Your First Session
Learn how to open your Agent, ask your first question, read the response, and recognize when your Agent needs more data.
Start Here · 2 of 3
Open your Agent, ask your first question, and learn how to read the response it gives back.
There’s nothing to set up before your first session. Type a question, or pick one of the starter cards, and watch how your Agent works through the answer.
1. When you open your Agent
Your Agent opens to an empty chat. There are four things to notice before you ask your first question.
The empty Agent · what you see on first open
The chat input
The “Message Agent…” field at the bottom. Type a question in plain English. Hit Enter.
Three starter cards
Sample prompts that span what your Agent is built to do — find inflection points, connect customer voice to revenue, or surface risk. Click one to drop it into the input.
Conversation history
The left rail shows past chats. New workspace, no chats yet — that’s fine.
Settings, files, voice
The icons under the input attach files, adjust settings, and enable voice input. Skip them for your first session.
2. Ask a real business question
Type a question and submit it. For example:
Example question
What is driving customer dissatisfaction?
When you ask, your Agent works through the request in stages.
Your Agent makes a plan
Before answering, it lays out the steps it intends to take. You’ll see a small task list appear — what it’s going to inspect, what it’s going to query, what it’ll synthesize at the end.
It inspects your data
It checks what tables, prompts, and dimensions are populated in your workspace, and what time range they cover. This is where it figures out whether your question is actually answerable yet.
It runs queries and pulls evidence
It writes and runs the queries it needs against your data — counts, filters, aggregations, transcript searches. You can watch this happen in real time.
It writes back with findings
A headline answer, the methodology it used, evidence with numbers, verbatim quotes from the underlying conversations, and a recommended next step. Below is what part of that response looks like.
Example response
Here’s an excerpt of what part of the response can look like.
You
What is driving customer dissatisfaction?
Agent
Across 4,388 conversations (Jan 1–Feb 1, 2026), the biggest drivers of dissatisfaction are billing, promo, and credits problems, followed by account access friction. Using a transcript-based proxy, 30.9% of conversations (1,356) show clear dissatisfaction language.
Important caveat: your usual sentiment dimensions aren’t populated, so this is based on patterns in transcript text — not your configured dimensions.
Representative verbatims
“BOGO promo isn't applying... being charged the full EIP...”
“I just want to pay my bill, not troubleshoot an app.”
“I canceled... but I'm still being charged $55.”
This is an excerpt. Full responses include weekly trend charts, more themes, and recommended actions.
3. Read the answer
Every answer your Agent writes back follows roughly the same shape. Once you’ve seen one, the rest are easier to scan.
Headline first
One or two sentences with the punch — what the data says, with the topline numbers.
Methodology disclosed
If your Agent had to use a proxy or work around missing data, it tells you. You’ll see phrasing like “based on transcript patterns, not configured dimensions.”
Evidence with numbers
Counts, percentages, time windows. Anything claimed comes with the number behind it.
Verbatim quotes
Direct quotes pulled from the underlying conversations. These are the rawest signal — read them.
Recommended next step
A path forward — usually either a deeper question to ask or a setup move to make the next answer more reliable.
4. Recognize when your Agent doesn’t have enough data
Your Agent refuses to answer rather than guess. The refusal is honest and specific: it tells you exactly what’s missing and what to do about it.
Sample refusal
Sample refusal
“I can’t yet answer ‘what’s driving dissatisfaction’ from your labeled dimensions, because the two tables that normally contain dissatisfaction and driver dimensions are currently empty in your workspace.”
What I checked
system_session_prompt(expected sentiment, rating, reason dimensions) — 0 conversationsunstructured_analysis(expected issue category, predicted CSAT) — 0 conversationstranscripts(raw transcript text) — 4,388 conversations
What I can do next
Run a lightweight theme extraction directly from transcript text, which is less reliable but available now, or set up the enrichment dimensions and rerun on the full dataset.
From here, you have two moves:
- Tell your Agent use available data to take the interim path. That’s how the example answer above was produced.
- Set up the missing dimensions in your workspace, then ask again.
Trust the refusals. Most analytics tools fail by quietly making things up. Your Agent fails loudly: it tells you what’s missing, then offers a path forward.
5. Optional: Explore before you analyze
If you’d rather understand what your Agent can access before asking an analytical question, start with discovery prompts.
- What’s in my workspace?
- List my prompts.
- What dimensions do I have for sentiment?
- Do we track competitor mentions anywhere?
Your Agent will inventory what’s available and tell you what kinds of questions it can answer well. The full discovery patterns — drill-downs, “where do we capture X,” and “help me find the right dimension” — are covered in What the Agent can see.
Up next
Updated 5 days ago