Deep Research
Turn a big question into a complete, evidence-backed report — planned up front, researched step by step, and delivered as a branded PDF you can share and schedule.
The Unified Co marketer provides conversational answers for campaign, journey, and segment analysis. Deep Research is designed for detailed reports such as quarterly reviews, channel audits, journey health checks, retention analysis, and executive summaries.
Instead of returning a single response, Deep Research creates a research plan, presents the plan for review, and then analyses the relevant data step by step. The findings are cross checked and assembled into a report, with each reported number traced to the available account data.
The completed report is available in the report library as a downloadable PDF. Follow up questions can also be asked in the same chat.
Dee Research Use CaseA standard chat response is suitable for a single question covering a specific period. Deep Research is suitable for:
- Reports, audits, MBRs, QBRs, deep dives, and executive summaries.
- Analysis spanning multiple periods, such as Q1 versus Q2 or month over month comparisons.
- Analysis spanning multiple domains, such as campaigns, journeys, and segments.
- Requests containing four or more defined sections.
A report shaped request entered in a standard Co marketer chat can also be identified automatically. The Co marketer returns the initial finding and offers to run the complete request through Deep Research.
Deep Research Data Access
- Deep Research follows the same data boundary as the Co marketer. It can read campaigns, journeys, and segments available on the current panel, along with the configured Brand Wiki where available.
- Deep Research does not access the web, industry benchmarks, CRM, order management, finance, or other external systems.
- Deep Research is an analysis only capability. It does not create segments, send campaigns, or edit journeys.
When a request requires an action, the Co marketer indicates that the action is outside the scope of Deep Research and directs the request back to standard chat, where supported actions can be drafted for review.
Run a Deep Research Report
- Log in to the Netcore CE dashboard and click Ask Co marketer in the top navigation bar.
- Click New Chat and then click the + icon to the left of the chat box.
- Enable Deep Research. The setting applies to the current chat.
- Enter the research objective and send the request.
Deep Research can also be activated automatically. A report shaped request entered in a standard chat is identified by the Co marketer, which then presents a research plan.
Enable Deep Research for Custom AgentsA custom agent requires the Deep Research capability to generate a report.
Enable Deep Research under Tools while configuring the custom agent. Without this capability, the agent cannot generate a report. For details, refer to Custom Agents.
Review The Research Plan
Deep Research first interprets the request and generates a research plan. Data retrieval does not begin until the plan is started.
The plan contains the following components:
| Component | Description |
|---|---|
| Title | A short name for the report, such as WhatsApp channel performance this quarter. |
| Objective | The requested outcome carried into the report so the analysis remains aligned with the original request. |
| Steps | The individual investigations required for the report. A plan contains between 2 and 12 steps. |
| Deliverable | The final step of the plan, defining the conclusion the report is expected to provide. |
Nothing runs until Start is selected.
If the plan does not accurately represent the intended analysis, the request can be rephrased to generate a new plan.
Provide A Complete Objective
A request such as "report" or "give me an analysis" does not provide enough information to create a research plan.
Deep Research requests the missing information, such as the topic, period, or metric required for the analysis, before creating the plan.
Run The Research
After the plan is started, the report runs in the background. Progress is reflected in the plan while the research is running.
The research follows these stages:
-
Plan
The approved steps are finalised. Campaigns and journeys named in the objective are resolved to the corresponding entities on the current panel before dependent steps run. -
Research
The defined steps are executed against campaign, journey, and segment data. -
Synthesise
Findings are organised into report sections, with figures linked to the evidence used to derive them. -
Verify
The report is checked against the stated objective to determine whether the analysis addresses the requested outcome. If gaps are identified, the report can be revised within the configured number of attempts. -
Render
The report is formatted using the configured brand theme and exported as a PDF. -
Save
The completed report is added to the report library.
Step Failure Handling
- Deep Research reports incomplete data rather than silently filling missing information.
- If some research steps return no data, the report can still be generated from the completed evidence. The missing information is identified in the report.
- If no relevant data is available, the report indicates that no data was found. This can occur when a specified campaign or journey does not exist on the current panel.
Read The Research Report
The completed report can be opened from the chat or from the report library and downloaded as a PDF.
A report is organised into three to six screens, with each screen focused on an insight rather than a data source.
Each screen can contain:
• Two to four KPI cards showing headline figures with relevant comparisons, such as the previous quarter or the account's trailing median.
• Charts or tables where the available evidence supports visualisation, including trends, channel breakdowns, rankings, and calendar heatmaps.
• Key findings that explain observations not immediately apparent from the numbers.
The final screen contains the deliverable defined in the research plan, such as recommendations, prioritised fixes, or the final finding.
Evidence Labels
Deep Research distinguishes between directly supported data and derived or estimated information.
| Label | Meaning |
|---|---|
| Grounded | The figure was read directly from account data during the research run and is supported by the underlying evidence. |
| Projected | The figure is an estimate, derived value, or statement that could not be directly tied to available evidence. It should be treated as a hypothesis rather than a confirmed fact. |
The labels are determined from the available evidence. A figure cannot be labelled Grounded unless supporting data exists.
Each report also includes a coverage line showing the proportion of claims supported by evidence. For example:
Coverage: 7/9 claims grounded in 12 evidence items (78%).
Evidence Rules
Deep Research applies a set of evidence rules to report findings.
| Rule | Description |
|---|---|
| Current State First | If a journey or campaign is marked as running but has not sent messages recently, the silent stop is treated as the primary finding rather than being reported only as missing recent data. |
| Then Versus Now | At least one research step considers the longest available history alongside the recent period. This allows the report to identify whether performance is declining, stable, or recovering. |
| Data Trust Before Attribution | Event ingestion and audience health are also considered. For example, a journey that stops receiving users because its trigger event is no longer arriving is reported as a tracking issue rather than a targeting issue. |
| No Invented Baselines | Comparisons are based on the account's own history. Deep Research does not use industry benchmarks, competitor figures, or typical rates as factual benchmarks. |
| Revenue Units | Revenue is reported in revenue units because accounts can use different currencies. |
| Empty Findings | When the data does not indicate an issue, the report states that instead of creating an unsupported finding. Missing data is identified rather than filled with assumptions. |
Ask Follow Up Questions
Once a report has been generated in a chat, follow up questions can be asked in the same conversation.
For example:
"Why did email drop in week 3?"
"Summarise the recommendations for the CMO."
Follow up questions are answered using the generated report rather than rerunning the analysis. This keeps the discussion aligned with the figures presented in the report.
If a report has not yet been generated or is still running, the Co marketer indicates the current report status.
The report can also be sent by email to a colleague.
Write An Effective Research Objective
The objective acts as the brief for the report. It is evaluated independently from the rest of the conversation, so it should contain enough information to define the intended analysis.
| Practice | Description |
|---|---|
| Name The Subject | Specify the subject being analysed, such as win back journeys, Diwali email campaigns, or lapsed customers by recency tier. |
| Specify The Period | Define the analysis period, such as the most recent full quarter, July to September 2026, or the available history. |
| Define The Expected Outcome | State the conclusion or recommendation required from the analysis. |
| Use Specific References | Name the campaign, journey, segment, or period instead of using references such as that campaign or last quarter. |
| Keep The Objective Concise | A single self contained sentence generally provides enough information to build the research plan. Additional recurring requirements can be added to a custom agent's instructions. |
Effective Research Objectives
The following examples provide sufficient scope for Deep Research:
• "Full performance diagnostic across campaigns, journeys and segments for the most recent full quarter of available data, including what is working, what is declining, likely drivers, and the top three fixes."
• "Executive summary of how this account is using the platform, with the three highest impact opportunities to improve results next quarter."
• "Compare channel performance across Email, WhatsApp, App Push, Web Push, and RCS for engagement and revenue over the recent period, and recommend where to shift volume."
• "Audit live journeys for completion rates, drop off nodes, and audience waste, and prioritise the fixes that could recover the most value."
• "Analyse lapsed and dormant customers by recency tier and provide a reactivation approach and expected revenue opportunity for each tier."
• "Identify the audience segments associated with the best and worst performing campaigns and identify areas for targeting focus."
Requests Not Suited For Deep Research
| Request | Behaviour |
|---|---|
| Report / analysis please | The request does not contain enough information. Deep Research asks for the topic and period. |
| Did the audit journey send yesterday? | The request requires a single lookup rather than a report. The Co marketer answers it directly in chat. |
| Create a segment of customers who lapsed in the last 90 days | The request requires an action rather than analysis. Deep Research does not create the segment. Standard chat can be used to draft the segment rule for review. |
Schedule A Report
A Deep Research report can be scheduled to run again with updated data and be delivered by email.
Scheduling is managed from the platform.
- In the left navigation, click Schedule.
- Click New Schedule.
- Select the required report.
- Enter the schedule details.
- Click Save.
A report can also be scheduled directly from the report page using the same options.
Each scheduled run uses the same objective against the latest available data, generates a new report, and emails the report to recipients on the active schedule.
Paused schedules do not send reports.
Each recipient receives the report individually. Other recipients are not visible in the email.
Schedule Management
Scheduling is managed from the Schedule section rather than through the chat.
A request to schedule a report through the Co marketer returns the steps required to configure the schedule manually. This keeps scheduled report delivery under account control.
Report Branding And Layout
Reports automatically use the configured brand settings.
Colours and fonts are taken from the saved theme, the configured Brand Wiki where available, or the platform default.
Six report layouts are available:
- Executive
- Minimal
- Editorial
- Compact
- Soft Cards
- Dark
- A custom layout template can also be uploaded, subject to a safety check that rejects scripts and external content.
A theme or template changes the visual presentation only. It cannot introduce, remove, or modify report numbers.
Important Points To Remember
- Deep Research reads campaigns, journeys, and segments, along with the configured Brand Wiki. It does not access the web or external benchmarks.• Deep Research is analysis only. It does not create segments, send campaigns, or edit journeys.
- Research does not start until the plan is started.
- An incomplete objective results in a clarifying request rather than an assumed report.
- Every figure is labelled Grounded or Projected, and each report includes a coverage line.
- A silent stop in a running journey or campaign is treated as a primary finding.
- Re running the same objective within 24 hours reuses the evidence already gathered. After 24 hours, the data is retrieved again.
- Reports are scoped to the panel and chat where they were created.
- Panel data refreshes on a scheduled cycle and is not real time.
- The Co marketer is AI powered and can make mistakes. Report findings should be reviewed against the relevant business context before action is taken.
FAQs & Troubleshooting
Q. Why was a normal answer returned instead of a report?
A. The request may have been interpreted as a single question. Enable Deep Research from the + menu and submit the request again, or provide a report objective containing multiple sections, periods, or domains.
Q. Why does a clarification request appear instead of a research plan?
A. The objective does not contain enough information. Specify the topic, period, and required metric or outcome.
Q. Why is Deep Research unavailable for a custom agent?
A. Deep Research is not enabled under Tools for the custom agent. Edit the agent and enable the capability.
Q. Why can't Deep Research complete the requested action?
A. The request requires an action rather than analysis. Disable Deep Research and use standard chat for supported actions.
Q. Why was no data found?
A. Verify that the specified campaign or journey exists on the current panel and that activity is available for the selected period. Consider using the most recent available data instead of a fixed period.
Q. Why are most figures labelled Projected?
A. The available data is limited for the selected period. Expand the analysis window or narrow the objective to the data available on the panel.
Q. Why is the PDF download unavailable?
A. The report is first available as a web page while the PDF is generated. If the PDF remains unavailable, contact support.
Q. Why does a report match a previous report?
A. The same objective was run within 24 hours and the existing evidence was reused. Wait until the reuse window has passed or modify the objective.
Q. Why was a scheduled report not received?
A. Verify that the schedule is active and that the intended recipient is included in the Schedule configuration.
Q. Why do my numbers appear stale?
A. Panel data refreshes on a scheduled cycle rather than in real time.
Responsible AI And Transparency
Deep Research inherits the safeguards of the Unified Co marketer.
Reports are generated only from configured campaign, journey, and segment data. Each figure is labelled as Grounded or Projected so the supporting evidence can be identified. Reports are also checked against the stated objective before publication.
Deep Research does not browse the web, take actions on account data, or configure schedules on behalf of an account.
Generated reports 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.
Updated about 7 hours ago