, then across conversations. Missing scores are omitted. Message counts include each eligible message once, and average start/finish times give each agent's working day equal weight in the project's timezone.

The total has neutral styling and does not enter statistical comparisons. Hover explanations still show category counts and unavailable-metric reasons. Include empty conversations for attribute-wide metrics controls the Base, Distribution and Weighted score columns; it is separate from Total row placement.

Heatmap is off by default and is saved per report. When enabled, each numerical column gets its own blue scale, including zero and excluding missing values and the total. Darker blue means a larger value, regardless of whether it is good or bad. Numbers keep their units, precision and hover explanations; distributions and progress bars keep their own colors and have no heatmap background. Turning Heatmap off restores each column's saved visualization. Pivot's separate Heatmap control is unchanged.

Agent table layout

Agent tables fill the report card. The agent column stays between 220 and 320 px; metric columns share the remaining space according to their headers, formatted values and visualizations. If the columns cannot fit at readable widths, scroll horizontally inside the card.

Names wrap to two lines, email addresses stay on one line, and metric titles wrap to three lines. Hover or focus shortened text to read it in full; header tooltips also retain the metric explanation. Numbers and units stay together. Sorting, moving the total between top and bottom, and toggling Heatmap preserve column widths. Percentage bars keep their display under Heatmap and stop growing at 320 px, including the number.

Statistical color settings

The meaning comes from your settings, never from the size of the number:

• Attribute values — the editor shows only controls used by the current aggregation. Distribution has order, meaning and colors; Percentage has meaning and target share for the selected category; Weighted score has the ordered scale, meaning and score points. Count and Total have no category settings block. Target share appears for a Positive or Negative category. In pivot tables, axes have order only. Stacked bars add colors on the column axis, plus either the selected category's target or weighted-score points. Choose Display, Metric beside bars and Compared value in the pivot editor to configure that view. Settings are retained when hidden and restored when switching back; changing the source clears them. Archived values remain in the scale.
• Numerical and time columns — in the dashboard editor, Average columns show Dashboard coloring below their main fields. Choose Higher is better, Lower is better, or Neutral (no status color), and optionally enter a Target average. These settings belong only to this column: the same attribute can have different targets in different reports. New columns start neutral with no target. Duration targets are entered in minutes; ordinary numbers use their original units. Clock-time columns have direction only.
• Criteria and total score — choose direction and an optional target below the column in the dashboard editor. Targets use percentages of the score scale (0–100%). New columns start neutral with no target. Changing project direction does not overwrite saved column settings.

When a target is set, cells compare with that target instead of the rest of the team. An empty target restores comparison with the rest of the team. Neutral disables comparison coloring while keeping the target saved. Switching from Average to another aggregation keeps the settings inactive until Average is selected again; changing the attribute clears them. Heatmap replaces statistical coloring for numbers and pills without erasing these settings; bars keep their colors.

Each column has independent settings, even when two columns use the same attribute. Pivot axes have their own settings. Changing a source resets its settings; changing aggregation keeps them for later use. Attributes and criteria contain only extraction and evaluation settings. New categories append without configured meaning; archived categories keep their position. Invalid points or targets block saving while their field is visible. Switching away restores the last valid value; valid hidden settings are retained.

Total, percentages and sorting

Percentages are shares of the conversations that have a value for the attribute, not of Chats count. Add a Total column (an Attribute categorical column with the Total aggregation and no category) to see that denominator: it counts the conversations with any value for the attribute. Hover a percentage to read it with its base, such as "4 of 6". A Total cell shows ! when more than a fifth of the agent's conversations have no value at all ("Not classified: 5 of 7") — usually a sign the attribute needs a better question or a re-run.

Empty conversations have no messages in the evaluation window. Count and Percentage columns keep their own Include empty conversations setting in Column settings; Base (labelled Total in the column editor), Distribution and Weighted score follow Include empty conversations for attribute-wide metrics in Report settings, one switch per report.

Click any column header to sort the whole table: first descending, then ascending, then back to the report's default sort. In Report settings you can pick the Default sort column and direction; without one, the first column that knows what "better" means leads with its best rows on top, and reports without any direction keep their usual agent order. Agents with small samples stay in the table; with the test enabled, a sample that cannot support the test is marked dark grey.

All percentages show one decimal (10.0%, 8.6%) in agent tables and in Pivot reports.

Distributions and visualizations

A whole attribute fits in one column. Add an Attribute categorical column with the Distribution aggregation and no category: it draws a 100%-stacked bar whose segments follow the order and colors you set directly below this column, so five outcomes no longer need ten columns. Each Distribution column carries its own legend in its header, which is why a report can show two distributions without confusing their values. A segment is labelled with a whole percent only when it is wide enough to be legible (8% or more); hover any segment for the exact reading, such as "Escalated — 63 of 667 (9.4%)". Conversations whose value has since been deleted from the attribute are kept together as Other values, so the bar always adds up.

Clicking a Distribution header sorts the table by the share of one value — by default the first value that means something bad, so the rows that need attention come up first. Pick a different one under Sort by value in Column settings; the header says which value is in use.

Percentage columns have a Show progress bar checkbox directly below the main fields. It is on by default. Turn it off to show just the percentage, which can receive heatmap coloring. When the bar is on, Heatmap keeps the bar, its statistical color and its tooltip, without coloring the cell background. The total row remains a neutral number.

Column settings also has Visualization. A share is drawn as a progress bar whose length is the value and whose color is the status (green, red or grey — never the size of the number). A criterion average is a score on its own scale rather than a part of a whole, so it reads as a plain percentage with no bar and no color; Status pill puts it back on a colored plate and Progress bar gives it the bar, for the reports that want them. Number always means just the number — the comparison stays in the tooltip either way. Counts and totals are numbers, a directed Avg wears a status pill, and a Distribution is always a stacked bar.

One score per agent

For a scale like CSAT or CES — values you have marked as ordered, best to worst — the Weighted score aggregation turns the whole attribute into a single number from 0 to 100. Each value is worth its score points, derived from the order and meaning you gave it: the positives step down from 100 towards 50, the negatives from 50 down to 0, and anything neutral sits at 50 — so Satisfied / Neutral / Dissatisfied gives 100 / 50 / 0, and adding Highly dissatisfied as a second negative gives 100 / 50 / 25 / 0. Edit Score points directly below this column or the pivot column axis when the weighted score is selected. The score is the average over the agent's classified conversations. It renders as a pill with one decimal, colored like every other cell: by direct comparison with the rest of the report, or by significance when Use significance test is enabled. Hover it to see the counts and points the score is made of.

For a single-choice attribute, select Weighted score first, then enable the ordered scale and assign at least one meaning before saving. Multiple-choice attributes cannot calculate a weighted score. Changing order, meaning or points affects only this column or pivot axis; other reports keep their own settings.

Deleting a value removes it from future selections while preserving its place in the scale and its contribution to historical scores. If an attribute's settings no longer support a weighted score, the saved report still opens and shows a dash with the reason. You can rename the report or choose another metric. Historical conversations with multiple selected values also make the weighted score unavailable for that filtered report, even if multiple selection has since been switched off.

Unavailable weighted scores stay blank in exports; the report parameters explain why they are unavailable. A saved report with an archived numerical axis can still be renamed and saved. When changing an attribute in the editor, an incompatible aggregation is cleared so you can choose a valid one. Numerical targets accept finite numbers or an empty value to clear the target.

Comparisons with the rest of the report count each conversation once and exclude every conversation in the current row. A conversation shared by several categories or agents does not become several independent observations. If it has separate agent evaluations, its comparison score is the average of those evaluations; each agent's own row keeps their personal score. Without a sufficient comparison sample, a colored value is dark grey with an explanation.

In Pivot reports with the Percentage metric, the Stacked bars button turns each row into one bar, with the legend above the table and the share of a chosen value on the right in its status color. Use Show to pick which value that is — it is the first negative one by default, and for an ordered column attribute you can show the row's score instead (CES score by enquiry topic, say). Bars always read along the row, so the Of picker steps aside in this mode. Table brings the grid back, and both modes save with Save changes.

Pivot reports

Open Dashboard → Custom dashboards, choose New dashboard, and select Pivot report. Choose two different categorical attributes for rows and columns, then choose a metric:

• Conversation count counts unique conversations in each intersection.
• Percentage uses the row as its default denominator. You can instead select the column or all conversations.
• Average, Median, and Sum aggregate a numerical or duration attribute. Hover a number to read how many of its conversations actually carry a value ("2 of 200 conversations have a value") — worth a look before trusting a small sample. Zero is a value; missing values are excluded. A dash means there are no numeric values.

Columns can also be a number instead of a set of values — pick one under Numbers in the same list. The report then has a single column: the chosen aggregate of that number for each row, which is how you answer "which topics take longest to resolve" (rows Enquiry Topic, columns Resolution Time, metric Median, then sort by the column). A number is aggregated on its own, so there is no second attribute to choose and no percentage to take; the Total column reports how many conversations are behind each row instead of repeating the number beside it. Rows always stay a set of values.

The report uses the same shared filters as the other custom dashboards and its own date range, interpreted in the project's reporting timezone. A one-off evaluation launch replaces the date filter. Include missing categories puts conversations without an axis value in No value; turning it off excludes them before calculating percentages. With a number as the column axis it applies to the rows alone — a number has no categories to miss, and conversations without one are still counted; hovering the number says how many carry a value.

The bottom row and right column show totals. When an attribute allows multiple categories, one conversation can appear in several cells. Totals count each conversation once, and percentages across overlapping categories can add up to more than 100%. Numeric totals are recalculated over the distinct conversations rather than averaged from cells.

Click a cell or total to open the existing Conversations page with that selection and the active report filters. Numeric cells open all conversations in the intersection, including those without the numeric value. Clear the pivot filter to leave the intersection. Browser Back restores the report settings. If someone edits the saved report after you open a cell, reopen the report to get a fresh selection.

Rows without conversations are hidden; Show empty rows brings them back, and a line under the table says how many are hidden. Empty columns stay visible. Click a row, column or Total header to sort the rows, and save the result with Save changes to make it the report's default sort. With Percentage, the Total column and the bottom Total row show the number of conversations instead of a trivial 100%; the percentage stays in the tooltip.

Dashboards open expanded whenever you enter Custom dashboards. Heatmap is enabled by default to highlight larger values; click it to turn the cell coloring off or back on. Change the metric or percentage direction above the table; users with editing access can select Save changes to make those settings the saved defaults. Use Edit to change axes or missing-value handling.

Export CSV and Export Excel export the complete filtered matrix, including totals and sample counts. Excel includes separate Report, Conversation counts, Parameters, and (for numeric metrics) Numeric samples sheets. Percentage values in Excel are numeric percentages.