To write effective instructions for the Analyst Agent, you need to guide it step by step, using explicit and structured language. Your instructions must specify which of the supported tools and capabilities the Agent is to use.

Essential and optional instructions

Your instructions must contain the following:

• What data to analyze.

• Which analysis steps to follow.

• How the output should be structured.

Optionally:

• Include examples of how the report’s wording should look.

• Specify a tone and style for the report (e.g. concise, executive-friendly, bullet-pointed).

• Define a language for the report.

• Identify a destination Metric for the report.

Instruction optimization

1. Be as explicit as possible:

   • When you know a specific Metric, Dimension or Application Variable is needed, use the Object picker: type @ to open a dropdown of eligible options. This way, even if the object names are updated at a later time, the Mission still runs. You can reference multiple objects in one instruction.

   • Only reference AI-enabled Metrics with the Number data type. Other types of Metrics are not supported at present. Metrics shared from other Applications are eligible. 

   • Directly reference in plain text the Dimension Items you want to filter on.

     Note

     Agents cannot see formulas.

     ​​​​​​

2. To improve clarity, paste your instruction into an LLM such as ChatGPT, preceded by guidance such as: “What do the following instructions for a report mean? Can you reformulate them for better readability?”

Supported tools and capabilities

Choose from the below tools and reference them by name in your instruction.

Contribution analysis

Contribution analyses work well when your Metrics have multiple Dimensions to drill into. A contribution analysis identifies the key drivers in an aggregated value. For example:

Choose which Dimensions are to be used to disaggregate your data by specifying them in the Object picker. If no Dimension is specified, the Agent automatically selects the three Dimensions that best explain the result. You can also ask the Agent to rule out certain undesired Dimensions, such as “currency”.

A powerful capability is the combined breakdown. The Agent looks for pairs of Dimensions (for example, Country–Product) that make key contributions.

Variance analysis

As with contribution analyses, variance analyses also work well when your Metrics have multiple Dimensions. Choose which Dimension is to be used to disaggregate your data by specifying the Dimension in the Object picker. If no Dimension is specified, the Agent automatically selects the three Dimensions that best explain the result.

Mention which two Dimension Items are to be compared, such as “actuals vs forecast”, “this month vs last month”, and specify the comparison Dimension. You don’t need to create a specific Metric that calculates the variance.

Filtering

You can ask the Agent to filter based on absolute and relative variances, using multiple conditions.

The Agent can also filter on pairs of Dimensions, such as “Filter top 10 SKU–Product pairs” or “Show Department–Vendor combinations with variance above ±10K”.

Time Series analysis

Compare data over multiple time periods, instead of between just two time periods or Scenarios.

Loops

The Analyst Agent can provide valuable insights by running analyses across multiple scopes, for example by Business Unit, Department or Cost Center. There is currently a limit of 15 items per iteration.

Charts

Charts generated for Mission output hold dynamic data while the text and figures are static. They are included by default unless you specify otherwise in your instruction. You can also manage charts and delete unwanted charts. The capability of referencing advanced existing Views, such as highly formatted tables and waterfall charts, is still in development.

Advanced calculations

Currently, certain advanced calculations are partially supported, such as:

• Difference between Metrics

• Show results as a %

Others are in development, such as:

• Cumulative operations (e.g. YTD, QTD, ...)

• Advanced ‘Show Value As’

• Ratios with different aggregation levels

If a calculation doesn’t work as expected, it could be due to these current limitations. A workaround would be to create a dedicated Metric for the calculation.

Dos and Don’ts

• Currently, the Agent can only access Metrics with data type Number.

• Snapshots are supported.

• Shared Scenarios are supported, but local Scenarios are not.

• Member-based variables are not supported.

• Avoid referencing pre-calculated variance Metrics. The Agent computes variances itself, and using pre-aggregated values can lead to incorrect results.

• Excluding Items when filtering is supported, though results vary in certain cases.

• You cannot reference Views in the Instructions.