Planning with the Planner Agent

Build and compare financial scenarios in plain language. Describe the change you want to model, and the Planner does the math, shows you the impact, and saves the result as a governed Cube scenario.

This guide covers how to turn the Planner on, how a planning session works, what each part of the dashboard shows you, and what to expect while the feature is in open beta.

Turning on the Planner

Access to the Planner is controlled by a single permission. An Admin grants it, and once granted, the user sees the Planner in their Workspace.

The permission an Admin grants

You will find Use AI Planner in the Cube AI permission group in User Management. There is no separate setup step inside the Planner itself. Once a user has the permission, the Planner is available to them in the Workspace.

Starting a plan

1. Open the Planner from the Workspace.
2. Click + Plan in the top right to begin a new planning session.
3. Tell the Planner what you want to model. You can anchor to an existing scenario as your starting point, for example "Start from our 2026 Budget," and then describe the changes you want to explore.

You do not need to write a perfect prompt. Describe the change the way you would explain it to a colleague:

• "Create a new forecast scenario that increases H2 Platform Revenue by 15% for GCo US, using the Budget as a baseline."
• "Create a new scenario based on the current forecast and add 5 new hires across Engineering, Sales, Product, and CS starting in July."
• "Duplicate our 2026 Budget as a new forecast, then increase Revenue and COGS by 10% in H2."
• "Model an aggressive case with S&M spend up 25% to 35% through Q4, and a moderate case at 10%."

How a session works

The Planner uses two panels. The conversation is on the left, and the dashboard is on the right. As you describe changes in the conversation, the Planner analyzes your Cube data, builds the scenario, and updates the dashboard to show the result.

The Planner will usually come back with its analysis and a clarifying question or two before it builds anything, for example asking you to confirm a growth assumption. Answer those, and it goes ahead and saves the scenario.

You can keep refining in the same session. Ask for a different number, a new comparison, or an additional scenario, and the Planner updates the model and the dashboard. You can also ask the Planner to compare more than one scenario against your baseline within a single plan.

The four dashboard tabs

The dashboard on the right is organized into four tabs. Each one answers a different question about your plan.

Overview

A summary of your plan, with the headline numbers, the assumptions behind them, and the Planner's recommendations.

• Key Metrics are comparison cards, for example Total Revenue, S&M Spend, Headcount, and Cost of Goods Sold. Each card shows the value alongside its variance against the scenarios you are comparing and against your baseline.
• Key Assumptions lists the inputs driving the model. Each assumption is tagged to show where it came from: User means you supplied it, and AI means the Planner inferred it from your Cube data. This tells a reviewer at a glance which numbers to double check.
• Recommendations are observations the Planner surfaces about your plan, each with a type: Risk, Opportunity, Data gap, or Info.

Analysis

The underlying Cube data the Planner pulled to support its analysis and the changes it made. A dropdown lets you switch between the different datasets the Planner referenced, each shown as a spreadsheet grid. This tab is where you trace a number back to the data behind it.

You can export any of these grids to Excel using the export action at the top of the tab.

Updates

Every value the agent changes, compared to the baseline. The rows are the account members the plan touched. The columns are your baseline scenario, each new scenario the Planner created, and a variance column for each new scenario. Negative variances appear in red so you can spot reductions quickly.

Open the same view in Builder

When you want to keep digging after the session, you do not have to rebuild this view by hand. Use the action in the top right of the Updates tab to open Builder in a new browser tab, already configured to match the Updates table.

What carries over:

• Rows: the same account members, in the same order.
• Columns: your baseline scenario, each new scenario from the session, and the variance columns.
• Filters: the same dimension filters the Updates table is resolved against, including Time, Department, Entity, Product, Market, and Currency.

Three things to know before you use it:

1. It opens as an unsaved range. Save it in Builder if you want to keep it.
2. Builder fetches live data when it opens, so the values may differ from the snapshot the Planner showed you if the underlying Cube data has changed since.
3. The handoff is one-way and copies configuration only. Edits, saves, and publishes you make in Builder never write back into the planning session, and re-running the Planner session will not change a Builder range you already have open.

The action only appears if you have access to Builder. If you do not see it, Builder is not in your permission set.

Impact

Charts and dashboards showing how your plan moves the numbers, for example, monthly spend by scenario or revenue trend lines. The Impact tab can contain its own set of sub-views that the Planner builds to fit your plan, so the charts you see will vary from one plan to the next.

Where your scenarios are saved

Scenarios the Planner creates are saved into the standard Cube Scenarios dimension, under a parent called Planner Scenarios. They are organized by user, then by session, then by the named plan, so you can always find which session produced a given scenario.

Because Planner scenarios are standard Cube scenarios, you manage them in the usual Scenarios area. From there, you can rename a scenario, change its parent, set its rollup type, or edit its description. Each Planner scenario is created with a description noting it was generated by the AI planning assistant.

Every Planner write is attributed

Changes the Planner commits appear in your audit trail with a prefix, so they are distinguishable from a manual publish at a glance:

• Planner writes are prefixed Published with Cube Planner:
• Writes that come from an external AI tool through the Cube MCP Server are prefixed Published via MCP (ClientName):
• Publishes you make yourself from Excel, Google Sheets, or the Web Workspace are unchanged and carry no prefix.

How the Planner handles formula accounts

A Planner scenario stores values, not living formulas or driver tables. So a Planner scenario is a snapshot of the numbers rather than a self-recalculating model.

Formula accounts are the exception worth understanding. When the Planner needs to change an account that is driven by a formula, it does not try to overwrite the formula. Instead, it reads the formula definition, follows the references to the accounts that actually drive it, and publishes there. That means:

• It follows drivers wherever they live in your account dimension, including branches outside the section you asked about. For example, a Salary account under your P&L driven by accounts under an HR Metrics parent.
• It works through nested formula accounts until it reaches accounts it can write to.
• It publishes at the correct dimensional and time intersection, without you having to ask for it.

If the Planner successfully publishes to the drivers, the formula account is not flagged. You will only see a Data gap recommendation when a driver genuinely cannot be published, and the recommendation names the reason.

Reviewing before you rely on it

The Planner uses AI to interpret your request, pull data, and summarize the result. As with any AI output, review the numbers before you act on them. Three parts of the product are built for exactly this:

• Use the User and AI tags on the Overview tab to see which assumptions the Planner inferred rather than ones you gave it.
• Use the Analysis tab to trace a headline number back to the Cube data it came from.
• Open the Updates table in Builder to interrogate the same lines against live Cube data, outside the planning session.

Planning from external AI tools

The Planner runs in the Web Workspace. It is not in the Excel or Google Sheets add-ins.

You can also drive publishes into Cube from external AI tools such as Claude and ChatGPT, through the Cube MCP Server. This requires the Use MCP (Write) permission, which an Admin grants separately from Use AI Planner. Writes that arrive this way are attributed in your audit trail, as described above.

See Connect Cube to AI Apps via MCP Server for setup.

Returning to past sessions

Your planning sessions are saved. Click History in the top right to open Your Planner History, a list of your prior sessions with the time each was last active. Select any session to reopen it and continue where you left off.

What's next

A few improvements are on the way:

• The Planner will start showing its individual thinking steps as it works, so you can follow its progress in real time.
• Planner-created formulas and tables, so a plan can produce a recalculating model rather than a set of values.
• Sharing a planning session with other users.
• Interacting directly with the recommendations on the Overview tab.
• Exporting the Overview and Impact tabs.

The Planner uses AI to summarize your analysis by extracting key data points. Review all generated content before use. AI may make mistakes.