In NocoBase, company background research can be turned into a traceable automation flow. Business users still work on the familiar company profile page, while workflows and AI employees fill in background information, record the processing steps, and keep every generated report for later review.
This scenario addresses a common problem: company background information is not a static field that only needs to be entered once. Public information changes, regulatory events may appear, and cooperation status keeps evolving as business moves forward. If everything depends on manual updates, key information can easily be missed. If AI is allowed to overwrite company profiles directly, it becomes hard to explain how a given judgment was made. The approach here is to store current information and the research process separately. The company record keeps the version business users are working with, while background check records keep the status, output, and history of every AI research task.
Start With Two Tables #
The company table provides the basic information about the research target, while the background check task table receives each research task. One stores the currently available information; the other stores the process and historical results.
💬 Hey, you’re reading the NocoBase blog. NocoBase is the most extensible AI-powered no-code/low-code development platform for building enterprise applications, internal tools, and all kinds of systems. It’s fully self-hosted, plugin-based, and developer-friendly. → Explore NocoBase on GitHub
companies
: Company Table
| Key Field | Purpose |
|---|---|
| Company name | Serves as the main identifier of the research target. |
| Website | Provides an official website clue and reduces misjudgment caused by companies with similar names or abbreviations. |
| Address | Helps identify region, legal entity, and business scope. |
| Company type | Marks business relationships such as customer, supplier, or partner, making it easier to prioritize follow-up. |
| Background information | Stores the company background report currently in use and presents structured content in Markdown. |
background_check_tasks
: Background Check Task Table
| Key Field | Purpose |
|---|---|
| Company ID / Company name | Records which company this research task is for, making execution and historical review easier. |
| Status | Tracks task transitions frompending to processing and completed ; it also helps prevent duplicate triggers. |
| Research report | Stores the full research report generated by AI for this run. |
| Summary | Stores AI’s summary of the research process, risk points, and information that still needs to be supplemented. |
| Previous background | Stores the previous version before write-back, supporting history tracking and report comparison. |
Enter the Research Flow From Company Profiles #
The company list is the most familiar entry point for business users. On the page, they can see company names, websites, company types, contacts, email addresses, and other information. After opening a company, users can view the current background report or actively start a new background research task.
On the edit page, Background information
is displayed with a Markdown editor. The AI output is not a short summary, but a structured report that can be read, copied, and maintained directly. Business users can still edit it manually, but every AI-generated result will leave a corresponding history record in the background check task table.
As a result, the page still looks like a normal company profile management interface, but the underlying structure has become “current information + research history.” The company table stores the current version, while the task table stores the process and evidence chain.
Three Trigger Methods #
Background research should not depend on only one manual button. In real business work, you may want to automatically complete information after a new company is added, rerun checks for historical records on a schedule, or actively start a new review before signing, revisiting, or reassessing a company.
The New company background check
workflow handles automatic research after a company is added or updated. It listens to data events in the company table and is triggered when a company name exists and background information is empty. After triggering, it does not call AI immediately. Instead, it first checks whether there is already an unfinished task for the same company. Only when no such task exists does it create a new background check record.
The Timing company background check
workflow continuously supplements historical data. It runs every 30 minutes, queries companies whose background information is still empty, and processes them in batches through a loop. Inside the loop, it also checks whether a task already exists before deciding whether to create a new one. This allows the scheduled task to run repeatedly without creating multiple concurrent processing records from repeated scans.
The Manual company background check
workflow is bound to the Run background check
button on the company detail page. It is suitable for business users who want to actively start a research task before visits, signing, or reassessment. Manual and automatic triggers use the same downstream chain: first create a background check record, then let the task execution workflow take over AI research.
These three entry points solve problems at different moments, but they all converge into the same background check task table. New-record triggers, scheduled triggers, and manual triggers only record that “research is needed.” The actual execution, status management, and result write-back are handled by the downstream workflow in a unified way.
Turn AI Research Into Tasks #
Do company background check
is the workflow that actually performs the research. It listens to pending
records in the background check task table. Once an automatic, scheduled, or manual flow creates a task, this workflow is triggered asynchronously.
During execution, the workflow first checks whether the company still exists. If the company no longer exists, the task is closed with an explanation. If the company exists, the task status changes to processing
, and the workflow then calls an AI employee to generate the report. The AI employee’s prompt requires two outputs: a Markdown report that can be written directly into the company background field, and a summary for manual review.
After AI returns structured results, the workflow first writes the report, summary, and previous background content into the background check record, then writes the new report back to the company record. This order avoids the problem of having only the latest result without a process record: the company page keeps the latest usable content, while the task record keeps the context of this generation and the content before write-back.
After research is task-based, batch processing also becomes more natural. The scheduled workflow does not need to wait for each company’s research to finish. It only creates multiple pending records; each record then independently triggers AI research. Multiple companies can move forward in parallel, and a failure or timeout in one task will not block the others.
Make AI Results Reviewable #
AI-generated reports follow a fixed structure: company overview, core business, development history and capital background, market position and competitive perspective, sales follow-up judgment, and reference links. Business users see more than a conclusion. They can also review the risk notes and missing information highlighted by AI in the summary.
The background check record detail page displays Research report
and Previous background
in separate tabs and provides a Copy
action. This makes it easy to copy the current report during discussion, review, or external communication, and to compare it with the previous version when needed.
The record detail page also includes two AI employee tasks:
- Improve background research report: supplement information through conversation, regenerate the report, and write the result back to the company record.
- Compare new and old background research reports: read both reports and let AI explain what substantive changes this update introduced.
This allows AI to go beyond generating a one-time text. It participates in continuous maintenance, review, and version comparison.
How the Workflows Fit Together #
At a high level, this set of workflows can be divided into four layers.
The first layer creates tasks. New company background check
handles companies after creation or update, Timing company background check
supplements historical data, and Manual company background check
supports active manual initiation. They all check for unfinished records before creating a task, reducing duplicate processing at the source.
The second layer executes tasks. Do company background check
listens to background check records, moves pending tasks into processing, calls the AI employee, and writes the report, summary, and current company background field after completion.
The third layer provides AI employees with controlled write-back capability. Update company background
works as a tool workflow, restricting AI to write into specified records with explicit parameters instead of granting overly broad data modification permissions.
The fourth layer cleans up exceptions. Clean overtime processing background check
runs every 30 minutes and cleans up non-completed tasks that have remained unfinished for more than 15 minutes, preventing interrupted tasks from staying in the processing state for too long.
Where This Pattern Can Be Reused #
This scenario is not just a standalone form or a single AI button. It combines several NocoBase capabilities: data tables carry business objects and history records, pages allow business users to view and trigger actions, workflows handle scheduling and write-back, and AI employees generate structured results that can be reviewed.
The same pattern can be applied to supplier onboarding, customer due diligence, contract risk pre-review, lead quality scoring, public opinion tracking, investment target screening, and similar scenarios. Whenever a business process requires continuously completing information, keeping traces of AI results, and preserving historical versions instead of overwriting them, this approach can be used to build a runnable, traceable, and extensible automation flow.