Analysis plans

An analysis plan is a written research proposal: the business question, the data the agent will use, the steps it will take, and the shape of the output. You describe the business challenge, the agent proposes the analytical approach, and you refine the scope together before anything runs. The result is a scoped, repeatable research process your stakeholders can trust.

The agent answers one-off questions directly in chat, and that is the best way to start out. A plan is for when the work needs more structure. Consider one when you have:

Situation	What it looks like
Complex analysis	The question needs several data sources, Dimensions, segments or analytical steps.
A repeatable deliverable	The output has to be rerun on a cadence, such as a weekly report or a monthly digest.
Stakeholder alignment	You want to agree on scope, data sources and expected outputs before the work runs.

For how plans fit next to exploratory questions, see Quick answers and analysis plans.

Planning is a capability, not only a workflow

You use the agent’s planning skills whenever you ask it to suggest an approach, improve a scope or say what else to include. Treat it as a partner with research expertise, and ask for feedback and guidance on what to do with the data you have:

• What would you suggest?
• How would you approach this question?
• Is there anything else I should include?

Asking for guidance on execution, optimization and troubleshooting works especially well, because the agent’s recommendations are grounded in the full context of your data.

Why write a formal plan

A formal plan defines a starting point for current and future research and fixes it in writing. You can get there two ways:

• Start with a plan. Some work is best scoped up front, and the agent is good at it.
• Build the plan at the end. Other work starts as a looser conversation. Once you have arrived at an interesting set of findings and want to repeat the process, ask the agent to turn the chat into a plan.

Tip

You can always generate a plan after the fact. If you develop a useful report through conversation, ask the agent to convert that chat into a reusable plan.

next.dimensionlabs.ioProduct footage

A plan run end to end: report and deck. Recorded in the product.

Before you start

Build the plan progressively. Start by understanding the available data, then narrow the scope, validate the right fields, and ask the agent to add to or revise the plan as you go.

Not every step is required, and some overlap. Use the sequence below to shape your approach when you are new to planning, or when the analysis needs a strong handoff to another team.

1. Identify available data
2. Get to know the data
3. Introduce the business question
4. Add guidance on actionability
5. Clarify the scope
6. Optimize the plan
7. Add stakeholder implications
8. Prepare for handoff
9. Run the plan

Step 1: Identify available data

Ask the agent which Dimensions, prompts and tables are available.

Example messages:

• Show me the Dimensions in the data.
• Show me the tables in the data.
• What metadata is available?

Your action: find out which Dimensions are available.

Why it matters: the main constraint on any analysis is the set of tables and columns (also called fields) in your data. If the field you need does not exist yet, you can add it. See Design custom Dimensions.

Key terms

Your data has three levels:

1. Tables. One per prompt, plus transcripts, structured_metadata and one per business data file.
2. Columns (also called fields). The columns in a prompt’s table are Dimensions, which the model fills in. The columns in structured_metadata are metadata you uploaded with your conversations.
3. Values, such as very_satisfied or 5.

Column names in these examples, such as CSAT_results or predicted_csat, are examples. Ask the agent for the names in your workspace.

What the agent can see lists the tables every workspace has and what one row of each means.

Step 2: Get to know the data

If you see options you don’t recognize, ask the agent for guidance. For example, if several metadata fields look related to CSAT or satisfaction, ask which one is best to use.

The structured_metadata table is often the best place to inspect the full set of metadata that comes in alongside the conversation text. Ask the agent to show you those fields directly, or investigate in the chat:

• Show me the Dimensions in the structured_metadata table.
• Show me here in the chat.
• What is the right CSAT Dimension to use?
• Do all of these Dimensions have data?
• Identify the most relevant fields.

Your action: test which Dimensions are reliable enough to support the analysis.

Why it matters: a Dimension existing is not proof that it is useful. Check that it is filled in, credible and relevant enough to support the plan. Ask as many follow-up questions as you need.

Step 3: Introduce the business question

Give the agent the business question you want to answer, and ask for a plan.

Example messages:

• I want to know: how satisfied are users with their experience with our bots, and what can we do to improve this further? Create an analysis plan to investigate.
• How would you go about answering this question: what knowledge gaps do interactions with customers highlight, and how can we fill them?
• Generate an analysis plan to address the question: what pain points are users’ conversations highlighting about our brands, or about their experience with them?

Ask for a file

Ask the agent to output the plan as a Markdown file, in the same message:

Output as a .md file named XXXX_analysis_plan.md.

Without that instruction the agent may write the plan straight into the chat. That works, but the point of this sequence is a deliverable you can reuse. If you forget, ask on the next turn: “Save the plan as a .md file.”

Your action: direct the agent to write a research proposal.

Why it matters: this is where the agent starts turning raw data into knowledge. You are asking it to review the available data, decide how to answer the question, and propose a scoped analytical approach.

For best results, brief it like a research agency

Treat this step as the kickoff for a bespoke research project. If you had hired a market research agency, you would not hand them only a question. You would explain the business context, who the work is for, and what a useful answer needs to make possible. Give the agent the same context:

• the project objective
• the value of the research
• who will use the findings, and which decisions the work should support

That context makes the plan more than analytically sound. It makes the plan fit the way the research will actually be used.

An example that combines the question, the primary measure, the audience and the file name:

Generate an analysis plan to address the question: how satisfied are users with their support experience, and what can we do to improve? Use predicted_csat as the primary lens. I am looking for insights to share with the leader of customer support for North America. Output as a .md file named support_satisfaction_analysis_plan.md.

Once the plan is generated, it opens as a file you can view alongside the chat while you keep revising and building on it.

Step 4: Add guidance on actionability

Tell the agent which Dimensions or metrics should be tied to action.

Example messages:

• Incorporate the enriched Dimensions for effort and predicted rating. Use these as a primary signal for identifying areas of user friction with the bot. (Check the scale of each score in your workspace first: ask “What range does each score use, and which end is best?”)
• Use the custom Dimension predicted_csat to run a root cause analysis of the drivers of customer dissatisfaction.
• Identify action items based on which topics are linked with product_dissatisfaction versus agent_dissatisfaction.

Your action: align the agent on a key metric or level of action, where one applies.

Why it matters: Dimension Labs is built to produce actionable findings, but you define what counts as actionable: which signals indicate a problem, which outcomes matter, and which Dimensions count as primary evidence.

Choose an outcome metric

Use this step to pick candidate outcome metrics. A predicted score, such as rating or effort, usually covers far more conversations than sparse metadata fields such as survey answers.

• Standard scores. The standard Dimensions (called System Prompts in the app) include predicted scores, stored in the system_session_prompt table. They run every day only when Dimension Labs has turned on daily enrichment for your workspace. If the table has no data, run a backfill from the Backfill tab. See Standard Dimensions.
• Custom scores. Create your own predicted score when you need a metric tailored to your business. See Predicted scores.
• Revenue outcomes. Tie actions to outcomes such as MRR, LTV, purchase amount, repeat purchase or other revenue measures from your business data. See Join business data.

Step 5: Clarify the scope

If needed, point the agent at specific prompts, tables or Dimensions, or tell it what to leave out.

Example messages:

• Focus on the fields in structured_metadata, like intents, answers, fallback and don’t-know responses. I don’t remember the exact names, but you’ll find them in the metadata.
• Include plan tier, region and channel in the analysis.
• Do not use any of the custom_fields labeled 2024.

Then ask the agent to check the field selection:

Confirm these are the right Dimensions for answering the question. Are there any others we should consider?

Your action: stop the plan from being built on an incomplete or weak set of fields.

Why it matters: the plan should include every Dimension relevant to the question. When your metadata is extensive and the opening question is deliberately open, expect to go back and forth on which fields to include.

The editing loop

This is the core loop for building a plan. Keep the plan file open beside the chat (open it from the File Manager in the message box), read it as a document, and keep refining it with the agent.

When you want a change, tell the agent to update the file itself, not only to reply in chat:

Revise the plan file XXXX_analysis_plan.md.

That keeps the reusable plan current as your thinking changes.

Step 6: Optimize the plan

Work with the agent to widen and deepen the proposal.

Example messages:

• Is there anything else we need to add to this analysis plan to make the insights more actionable or more specific to the business question?
• How would you further optimize this plan?

Your action: let the agent suggest improvements.

Why it matters: open-ended review questions are one of the best ways to improve a plan. The agent may find missing segments, better outcome metrics, stronger comparisons or a clearer structure for the deliverable.

Step 7: Add stakeholder implications

Ask the agent to go beyond the analysis and answer “so what?” for the people who will read it.

Example message:

Include a section with clearly defined, stakeholder-ready implications and action items for the business.

Your action: ask for a plan that produces stakeholder-ready findings.

Why it matters: the plan can then produce a report tuned to the people who use the findings, such as product managers, department leads or executive leadership, and point them toward concrete actions.

Step 8: Prepare for handoff

Ask whether operational details should be added so the plan runs smoothly.

Example message:

This plan will be executed in a separate chat. Add any operational details at the start needed to ensure a smooth handoff to the agent.

Your action: make sure the document holds all the context it needs.

Why it matters: a plan has to make sense as a standalone document, read outside the conversation that drafted it. Ask the agent to suggest anything else worth including.

Tip

Start a new chat at key points in a process. A clean chat keeps planning separate from execution, and the agent running the plan starts from the plan alone instead of everything you tried on the way.

Step 9: Run the plan

Start a new chat in the same project and give it the plan. Every chat in the project can read the project’s files. If you can’t see the plan in the File Manager, ask the agent in the planning chat to save it to the project, or download it and attach it to the new chat.

Example message:

Execute the plan file XXXX_analysis_plan.md.

Your action: kick off the analysis.

Why it matters: a clean handoff from planning to execution makes the analysis easier to manage as its own piece of work.

Tip

Give the plan file a name that says what it is for, such as support_satisfaction_analysis_plan.md, so you can find it again in the File Manager.

Finished deliverables are saved in the project’s Final Reports folder. To share one, publish it to Artifacts. See Reports, decks and sharing.

Run it again

Next month, reuse the same plan on new data:

Update the plan for October and run it.

Same plan, new data, results you can compare from one run to the next. You don’t edit the file by hand: tell the agent what to change (“Change this plan to cover May”, “Remove the section on email”) and it rewrites the file in place.

To run a plan on a schedule, see Run it on a schedule.

Related

• Quick answers and analysis plans: when to explore and when to build a repeatable deliverable
• Context engineering: the plan file and the other files that make repeat work consistent
• Projects and standing instructions: where plans live, and the AGENTS.md file every chat reads
• Ask good questions