Count AI

Spend less time remembering that date function or tweaking that visual.

Count AI is a suite of AI-powered features that can perform various analytical tasks, such as editing visuals, editing SQL or Python, explaining the logic behind cells and even generating whole analyses. It's designed to both accelerate expert users in their work, as well as help beginner users get started with Count more quickly.

Count AI is currently in beta. We'll continue to add more functionality and improve the experience based on user feedback.

Availability

• Count AI is available for all paid accounts.

• Anyone who can edit code in a canvas can use Count AI. (We plan to open up Count AI to other users in the future.)

Enabling Count AI

By default, Count AI is disabled for all workspaces created before 15 July 2024. Workspace admins/owners can enable AI in the workspace settings (under Other settings):

Once enabled, Count AI can be disabled at any time by toggling the same switch — this immediately disables the feature for all members of the workspace.

Using Count AI

Count AI can help you analyse your data in various ways:

• Generating analyses from a text prompt

• Generating analyses from a canvas prompt

• Editing visuals

• Editing SQL and Python cells

• Explaining the logic behind cells

Generating analyses from a text prompt

Count AI can generate whole analyses, including cells, visuals, controls and text, from a simple text prompt. Click on the AI icon in the control bar at the bottom of the screen and describe the analysis or outputs you want to create. Count AI will then automatically add objects to the canvas.

Click on the AI button in the canvas control bar

Describe what you'd like to know or create

Count AI knows about other cells in the canvas, as well as tables in your database — including all column names and types. However it does not have access to row-level data, either in cells or database tables.

Generating analyses from a canvas prompt

Count AI can also generate analyses from sketches or instructions in the canvas itself. Start by selecting the objects you wish to use as the prompt for the AI, and click on AI -> Generate in the floating controls. New objects will be generated in a similar way to above.

Click on AI -> Generate.

You can use this to sketch out multi-stage analyses, draw schematics of visuals, and lay out entire dashboards.

Generating a multi-stage analysis from a flow diagram

Generating a dashboard from a schematic

When generating outputs from canvas objects, table and visual outputs are removed before being sent to the AI model.

Editing visuals

You can use Count AI to edit a visual by selecting it and clicking on AI -> Edit in the floating controls, or clicking on the AI button in the canvas control bar at the bottom of the screen when the visual is selected.

Explaining

Count AI is able to change many aspects of visuals, including the query, the type of chart, colors, labels, axes names, ranges and types, orientation, subplots, etc.

Describe how you'd like to change the visual and save yourself many UI clicks.

Editing SQL and Python cells

Editing SQL and Python is also possible with Count AI. As above, select the cell, click on AI -> Edit in the floating controls and describe what you'd like to change.

Click on AI -> Edit

Describe how you'd like to change the query.

Explaining the logic behind cells

In complex canvases, or when viewing someone else's work, Count AI can help you quickly understand the logic behind a cell. Simply select the cell, click on AI -> Explain, and a sticky will be added to the canvas that provides a summary of the analysis, as well as a cell-by-cell explanation of each dependency. Clicking on any of the cell names will take you to that particular cell, so you can quickly interrogate dependencies.

Click on AI -> Explain.

Count AI gives a cell-by-cell explanation of the logic behind any SQL, Python or visual cell.

Tips

• Provide as much context as possible. The more specific you are, the more likely it is that the AI will return what you're looking for.

• When generating analyses, try to name the cells you wish to reference.

• Explain the definition of any metrics you wish to be created — for example, "Show me how margin (Profit / Sales) varies by customer".

• For complex analysis, try to break down the analysis into distinct steps. These will become cells in the generated output. If they reference one another, give each a name "call this cell..." and state which other cell it references — this can be done in words or, when generating from a canvas schematic, arrows or connectors.

Queries

Count AI may create or edit cells or visuals that are attached to a database connection. However, unlike other cells, these will not execute without your approval, so you can verify any queries before they are sent to your database. Local cells and visuals (which are executed by DuckDB in the browser), as well as Python cells, will be executed automatically as normal.

Privacy

• Count AI is fully GDPR compliant.

• Neither Count or our model partners (OpenAI, Google) train models using data from Count customers.

• Admins can disable AI features from the workspace settings if necessary.

• Count AI is secure, built on top of our SOC 2 accredited platform.

Learn more in our Terms of Use and Security pages.

Accuracy

Count AI is powered by artificial intelligence algorithms, whose accuracy and reliability may vary. Outputs produced by Count AI may be misleading or include inaccuracies, and should not be regarded as fact or used as a substitute for expert judgement. It's essential to verify the information provided by Count AI before making any decisions or taking any action.

Learn more in our Terms of Use.