> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tryhamsa.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Analytics: how your agents' calls are going

> Read the Analytics page: call volume, reply time, call length, sentiment, CSAT, NPS, resolution and errors, and exactly how each figure is counted.

Analytics shows how calls went across your agents, or for one agent, over a period you choose. Use it to spot changes in volume, slow replies, unhappy callers and calls that needed a person. Most cards link into the [Conversation log](https://docs.tryhamsa.com/agents/monitor/conversation-log.md), so you can open the calls behind a figure.

![The Analytics page for All agents and Today, with the Calls chart, When calls come in, Reply time, Call length, Talk time, How calls ended, Forwarded to a person and Channels cards](https://mintcdn.com/hamsa-452fa79a/Rl78D_qJIpinmg19/images/agents/monitor/analytics-light.webp?fit=max\&auto=format\&n=Rl78D_qJIpinmg19\&q=85\&s=0bf7550de0d79ef161a2e7311ab35c42)

## Before you start

You need the **View** permission on **Statistics**. Owners, Admins and Users have it by default. Without **View** on the **Conversation log**, you still see Analytics, but without the **How calls ended** and **Channels** cards and without links into the log. See [Roles and permissions](https://docs.tryhamsa.com/overview/projects/roles-and-permissions.md).

## Choose the agent and the period

**1. Open Analytics**

Click **Analytics** in the sidebar, or the **Analytics** card on [Home](https://docs.tryhamsa.com/agents/monitor/home.md).

**2. Pick an agent**

The agent filter starts on **All agents**. Pick one agent to see only its calls. You can search the list; each agent shows whether it is a flow or a single prompt agent.

**3. Pick a period**

Choose **Last hour**, **Today** (the default), **Yesterday**, **This week (from Sunday)** or **This month**. For other dates, choose **Custom**, pick a start and an end on the calendar, and click **Done**. Future dates can't be picked.

![The period menu open under Today: Last hour, Today, Yesterday, This week (from Sunday), This month and Custom](https://mintcdn.com/hamsa-452fa79a/Rl78D_qJIpinmg19/images/agents/monitor/analytics-period-menu-light.webp?fit=max\&auto=format\&n=Rl78D_qJIpinmg19\&q=85\&s=082555c7d915bfd446dad549b7cbb034)

The agent and the period are kept in the page address, so you can bookmark a view or send it to a teammate.

Periods follow the time zone of your browser. "Today" starts at midnight where you are, and **This week** runs Sunday to Saturday.

You can also change the agent, the period or export from the command menu (Cmd+K on Mac, Ctrl+K on Windows): **Filter by agent**, **Filter by date**, **Export analytics**.

## Read the change since the period before

Next to each headline figure, a small badge shows how it moved compared with an earlier stretch of the same length. It appears once the period has calls.

| Period | Compared with |
| - | - |
| **Last hour** | the hour before |
| **Today** | yesterday, up to the same time of day |
| **Yesterday** | the day before |
| **This week** | last week, up to the same point |
| **This month** | last month, up to the same point |
| **Custom** | the same number of days just before |

## What each figure means

Every card has a **?** button ("What do these figures mean?") with a short definition. The tables below say exactly what is counted and what it is divided by.

Two terms come up again and again:

* **Every call** means every call in the period in any status: completed, forwarded, failed, no answer, terminated, in progress and pending.
* **Finished calls** means calls with the status Completed or Forwarded. A forwarded call finished like a completed one; it was transferred to a person on the way out. Length, reply time, scores and resolution are measured on finished calls only.

### Volume and timing

| Card | What it shows | How it's counted |
| - | - | - |
| **Calls** | The number of calls, an average per hour, day or week, and "N live now". | Every call in the period, whatever its status. "Live now" counts calls in progress that started in the last 10 minutes; click it to open [Live calls](https://docs.tryhamsa.com/agents/monitor/live-calls.md). **See these calls** opens them in the log. |
| **When calls come in** | A grid of calls by hour of day, the busiest hour, and "Half of the calls came between ... and ...". | When each call started. For a period under a week, one row per day. From a week up, one row per weekday (Monday first), and each cell is the average over those days. Needs a period of a day or longer. |
| **Talk time** | Time "spent on calls" and "Calls average N messages". | The lengths of all finished calls, added up. Messages are the caller's and the agent's turns per finished call, on average. Tool calls are not counted as messages. |
| **Call length** | The "average call", a bar of **Less than 30s**, **30-120s** and **More than 120s**, and "Agent replies average N words". | Average length of finished calls, and how many fall in each band. Words per reply is all the words the agent said in finished calls divided by the number of agent replies. Click a band to list its calls. |

The **Calls** chart has one column per hour for periods up to two days, one per day up to 62 days, and one per week beyond that. **Last hour** has no chart. Click a day's column to narrow the whole page to that day; **Back to ...** returns to your period.

### How calls ended and where they came from

| Card | What it shows | How it's counted |
| - | - | - |
| **How calls ended** | The share "of calls completed", and a bar of Completed, Forwarded, No answer, Terminated and Failed. | The calls in the period that have ended, by status, counted the same way as the Conversation log. Calls still in progress or pending are left out until they end. Click a status to list its calls. |
| **Forwarded to a person** | The share "of calls" handed to a person, as **Forwarded** and **Kept by the agent**. | Calls handed to a person, divided by the finished calls the analysis could judge (see [How calls are scored](#how-calls-are-scored)). A call is handed to a person when it was transferred, or when the caller asked to speak to a person during the call. This is the same figure as **Escalation rate** below. **See the forwarded calls** lists calls with the status Forwarded. |
| **Channels** | The share of calls that "came by phone", "came from the web" or "came through the chat agent". | Every call in the period, by channel, as the Conversation log's **Channel** filter counts them. Each share is out of all three channels together. Click a channel to list its calls. |

### Reply time and errors

| Card | What it shows | How it's counted |
| - | - | - |
| **Reply time** | The "average per reply", split into **Speech to text**, **Model** and **Speech**. | On each call, the platform records three times per reply: how long speech to text took to turn the caller's words into text, how long the model took to produce its first words, and how long speech took to produce the first sound. Each call keeps its average for each stage. The card averages those across finished calls and adds the three together. |
| **Calls with errors** | The share "of calls had an error", or "No errors", and "N calls ended in failure". | Finished calls where speech to text, the model or speech reported an error at some point, divided by all finished calls. The second line counts calls with the status Failed. They are not the same thing: a call can have an error and still finish, and a failed call never finished. **See failed calls** lists them. |

When the average is over 1.5 seconds, the card warns that replies are slow. The same limit marks a call **Slow replies** in the Conversation log.

### How callers felt and how calls were resolved

| Card | What it shows | How it's counted |
| - | - | - |
| **Sentiment distribution** | The share "of calls positive", and a bar of **Negative**, **Neutral** and **Positive**. | Each finished call gets a sentiment score from 0 to 100. 75 and up is positive, 50 to 74 neutral, below 50 negative. The shares are out of the finished calls that have a score. |
| **CSAT score** (in **Outcomes**) | Average satisfaction, from 0 to 100%. Higher is better. | The average of each finished call's CSAT score (0 to 100), over the calls that have one. The score rates helpfulness, professionalism, how well the issue was resolved, clarity and promptness. |
| **First-call resolution (FCR)** (in **Outcomes**) | The share of calls the agent resolved without help. Higher is better. | Finished calls judged resolved and not handed to a person, divided by the finished calls the analysis could judge. A call is not resolved when the conversation says plainly that it must go to a person: the caller asks for a human, or the agent says the case will be transferred, escalated, or followed up or called back by a person. |
| **Escalation rate** (in **Outcomes**) | The share of calls handed to a person. Lower is better. | Finished calls handed to a person (transferred, or the caller asked for a person), divided by the finished calls the analysis could judge. |
| **NPS** (in **Outcomes**) | Promoters minus detractors, from -100 to +100. | Each finished call gets a 0 to 100 "likelihood to recommend". 90 and up is a promoter, 60 and below a detractor. NPS is the percentage of promoters minus the percentage of detractors, over the finished calls that have a score. |

![The lower Analytics cards: Sentiment distribution, Outcomes with CSAT, first-call resolution, escalation rate and NPS, and Calls with errors](https://mintcdn.com/hamsa-452fa79a/Rl78D_qJIpinmg19/images/agents/monitor/analytics-scores-light.webp?fit=max\&auto=format\&n=Rl78D_qJIpinmg19\&q=85\&s=ba4ada9c48ec64f5a22fce61eeaa342a)

FCR and escalation rate don't have to add up to 100%. A call the agent didn't resolve isn't always a call the caller asked to hand off (for example, the agent promised a callback), and a call the analysis couldn't judge counts in neither.

## How calls are scored

CSAT, NPS, sentiment and resolution are not caller surveys. Nobody asks the caller anything. After each call ends, an AI model reads the transcript and answers two sets of questions:

1. **Scores**: sentiment, CSAT and likelihood to recommend, each from 0 to 100.
2. **Resolution**: whether the call was handled by the agent alone, whether the caller asked to speak to a person, whether a gap in knowledge stopped the agent from answering, and which topics the caller asked about that the agent couldn't handle.

A few rules decide what counts:

* **Calls where the caller never spoke are not scored.** A call with no answer, a lone greeting or an instant hang-up gets no scores and no verdict, so it doesn't pull any rate up or down.
* **A transferred call always counts as handed to a person**, whatever the transcript says, and it always counts as judged.
* **A call the analysis couldn't judge counts in no rate.** If the analysis fails for a call, that call drops out of the sample instead of counting as resolved.
* **The analysis is strict about escalation.** Frustration or a neutral ending is not enough: the transcript has to say that a person must take over.

The results for one call appear in its **Outcome** tab under **Call analysis**. See [Call details](https://docs.tryhamsa.com/agents/monitor/call-details.md#outcome).

## Export the figures

Click **Export** to download an Excel file (.xlsx) titled "Hamsa Analytics Report", with an **Overview**, a **Performance** and a **Satisfaction** sheet for the agent and period on screen. The file name has the period in it, for example `analytics_report_today_2026-10-31.xlsx`.

## When there's nothing to show

* A period with no calls says so ("No calls today", "No calls this week" and so on), offers **Today**, **This week** and **This month**, and tells you when the last call was, with **See it**. A project or agent that has never had a call says "This project has no calls yet." or "This agent has no calls yet."
* Reply time, length, scores and resolution fill in as calls finish. A call that is still in progress counts in **Calls** but nowhere else yet.
* The analysis runs after the call ends, so scores for the last few calls can take a moment to appear.
* If one card fails to load, it says "Couldn't load this." with **Try again**. The other cards still work.

## Good to know

* The only threshold the product sets is reply time: over 1.5 seconds is slow. There are no built-in targets for CSAT, NPS or resolution; compare your agent with its own earlier periods.
* Satisfaction scores are an AI reading of the transcript. Use them to find calls worth listening to, not as a substitute for asking customers.
* Analysis looks at the transcript only. If the agent did something off the call (for example, a tool booked an appointment) but never said so, the analysis doesn't see it.

## Related

> [card] [Conversation log](https://docs.tryhamsa.com/agents/monitor/conversation-log.md)
>
> Open the calls behind any figure, with the same analysis per call.

> [card] [Call details](https://docs.tryhamsa.com/agents/monitor/call-details.md)
>
> Transcript, recording, call analysis and reply time for one call.

> [card] [Outcome dashboard](https://docs.tryhamsa.com/agents/monitor/outcome-dashboard.md)
>
> Build charts for one agent from its own outcome fields.

> [card] [Write a prompt that works on calls](https://docs.tryhamsa.com/agents/build/write-a-prompt.md)
>
> Fix what the analysis finds.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.