Skip to main content

Moby Deep Dive

Create Plan-Driven Analytics Reports with Moby Chat

R
Written by Rebecca Maybaum
Updated today

Overview

Deep Dive is Moby’s almost fully-autonomous analytics mode. Instead of replying one message at a time, Deep Dive acts as an agent: it designs a plan, executes every step, auto-debugs issues, pivots when it finds new angles, and delivers a polished, share-ready report—text, charts, Gen UI, and code—all in one go.

Benefit

Why it matters

Plan‑first workflow

Review the entire plan up front. Add, remove, or reorder tasks before anything runs.

End‑to‑end execution

Deep Dive executes the whole dependency graph, running independent steps in parallel and chaining dependent ones automatically.

Smart pivots & auto‑debugging

It detects errors or dead ends, fixes them on the fly, and explores new angles it discovers.

Shareable & schedulable results

Copy the report, share a link, or save the plan as a scheduled Agent to have it re‑run automatically.

Deep Dive vs. Build Mode

Deep Dive

Build Mode

Experience

Hands‑off & autonomous. Provide a target, wait a short time, receive a comprehensive report.

Manual & conversational. Ask one question at a time, debug as needed, request reports yourself.

Execution model

Dependency‑graph execution with parallel tasks and auto‑debugging and smart pivots.

One SQL/chat step at a time.

Extra capabilities

Router Chat → Steps Planner → Execution Engine → Final Report pipeline.

Can generate Gen UI when you explicitly ask.

Ideal for…

In‑depth investigations, strategic deep dives, recurring insights.

Quick explorations, ad‑hoc SQL questions, iterative widget building.

Key Components

  1. Router Chat – gathers context to define the analysis objective.

  2. Steps Planner – translates the objective into an ordered list of tasks (data pulls, analyses, visualizations).

  3. Execution Engine – runs each task, auto‑debugs, and pivots when it finds richer insights.

  4. Final Report – delivers Gen UI, Gen Text, Code, and Reasoning layers in a single view.


How It Works

  1. Ask a question – type your own prompt or choose from the Deep Dive Prompts list.

  2. Deep Dive drafts a plan – a numbered list of steps (8–15 is typical) that cover data pulls, analyses, and visual outputs.

  3. Review or edit – keep, remove, or tweak any step in the plan.

  4. Run the plan – each step queries data or performs an analysis (in parallel when possible).

  5. Results appear – Gen Text first, then Gen UI, alongside Code & Reasoning tabs.

  6. Iterate – adjust the question or plan; only the changed steps rerun.

  7. Share or schedule – copy the share link, or convert the plan to an Agent you can schedule to run automatically.

What You’ll See

Left Pane

Right Pane

Prompt & Plan – your question, editable step list, and chat.

Output – Gen Text, Gen UI, Reasoning, and Code

Behind the Scenes

Deep Dive builds a dependency graph for the plan. Steps that depend on each other run in order, while independent steps run in parallel to accelerate the overall run time.

Once every data‑analysis step finishes, Deep Dive produces two synchronized report layers:

  1. Gen Text – the narrative summary with tables, insights, and recommendations.

  2. Gen UI – an interactive visual report generated immediately after the text layer is complete.

Use the Reasoning tab to inspect step‑by‑step logs, view raw queries, and trace how each insight was produced. After the run you can keep asking questions (e.g., “How did you calculate ROAS in step 7?”) and Deep Dive will reference the finished results.


Start Using Deep Dive

Open Deep Dive

  1. Navigate to Moby from the left navigation panel.

  2. Click Deep Dive just below the chat input field.

Write or Pick a Prompt

Option

Details

Choose a template

Select one of the starter prompts from the Deep Dive Prompts below the chat input field.

Write your own

Type a plain‑language prompt describing the analysis goal and desired outputs.

Example Prompts Deep Dive Handles Well

  • “Which attribution model should I use to analyze my ads?”

  • “Why is my ROAS and blended ROAS down?”

  • “Which Meta campaigns should I scale?”

  • “What are my best opportunities to increase Average Order Value (AOV)?”

Full Prompt Example

Which creatives are performing best on Meta?
1. Break down performance by Pixel CTR, Pixel Conversion Rate, Pixel ROAS, Engagement.
2. Compare image, video, carousel, and UGC formats.
3. Evaluate audience‑specific performance.
4. Identify creative fatigue trends.
5. Recommend creative iterations.

Run the prompt

It can take Moby 1-2 minutes to run the initial prompt and generate the Deep Dive plan.

Review the plan

You can run all steps, pick a subset, or edit the list before running.

Example plan

Step details

For each step you may see:

  • Step Type - Warehouse, Vision, Analyze Data, Preload Dashboard Data, Forecasting, Asking the MMM, Final Report

  • Status - Pending, In Progress, Completed

  • Step Dependencies - Steps whose outputs feed this step (those steps must be completed before this step can begin).

  • Request - The full prompt text for that step

  • Dashboard Data - Dashboard Name, ID, Date Range, and Period to Compare

  • Output type - Text, Data Visualization

Best-practices for an effective plan

Tell Deep Dive what to change in plain language (e.g., “drop the creative analysis step and combine the two dashboards”).

Why

Tips

Keep it focused

Extra steps slow things down.

Remove anything that doesn’t answer your question.

Watch the count

8–15 steps usually cover most needs.

If you see 20+, combine or drop steps.

Cover all data

Missing a key metric, channel, or time window skews results.

Add retrieval steps for Email, Amazon, etc.

Limit costly ops

Vision analysis & large forecasts are slower.

Include them only when they add clear value.

Run the plan

Once you are happy with the plan, tell Moby in plain language to execute it (e.g. "run the plan").

Explore the results

After the plan runs, Deep Dive returns four synced layers you can click through in the right-hand pane:

Layer

What you see

GenUI

Beautifully rendered UI report including interactive charts, tabs, insights, and recommendations.

GenText

Detailed text report including tables, insights, and recommendations based on the Deep Dive analysis.

Code

The full code that renders the GenUI report.

Reasoning

Detailed reasoning behind each step of the Deep Dive plan, including the data behind every part of the analysis.

GenUI

GenText

Code

Reasoning

Iterate and refine

  • Logic tweaks — Ask follow‑up questions or change the prompt. Deep Dive reruns only the steps that changed, so iterations are efficient.

  • UI tweaks — Open Edit UI from the three dots menu in the top right. Use natural language to prompt UI changes in the GenUI report.

    • Preview desktop and mobile views before saving.

    • Save Design as Default to reuse the style in future Deep Dive runs.

    • Set Brand Guidelines once and they’ll apply to every GenUI report your team creates.

Share or Convert to Agent

Once your Deep Dive report looks the way you want, you have two options for making it available to others:

Share a view-only link

  1. Click Share in the upper-right corner of the report.

  2. Configure any extras you need:

    • Restrict to specific email addresses

    • Require a password for access

  3. Click Generate Link and drop it in Slack, email, or any doc.

Convert to an Agent

Note: You'll need access to Agents to use this feature. View plans

Saving your Deep Dive plan as an Agent lets you:

  • Automate recurrences – schedule the analysis to run (e.g., every Monday at 9 AM).

  • Add destinations – after conversion, open the Agent builder to add delivery steps such as Email, Slack, or Google Sheets (Deep Dive does not assign destinations automatically).

  • Build on it - Connect your Deep Dive output with other Agents to create even more powerful mega-Agents.

  1. Click Create Agent in the upper-right corner of the report.

  2. Deep Dive turns every step in the plan into an Agent workflow.

  3. In the Agent builder, you can finalize the Agent configuration, including:

    • Edit or reorder steps

    • Add destinations (Email, Slack, Feed)

    • Set a schedule (e.g., every Monday at 9 AM)

The Agent now lives in your shop's Agent collection.

Did this answer your question?