# Test your agent

> Play the lead in the Test tab of an agent: first message, conversation, follow-ups and booking, with nothing sent to your real contacts.

Source: https://docs.whatsetter.com/dashboard/playground/

The **Test** tab lets you talk to your agent before a real contact does. You
play the lead, the agent answers with its real instructions and knowledge, and
nothing is sent for real. The Playground of the old menu is now this tab,
inside each agent.

## Open the Test tab

1. Open **Agents** and click your agent.
2. Click the **Test** tab, or the **Test** button at the top of any other tab.

The tab has three steps, the same as the **Prompt** tab: **First message**,
**Conversation** and **Follow-ups**. One conversation runs through them: you
go from one step to the next as the test moves on. On the right, **This test**
shows what the agent concluded, the lead you play and the version tested. On a
narrow window, **Test details** opens that panel.

Any member who can see the agents can test them. The Playground is now this
tab: an old link to it opens the Agents list, where a click on the agent opens
its **Test** tab.

## What is real and what is not

A test is a simulation. The first screen says it: "Simulation: no lead
contacted, no real meeting, no effect on your stats."

| Same as in a real conversation | Not real |
|---|---|
| The agent's instructions and its first message | No WhatsApp message is sent |
| Its knowledge | No lead is contacted |
| Its style | No meeting is booked in your calendar |
| How it qualifies the lead or stops | The figures of an A/B test do not move |
| The free slots of your calendar, read only | No waiting: a follow-up is generated at once |

One thing is real: each answer the agent generates uses half an assistant
credit.

## Start a test: who writes first

Open **Agents**, click your agent, then **Test**. An empty test asks
**Who writes first?**

- **Julia writes first**, with the name of your agent: like a new lead from one of your lists, the agent sends its first message. The test opens on the **First message** step.
- **Alex writes first**, with the first name of the lead you play: like a lead who writes to you after an ad, a link or your website. The test opens on **Conversation**.

If the agent has no first message, only the second choice is offered: such an
agent answers when someone writes to it.

You change the first name you play in **This test**, under
**Lead's first name**. The agent uses it like the first name of a real lead.

## Test the first message

1. Open **Agents**, click your agent, then **Test**, and stay on the **First message** step.
2. Click **Generate the first message**. The message appears in the conversation, marked **First message**.
3. Click **Generate another one** to get a different message, or **Answer as Alex** to go on as the lead.

The agent writes a new message each time, as it does for each real lead.

If your first message uses columns of your lists, **This test** shows a line
such as **Lead details (2)**: open it and type the value of each column for
the lead you play. A column left empty takes an example from your lists.

Once you have answered, the first message of this test no longer changes. To
try another one, click **New test**: the current test stays in the history.

## Test the conversation

1. Open **Agents**, click your agent, then **Test**, and open the **Conversation** step.
2. Write in the field at the bottom as if you were the lead, then press Enter.
3. The agent answers in one or several messages, depending on its **Split long replies** setting, and at most as many as **Messages per reply** allows.
4. Go on: an objection, a question off the subject, a request for a call.

Above the field, message ideas such as **Ask the price**,
**Say you're short on time**, **Accept a call** or **Say no** fill the field
for you. Nothing leaves until you send, and an idea costs nothing by itself.

Replies arrive faster than on WhatsApp: the test skips the waiting time of a
real conversation.

## Test a follow-up

A follow-up is tested by pretending that the lead stays silent.

1. Open **Agents**, click your agent, then **Test**, and get at least one message from the agent.
2. Click **Don't answer**, or open the **Follow-ups** step.
3. The step says how long the lead stays silent and which follow-up comes next. Click **Generate follow-up 1**.

The conversation shows the silence, for example "1 day later, simulated", then
the follow-up. No time is really waited. Click again for the second follow-up
when the agent has one.

The follow-up tested is the one for **Never replied** when the lead never
wrote in this test, and the one for **Stopped replying** when they did.

Follow-ups are written in the **Prompt** tab: see
[Automatic follow-ups](/dashboard/follow-ups/).

## Why a follow-up cannot be generated

On the **Follow-ups** step of the **Test** tab, the bottom of the conversation
names the reason:

| What you read | What it means |
|---|---|
| **Nothing to follow up yet** | The test is empty. Start with the first message or with a conversation. |
| **Follow-ups turned off** | The agent does not follow up in this situation. **Set up follow-ups in Prompt** opens the right place. |
| **Follow-up 1 has no instructions** | The follow-up is on but its text is empty. |
| **Julia has to answer first** | The last message is the lead's. A follow-up comes after a message of the agent. |
| **No more follow-ups planned** | The agent has sent all its follow-ups for this situation. |
| **Julia ended this conversation** | The agent qualified the lead or decided to stop. No follow-up goes out after that. |

## Test booking without booking anything

When booking is set up for the agent, in its **Settings** under **Meetings**,
**This test** shows **Test options** with the switch **Simulate booking**.

- On: the agent offers real free slots from your calendar. When it books one, the conversation shows "Meeting booked in simulation. Nothing is booked in your calendar."
- Off: the agent does not offer slots in this test.

The **Meeting** line of **What the agent concluded** follows the booking:
**No slot offered**, **Slots offered**, **Waiting for their email**, then
**Booked (simulation)**.

Setting up booking is covered in [Meetings](/dashboard/bookings/).

## Read what the agent concluded

In the **Test** tab, the block **What the agent concluded** shows the decision
the agent took about the lead after each exchange.

| Line | Values |
|---|---|
| **The analysis's verdict** | **Not assessed yet**, **To be determined**, **Qualified**, **Outside your criteria** or **Declined**. |
| **Next action** | **Keep the conversation going**, **Mark as qualified**, **Stop** or **Pick the conversation back up**. |
| **Stage in Leads** | The stage the agent would set on a real lead, under its name in **Leads**: for example **In conversation** or **Qualified**. |

The agent assesses the lead after a few exchanges, not on the first reply.

"Your criteria" are in the agent's prompt, in its **Conversation** step: its
goal and the path of the conversation, or a "Qualification criteria" section
in an older prompt. Labels play no part in it. To change them, see
[Deciding when a lead is qualified](/dashboard/prompt/#deciding-when-a-lead-is-qualified).

The conversation also notes what the agent does without writing:

- "Julia qualified this lead." The lead meets your criteria: the agent no longer follows up.
- "Julia stops here: the lead is outside your criteria."
- "Julia doesn't answer this message." The agent chose to stay silent.
- An emoji placed on your message: the agent reacted instead of writing.

## See why the agent gave an answer

In the **Test** tab, open **Why this answer** under a reply of the agent. It
shows what the analysis read before that reply:

- **Criteria noted by the analysis**: each criterion of your prompt, **Met**, **Not met** or **Missing**, with the words of the lead that count;
- **Knowledge base excerpts**: what was given to the agent from your knowledge base to write that reply.

The criteria analysis starts at the 4th message exchanged, not counting the
opener and the follow-ups. It starts earlier when the lead talks about
booking a meeting or asks to stop. Once the lead is qualified, the analysis no longer
reads the criteria. A follow-up is written without it.

## Choose which version you test

In the **Test** tab, the block **Version tested** has one line for
**First message** and one for **Conversation**.

- **Live version**: the text has a single version, the one your leads receive.
- A list appears when there is a choice. **As live** tests what your leads receive today. Each variant is listed with its state: live, saved draft or paused.

When you saved a draft in the **Prompt** tab, the test uses it by default and
says so: "This test uses your draft. Your live agent doesn't change."

Changes you did not save in **Prompt** are not tested. Changing the version
once it has spoken starts a new test. **Edit the prompt**, at the top right,
opens the text of the step you are on. Your test stays saved.

## Start a test over

In the **Test** tab, click **Start over** above the conversation. A new, empty
test begins with the same lead and the same versions. The test you leave is
not lost: it stays in the history.

**Start over** is greyed out while the test is empty or while the agent is
writing. **New test**, offered at the bottom of some steps, does the same.

Nothing has to be reset on the agent itself: a test leaves no trace in its
real conversations.

## Find or delete a past test

Tests save themselves after each exchange: **Saved** appears next to the title
of the test.

1. Open **Agents**, click your agent, then **Test**.
2. Click **History**. The panel **Your tests** lists the 20 most recent tests of this agent.
3. Click **Resume** on a test to reopen it and go on.

Each test shows the first message of the lead, who wrote first, the number of
messages, its age and, when there is one, its outcome: **Qualified**,
**Outside your criteria** or **Meeting booked**.

To delete a test, click the bin on its line, then **Delete**. **Keep** cancels.

A reopened test made with an earlier version of your text says so. What you
write next uses the current version.

## What a test costs in credits

Each generation in the **Test** tab uses half an assistant credit: one reply
of the agent, even when it is split into several messages, one first message
or one follow-up. The tab says it next to the buttons: "Each generation uses
credits."

- A message idea, a change of the lead's first name or a reopened test costs nothing.
- A generation that fails is not charged.
- Credits belong to your team and refill every month.

When none are left, the field reads "No credits left to test" and the buttons
are greyed out. Credits come back when your plan renews, or with a higher
plan: see [Plan & billing](/dashboard/billing/).

## When a reply cannot be finished

If the agent's answer fails, the conversation shows "The reply couldn't be
finished." with **Try again**. The retry runs the same step again without
sending the lead's message twice, and the failed attempt is not charged.

If the tab itself does not open and reads "The test couldn't load", check your
connection and click **Reload**. Your agent is not affected: tests and real
conversations are separate.

If a past test does not reopen, a message says "This test couldn't open. Try
again."

## Test a group agent

A group agent has two situations in place of the three steps. Open **Agents**,
click **See group agents**, click the agent, then **Test**.

- **In the group**: write like a member who mentions the agent in the group.
- **In private**: write like a member of the group who messages it privately.

Nothing is sent to your real group. Going from one situation to the other
starts a new test. See [WhatsApp groups](/dashboard/groups/).

## Related

- [Your agent's prompt](/dashboard/prompt/)
- [Agents](/dashboard/agents/)
- [Knowledge bases](/dashboard/knowledge-base/)
- [Meetings](/dashboard/bookings/)
