Sona AI Chat (also called Ask Your Data) is a natural-language assistant built into the platform. Instead of building a report or writing a query, you ask a question in plain English and Sona AI answers using your connected data. It's the fastest way to interrogate your revenue, attribution, account, and pipeline data, and every answer is grounded in a data source you choose, so you stay in control of what the AI is reasoning over.
New to the chat screen? Start with Getting Started with the New Chat Screen, which covers the New Chat landing screen — the chat input, connecting your data, and the starter templates. This article picks up from there.
Anatomy of the chat screen
The screen has three zones:
1. Chat rail (left). Manage your conversations.
New Chat — start a fresh conversation.
Search Chat — find a past conversation.
Playbooks — run a pre-built analysis template instead of writing a prompt from scratch.
2. Data context (centre). This is what makes Sona AI's answers specific to your data.
Model selector (top) — choose which AI model answers (e.g. GPT-4.1 mini). Other models are available from the dropdown.
Analysis type — the dataset the chat analyses (e.g. Revenue Attribution). This is the same selector you set on the New Chat screen; see the Getting Started article for the full list of analysis types.
Attribution model — for analysis types that support it (such as Revenue Attribution), a second selector sets the attribution model (e.g. First Touch Credit), which changes how credit and values are assigned.
Upload files — bring your own file (e.g. a CSV) into the conversation to analyse alongside your platform data.
Data preview — a live preview of the selected data so you can confirm you're pointed at the right thing before you ask. The preview shows the first 25 rows; use the row selector and pagination to scan more.
3. Conversation (right). Your messages and Sona AI's responses. Type in the box at the bottom — Enter sends, Shift+Enter adds a new line.
Setting your data context before you ask
Sona AI answers within the context you select, so set it first:
Pick an analysis type (e.g. Revenue Attribution).
If the analysis type supports it, pick the attribution model (e.g. First Touch Credit) — this determines how credit and values are assigned in the answers.
Confirm the data preview below shows the data you expect.
Because answers are scoped this way, the same question can return different results under a different lens. For example, asked which model it's using, Sona AI will respond based on the attribution model you've selected — first-touch credit attributes revenue and conversions to the first marketing touchpoint a customer interacted with.
Choosing the AI model
The model selector at the top lets you choose the underlying AI model that generates answers. Models differ in speed and depth, so if an answer feels too shallow or too slow, switching models is the first thing to try.
Uploading files
Use Upload files to bring your own data into the conversation. This is useful when the data you want to analyse isn't already in a Sona data source — upload it, then ask Sona AI questions about it the same way.
Playbooks
Playbooks are pre-built analysis workflows you open from the left rail — a fast way to run a common analysis without phrasing the prompt yourself. (These are different from the starter template cards on the New Chat screen, which are covered in Getting Started with the New Chat Screen.) Open Playbooks, pick one that matches what you're trying to learn, and run it against your selected analysis type.
Managing your chats
New Chat clears the context and starts over — use it when you switch topics, so earlier questions don't colour new answers.
Search Chat lets you find and return to a previous conversation.
Tips for better answers
Set the analysis type (and attribution model, if shown) first. The answer is only as relevant as the context it's grounded in.
Ask one thing at a time. Narrow questions get sharper answers than broad ones.
Be specific about time frames and segments (e.g. "by marketing channel, last quarter").
Check the preview if an answer looks off — you may be pointed at the wrong analysis type or attribution model.
Sona AI can make mistakes. Responses are AI-generated. Verify important numbers against the underlying report before you act on them.