Query Mode Selection

ADM offers several ways to work, depending on your goal: quick answers, deeper investigation, repeatable tasks, or work with your team. When you choose the right mode, you get accurate results faster and can reuse them more easily.

This guide helps you:

• Decide which mode to use for data questions vs business questions
• Write prompts that produce consistent results
• Understand when to bring in context (Knowledge Base, MCP sources, people)

How to access these features

• Conversation: In the left navigation, select Agentic Data Management > Home, and then type your question in the chat box. To start another conversation, click New Conversation.
• Workflows: Select Agentic Data Management > Workflows.
• Business Notebooks: Select Agentic Data Management > Business Notebooks.
• Knowledge Base: Select Agentic Data Management > Knowledge Base to upload documents. In a conversation, use Knowledge base in the chat box to choose which files ADM searches.
• Multi-User Collaboration: Open a conversation, and then click Share with users.
• Agents: To learn which agents are available and when ADM uses them, see Understanding Agents.
• MCP Servers: Navigate to Control Center > Integrations > MCP Servers.  For more information, see Understanding MCP Server.

Choose the right mode

Use Conversation when you want fast answers or iterative investigation

Best for:

• Asking “what is happening?” and “why did this happen?”
• Exploring root cause hypotheses
• Getting suggestions for next steps
• Summarizing, translating, or explaining results to stakeholders

Not ideal for:

• Repeatable, standardized execution (use Workflows)
• Producing a structured artifact every time (use Business Notebooks or Workflows)

Examples

• “Why did the customer freshness alert trigger today?”
• “Summarize the top issues impacting revenue metrics this week.”

Use Workflows when you want repeatability and consistency

Best for:

• Standard operating procedures (SOPs)
• Running the same investigation steps across assets/teams
• Turning analysis into an operational process

Not ideal for:

• Exploratory “brainstorming” without a clear endpoint

Examples

• “Every morning: check freshness + reconciliation + quality score for Tier-1 tables and summarize failures.”
• “Run a standard triage workflow whenever an incident is created.”

Use Business Notebooks when you want a shareable narrative + structured outputs

Best for:

• Executive-ready summaries
• Consistent reporting format (tables/sections)
• Combining business context with technical detail

Not ideal for:

• Simple one-off Q&A

Examples

• “Create a weekly reliability briefing for Finance domain: top failing assets, trends, recommended actions.”

Use Multi-User Collaboration when the work requires multiple roles

Best for:

• Incident response across data/platform/business stakeholders
• Shared investigation context and decisions
• Keeping a living record of “what we tried and why”

Examples

• “Bring in the data engineer + business analyst to validate whether this is expected seasonal drift or a pipeline defect.”

Best practices for data and business questions

Data / Technical queries (Engineers, Platform teams)

Use Conversation for exploration, then Workflows for operationalization.

Good fits

• “Identify which upstream dependencies changed before the schema drift alert.”
• “Show which assets are failing policies and whether failures share a pattern.”

Recommended pattern

• Conversation: explore and confirm the hypothesis
• Workflow: standardize the triage steps
• Notebook: publish a stable summary for stakeholders

Business / Stakeholder queries (Analysts, Data Stewards, Leadership)

Use Business Notebooks or Conversation with clear context and constraints.

Good fits

• “Which customer metrics are impacted by delayed data today?”
• “Summarize reliability risks for Finance domain in plain language.”

Recommended pattern

• Provide business context (which metric/process matters)
• Ask for a stakeholder-ready output (bullets/table)
• Request citations (Knowledge Base, sources) where applicable

Prompting patterns that work (reusable templates)

Template A — Investigation prompt (conversation)

Context + Question + Constraints + Expected output

• Context: asset / domain / incident link
• Question: what happened / why / what changed
• Constraints: timeframe, environment, severity
• Output: bullets, table, short summary + next actions

Example “For the customer_360_view asset, investigate why quality score dropped in the last 24 hours. Compare today vs yesterday, identify the top failing rules, and suggest the most likely root causes. Output: (1) 5-bullet summary, (2) table of failing rules and impact, (3) recommended next steps.”

Template B — Business translation prompt

“Explain this issue for [audience] focusing on [business impact], using [constraints], and include [links/citations].”

Example “Explain today’s freshness delay for Finance leadership: what business reports are impacted, estimated risk, and what teams are doing. Keep it under 10 bullets, and include links to the affected assets and incidents.”

Template C — Workflow design prompt

“Create a repeatable workflow that takes [input] and produces [output], including decision points.”

Example “Design a workflow: given an incident on a Tier-1 table, run triage steps (freshness, schema drift, policy failures), produce a summary, and assign recommended owner teams based on domain tags.”

Common mistakes and how to fix them

Mistake: Asking without scope

Bad: “Why did it fail?” Better: “Why did customer_validation fail today at 2 PM? Show the failure pattern and what changed upstream.”

Mistake: Mixing business and technical asks without structure

Better: break into two steps:

• Technical diagnosis
• Business impact summary

Mistake: Not specifying output format

Ask for exactly what you need:

• Table with columns
• Summary length
• Include links / citations
• Include “what to do next”

Quick decision guide

Use this when you’re not sure:

• Need a fast answer now? → Conversation
• Need consistency / repeatability? → Workflows
• Need a polished narrative for sharing? → Business Notebooks
• Need multiple roles to decide/act? → Multi-User Collaboration
• Need company-specific grounding? → Knowledge Base (and request citations)
• Need ADM to reuse a key finding in later conversations? → Click Remember for this workspace under the answer to save it to Workspace Memory
• Need external systems context? → MCP (if configured)

Examples: “data” vs “business” version of the same request

Data version

“Identify which rules failed on orders today, how many rows were affected, and whether failures correlate with a pipeline run or schema change.”

Business version

“Summarize today’s orders reliability issue: impact on dashboards, customer experience risk, and expected resolution timeline. Keep it non-technical.”