Overview

The Unified Co-marketer provides a conversational experience for campaign, journey, and segment analysis. A custom agent uses the same underlying intelligence for a specific, recurring analysis and follows a defined set of instructions.

A custom agent can be configured with a specific role, selected data domains, and instructions that define what to examine, what not to assume, and how to structure the response. This enables recurring activities such as weekly audits, monthly reviews, and pre-campaign checks to follow a consistent analysis format.

Custom agents are available in Agent Studio, within the full-screen Co-marketer experience.

• 📘

  👍Start With a Ready-Made Agent

  Each panel includes a set of Netcore-built sample agents under Get started with these on the Agent Studio home. These agents can be used as starting points instead of building an agent from scratch.

  • Run an agent to review the type of output produced by a configured agent.

  • Clone and modify an agent to copy its setup and customise the name, description, domains, and instructions for the required business context.

    Cloning a sample agent provides a predefined structure and output format that can be adapted for a specific use case.

Custom Agent Data Access

A custom agent inherits the Unified Co-marketer data boundary. It can read campaigns, journeys, and segments available on the panel where the agent was created. It does not have access to CRM, order management, finance, or other external systems.

During agent creation, the data boundary can be narrowed further by selecting only the domains required for the intended analysis.

Create a Custom Agent

Follow the steps below to create a custom agent for your panel.

1. Log in to the Netcore CE dashboard.
2. Click Ask Co-marketer in the top navigation bar. The Co-marketer opens as a side panel.
3. Click the Expand icon at the top right of the panel to open the full-screen experience.
4. In the left navigation, click Agents.

📘### Note

The Agents section is available only in the full-screen experience. The side panel provides the chat experience only. Expand the panel to access Agent Studio.

The Agent Studio home displays:

• Your agents — Agents already created on the current panel.
• Get started with these — Netcore-built sample agents available as starting points.

Create an Agent

Click Create Agent at the top right of the Agent Studio home.

Existing agents appear as cards displaying the agent name, description, and selected domains.

Configure the Agent

The agent creation form includes the following fields:

❗️### Enable Deep Research for PDF Reports

Deep Research must be enabled for the agent to generate a downloadable PDF report.

Domains and capabilities are unchecked by default. Deep Research can also be enabled later by editing Tools on the agent page.

Select Data Domains

Select only the domains required for the intended analysis. A narrower domain selection can produce a more focused response.

Add Agent Instructions

After the agent is created, the agent-specific page displays Instructions and Tools in the right rail. Tools contains the configured domains, capabilities, and visibility settings.

1. Click the Pencil icon next to Instructions.
2. Enter the instructions in the 5,000-character field.
3. Click Save Instructions.

A confirmation appears after the instructions are saved, and the saved content is displayed in the right rail.

📘### Configure Instructions

An agent without custom instructions can still respond, but the response remains general. The Instructions field defines the agent's role, analysis requirements, boundaries, and response format.

Most effective configurations use approximately 2,000–2,600 characters. The full 5,000-character limit is not required for most use cases.

Run a Custom Agent

A custom agent can be run from its own page or from a new chat.

From the Agent Page

The chat box on the agent home has the agent already attached. Enter the required prompt and send it.

From a New Chat

1. Click New Chat.
2. Click the + icon to the left of the chat box.
3. Select Custom Agents.
4. Select the required agent from the list.

The + menu also provides Deep Research as a per-chat option.

Write Effective Agent Instructions

Well-defined instructions help produce consistent and focused analysis.

Agent Instruction Template

The following template can be adapted to a specific use case and remains within the 5,000-character instruction limit.

ROLE
Audit [what] across [campaigns / journeys / segments]. Work only
from campaign, journey, and segment data. [External system] data
is unavailable. Never infer or estimate [specific forbidden
inferences].

REFERENCE BENCHMARKS & DEFAULTS
Treat the following as conventions rather than measured values.
Verify against the panel and report deviations:

- [Structural convention, e.g. reminder contact commonly begins
  ~30 days before the due date, escalating at 15, 7, 3 and 1 days]

PRIMARY RULE
The account's own trailing baseline always overrides external
figures. Use published ranges only to place the account within a
wide band, never to declare underperformance.

WHAT TO EXAMINE
1. Inventory — identify every [X]. Report anything that could not
   be classified; never force it into a category.
2. Coverage gaps — identify [cohorts] served by nothing at all,
   with sizes. Look for these gaps deliberately.
3. Structure — [cadence / silence gaps / dead branches /
   stage progression].
4. [Domain-specific check].
5. [Domain-specific check].

OUTPUT
1. Headline — [single quantified finding], with the number.
2. [Table] — [columns].
3. Uncovered [cohorts], sized.
4. Closing statement — [required conclusion or finding].

RULES
- Every finding must cite a name and a number. If it cannot be
  evidenced, do not state it.
- Where [X] is well built, state this rather than manufacturing
  a finding.
- An empty finding is valid and valuable. Do not fill space.
- Use plain, factual, board-readable language. Avoid marketing
  language.

Agent Naming Configuration

The agent name and description appear on the agent card and in the agent picker. The name has a 50-character limit and should clearly communicate the business problem or analysis performed.

Useful names generally fall within 30–44 characters.

For the 300-character description, describe what the agent surfaces rather than the underlying processing.

Example:

Audits win-back journeys for cohorts entering no journey at all, sequences that stop before the reorder window, and dead branches.

Manage Custom Agents

👍### Important Points to Remember

• Custom agents read campaigns, journeys, and segments only. No other data domains are available.
• Deep Research must be enabled for PDF report generation.
• An agent without custom instructions can still respond, but the response remains generic.
• An empty result can be a valid finding. For agents designed to identify gaps, the absence of a matching result should be reported rather than treated as an error.
• Agents are scoped to the panel where they were created and are not shared across panels.
• Panel data refreshes on a scheduled cycle and is not real-time.
• The Co-marketer is AI-powered and can make mistakes. Generated findings should be reviewed against the relevant business context before action is taken.

Troubleshooting

Responsible AI And Transparency

Custom agents inherit the safeguards of the Unified Co-marketer.

Custom agents generate insights only from the configured campaign, journey, and segment data. Supporting metrics are presented with recommendations where applicable, and agent access remains isolated to the associated account.

Generated insights should be reviewed in the context of organisational goals and business requirements before strategic decisions are made.

For more details, refer to AI Security and Governance.