| Audience | Platform admins, data teams, marketing teams, and creative teams |
| Prerequisites | The Configure Agents grant, and at least one agent product enabled |
You don't have to document your whole business before agents are useful. Set up one workflow at a time, and let real usage tell you what to add next.
Overview
Start with one workflow. Configure only the information that affects that workflow, then run it with an agent. Review the result, and add more context only where the agent is missing information or applying the wrong business rule.
Repeat this as you take on more workflows. Context Hub grows alongside your use of agents, instead of becoming a project to finish before anyone can start.
The rest of this page walks that loop:
- Start with one workflow
- Configure only what that workflow needs
- Run the workflow and check the result
- Fill the gaps the agent reveals
- Put each kind of information in the right place
For how to configure each surface in detail, see Teach agents your business and Set content and creative rules.
Start with one workflow
Choose a recurring task where "correct" is easy to describe and valuable to the team. For example:
- A weekly performance report
- A campaign or channel comparison
- Drafting email, SMS, or push copy
- Generating on-brand ad creative
Before you configure anything, write down:
- The question the agent should answer, or the content it should produce
- Who will use the result
- The data or brand assets involved
- What a good answer or a usable draft looks like
- A trusted report, dashboard, or example you can check the result against
A known-good reference is what lets you judge whether your context is working.
Configure only what that workflow needs
Configure only the information the agent needs to complete this workflow correctly.
For an analysis workflow, start with the business rules and definitions that affect the answer. Add them in Context Hub → Context by selecting Add context — see Add business context for step-by-step instructions.
| Start with | Why it matters |
|---|---|
| The three to five metrics that matter | Focuses the agent on the numbers this workflow is about |
| A definition and formula for each | Keeps the agent from confusing similar metrics or inventing a formula |
| Your source-of-truth rules | Keeps warehouse and platform numbers from being mixed silently |
| Inclusion and exclusion rules | Excludes test, internal, refunded, or suppressed records |
| Fiscal calendar, timezone, and relative dates | Makes "last month," "YTD," and "prior period" unambiguous |
For a content workflow, start with the brand and channel rules that shape the first deliverable — voice, required disclaimers, approved claims, and message limits. You configure these in Set content and creative rules. Don't build a full brand library before you've generated anything.
Set each item's Agent scope so it reaches the right product — Agents, Ad Studio, or Lifecycle Studio — or leave it unscoped to apply everywhere.
Run the workflow and check the result
Run the real question, report, or content task, then compare the result against the trusted reference you chose.
Before you decide what to fix, ask the agent what it worked from:
For this result, tell me which data, context, and assumptions you used, and where you were missing information.
The response helps you identify where the agent relied on assumptions or lacked information. Those are good candidates for additional context.
Fill the gaps the agent reveals
Add context when a real workflow shows the agent is missing information or applying the wrong business rule. This keeps Context Hub focused on information that improves results.
| When the result shows… | Add… |
|---|---|
| Channels compared on mismatched numbers | Attribution model, channel taxonomy, and source-of-truth rules |
| A metric moved but no explanation | Benchmarks, change thresholds, and the dimensions to break down by |
| Lifecycle stages the agent misread | Stage definitions, re-entry rules, and conversion definitions |
| Off-brand or non-compliant content | Brand profiles, styles, copy examples, and guardrails |
| The same correction, over and over | A confirmed memory |
Run the workflow again after each change. Once the result is reliable, move on to the next workflow and repeat.
Put each kind of information in the right place
Not everything belongs in Context Hub. When something's wrong, the symptom tells you which layer to fix:
| The problem | Fix it here |
|---|---|
| The agent can't see a table, field, or model | Sources, Models, the schema, or access settings |
| The table grain, join, or field meaning is unclear | Schema or model metadata |
| The agent has the data but uses the wrong metric, join, or source | A business context item |
| A rule applies only to one recurring report | The report's own instructions, in Agents |
| A rule matters only for today's analysis | The prompt, a chat, or an uploaded file |
| A confirmed learning should be reused | A memory |
| Generated content uses the wrong brand rules | Brands, styles, or guardrails |
The short version: data structure belongs in the data layer, business meaning belongs in Context Hub, confirmed learnings become memories, and task-specific instructions stay with the task.
Keep context current
Assign an owner for your business rules, definitions, and brand context, even when several people contribute. Keep each item short, and give it a description that says when it applies — the agent uses that description to decide what's relevant.
Review your definitions after major changes to tracking, models, attribution, products, or legal requirements. Treat memories as knowledge you maintain, not a permanent record. When two sources disagree, have the owner decide which one wins, or run a context review to surface the conflict.
Next steps
- Teach agents your business — add business context, skills, memories, and signals.
- Set content and creative rules — brands, styles, copy examples, and guardrails.