Workflows

​

Overview

Turn manual, repetitive processes into automated workflows.

​

Workflow Types

There are three types of workflows.

​

Workflow Modes

There are two types of Run mode:

• Run Live: Automatically execute the workflow live. Process termination occurs if the browser session is closed or the URL changes.
• Run in the Background: Workflows run server-side and will continue to process even if the browser tab is closed or the device is powered off.

​

Nodes Within a Workflow

There are two types of nodes within a workflow - Action and Event.

​

Action Nodes

An Action is a specific task or operation that a workflow performs in response to a trigger:

​

Event Nodes

An Event allows you to set the conditions when to trigger the workflow to run.

​

Action Node Details

​

Instruct

Generate a response from a pre-filled prompt or previous step. Use this when you already know exactly what and how you want the AI to respond.

• Workflow automation frequently requires synthesizing data from multiple sources. For this reason, the Instruct node, which is designed to perform this synthesis, will likely be one of the most used nodes in any workflow.
• The Instruct node is the core of your workflow’s logic. It takes in data and performs a specific action, such as synthesizing information, generating text, or making a decision.
• The more precise your instruction, the better the output will be. Instead of a general command like “summarize this,” specify what kind of summary you want. For example: “Summarize this article into three key bullet points, focusing on the main arguments and conclusions. The tone should be concise and professional.”
• For multi-step workflows, chain multiple Instruct nodes together. This breaks down a complex task into smaller, manageable parts. For instance, instead of one node trying to do everything, you could have the first node to summarize a long document. The second node to extract key facts from the summary. The third node to use those facts to draft an email.
• Tell the Instruct node exactly how you want the final output to be formatted. You can ask for a response in JSON, a list, a table, or simple text. For example, you can specify: “Provide the output as a JSON object with ‘product_name’ and ‘price’ as keys.” This is essential for ensuring the data can be used by subsequent nodes in your workflow.

​

Upload Files

Ask you or other users to upload a file so it can be indexed and used in the workflow (for example, a new report or document each run).

• Use Upload Files node for files that change with each use, like a new report or a daily log.
• This node treats the file as a variable, allowing you to handle different content each time the workflow runs.
• In contrast, if you need to access a fixed set of information repeatedly, use the Contextualize From Library node. This is ideal for consistently referencing static information, such as a company’s product catalog or a list of standard operating procedures, because it treats the file as a persistent source of reference data.

​

Process OCR

Turn text inside images (like scanned documents or photos) into editable, searchable text the AI can read.

• When a task involves extracting text from images, such as scanned documents or photos, use Process OCR. Optical character recognition (OCR) technology analyzes the image to recognize and convert the text into a machine-readable format. This is especially useful for digitizing physical documents, automating data entry, or making scanned content searchable.

​

Input Text

Ask users to input information during Run mode.

• The Input Text node is designed to capture and store a user’s input, which then acts as a variable. This node is not a standalone function; instead, it serves as a crucial component that connects to other nodes within a larger system.

​

Download Document

Turn workflow results into a downloadable document that you can save or share. Available in PDF, DOC, or XLSX formats.

• The Download Document node is suitable for converting synthesized data from an Instruct node into a downloadable report. This allows you to transform the processed information into a polished, shareable document.
• The Download Document node is most powerful when used as the final step in a process.

Powerful Combinations:

• Pair with Instruct node to turn text output into professional reports
• Use after Process OCR node to format extracted text into clean documents
• Combine with Send Email to automatically attach and distribute reports

​

Contextualize From Library

Pull context from your indexed knowledge sources (like SOPs or reference docs) so the workflow can use that information.

• The Contextualize From Library node provides a centralized way to access all indexed data from various sources. Use this node when you need to select specific information as context for your automation tasks. It allows the user to access the institutional body of knowledge to inform or guide a workflow.

• The information provided by this node is only as good as the data it contains. Regularly update your integrated documents or databases to ensure your workflows are using the most current and accurate information.

​

Send Email

Send an email from the workflow to one or more recipients with a subject and body. Replies are not processed by this node.

• Set up an email sender directly within your workflow. This feature allows you to input a subject line and a body, then send the email to one or multiple recipients. You also have the option to include CC recipients.

Powerful Combinations:

• Combine with Scheduled Execution for recurring emails
• Pair with Download Document to attach and send reports

​

Request Email Reply

Send an email and wait for a reply. The recipient’s response is captured so the workflow can continue based on their answer.

• Request Email Reply node is used to send an email to the email recipient with the intention of getting replies in order for the workflow to continue.
• Since receiving email replies may take some time, find the workflow in the execution history at the right side of the screen.

​

Fill In Form

Show a custom form during Run mode so users can fill multiple fields, which are then stored as structured data for the workflow.

• The Fill In Form node acts as a dynamic form builder that prompts users to fill in information based on the form.
• Before you start, identify the goal of your form. What data do you need to collect? What’s the logical order for your questions?
• Use the right input fields. Matching the field type to the data you need is crucial for a good user experience.
• Use labels and placeholder text to guide the user. A clear label like “Enter your email address” is much better than a generic “Input.”
• Mark any critical fields as Required. This prevents the workflow from failing due to missing data.
• Only ask for information you truly need. A long, complex form can lead to user fatigue and a higher drop-off rate.

​

Available Form Fields

​

Get Approval

Request an approve or reject decision from a Shieldbase user and branch the workflow based on their response.

• The Get Approval node is a decision-node used to get a sign-off on a workflow’s output.
• There are only two outcomes - approved and unapproved. Hence the Get Approval node is a decision that branches to two different paths.
• Since receiving approval from the approver may take some time, find the workflow in the Execution History at the right side of the screen.

​

Create a Report

Analyze data and generate a visual report (with charts and tables) that summarizes insights.

• Before you even start, make sure the data you’re feeding into the Create a Report node is clean and well-organized. Use a clear, consistent format, and label your data columns with descriptive names (e.g., Revenue_2024, Region, Customer_ID).
• Explicitly state the purpose in your instructions. For example, instead of just “develop a report,” specify: “Create a sales performance report analyzing monthly revenue trends and showing top-performing regions. The report should include a line graph for revenue over time and a bar chart for regional sales.”
• Use Create a Report node together with Input Text node, Fill In Form node or Contextualize From Library node to generate data analysis and visualization using the information as context for analysis.

​

Store Into a Dataset

Save structured data from the workflow into a dataset so it can be reused, queried, or reported on later.

• Before utilizing the Store Into a Dataset node, ensure that the previous steps in your workflow generate structured data.
• When creating the dataset, always define custom column names that reflect the data’s purpose (e.g., “Customer_ID,” “Transaction_Amount”). This improves readability and compatibility when integrating with tools like spreadsheets, databases, or reporting software.
• If your workflow deals with dynamic inputs (e.g., varying numbers of records), use conditional logic in prior steps to batch or filter data. This prevents overwhelming the dataset with irrelevant entries and keeps performance optimal.
• Be mindful of dataset size in enterprise environments, as large datasets can impact workflow speed.

​

Start a Workflow

Trigger to run another workflow as part of the step.

• Toggle the wait checklist to continue running the first workflow after the triggered workflow has completed.
• The Start a Workflow node is ideal to build a reactive, decoupled workflow capable of running by itself automatically in the background as the main workflow continues running.

​

Refer to a Parent Workflow

Use data or files from specific steps in the workflow(s) that started this one. Useful for aggregating info from multiple parents.

• Name and label key steps in parent workflows clearly so it’s easy to choose the right “Parent Step” when configuring this feature.
• When multiple parent workflows can trigger the same child, standardize output schemas (field names, types) to avoid mapping errors.
• Use this feature to aggregate results from parallel parent branches (e.g., multiple approvals) into a single consolidated summary step.
• Always validate that parent outputs exist before referencing them (e.g., add guards or conditional branches for missing/empty data).
• Log the parent workflow ID and triggering step in the child workflow for easier debugging and audit trails.

​

Loop Action

Loop over a list and run the same actions once for each item (for example, send an email to every address in a list).

• Keep loops focused. Place only the actions that truly need to repeat inside the loop to avoid unnecessary processing time.
• For large lists, add safeguards such as item limits or batch sizes to prevent timeouts and rate-limit issues with downstream systems.
• When calling external APIs with Pull Data From API in a loop, implement throttling or short delays if you notice rate-limit or quota errors.

​

Pause Until

Pause workflow execution until a specified condition is met.

• Use precise, machine-detectable conditions (status fields, timestamps, flags) instead of vague text values to avoid workflows pausing indefinitely.
• Set a maximum wait duration with a failover path (e.g., escalate, notify, or auto-close) to handle conditions that are never met.
• Prefer event-based triggers (webhooks, status updates) over long fixed delays to reduce latency and resource usage.
• Add clear logging messages when entering and exiting the pause so operators can understand where and why runs are waiting.
• When pausing for user input, send proactive notifications (email, chat, or in-app) that include a direct link to complete the required action.

​

Pull Data from API

Pull specific fields from third-party apps to be utilized as context.

• The initial permission to pull data through API from third-party apps must be configured in Integrations.
• Start by pulling only the fields you truly need. Over-fetching unnecessary fields can slow down workflows and increase API costs.
• Normalize and validate API responses immediately after pulling (e.g., type-check, null-check) to prevent downstream failures.
• Log the raw response (or a sanitized version) for debugging, but mask or omit sensitive fields (tokens, PII) from logs.

​

Push Data to API

Push data to third-party apps to be written.

• The initial permission to write data to third-party apps must be configured in Integrations.
• Map fields explicitly and document the mapping, especially when integrating with multiple APIs or versions of the same service.
• Validate and sanitize data before sending to prevent hard failures due to schema or format mismatches on the target side.
• Capture and store response IDs or confirmation tokens from the API for traceability and future updates or deletes.

​

Templatize

Fill in a downloadable template with content while preserving its format.

• For the most accurate results in ensuring content is mapped to the template, use spreadsheets or structured text documents as the template.
• For optimal accuracy, use an empty, tabulated template. This structure ensures the system correctly maps each data field while preventing data errors.
• Design templates with clear placeholder markers (e.g., {field_name}) and maintain a mapping document for template authors and builders.
• Keep formatting logic (fonts, spacing, layout) in the template file and limit dynamic content to text and images where possible.
• Test template fills using edge-case values (very long strings, special characters, empty fields) to prevent broken layouts in real usage.
• Use conditional sections or fallback text for optional fields so templates still look polished when some data is missing.
• Version your templates and track which workflow version uses which template to avoid unexpected formatting changes in live flows.

​

Translate Document

Translate documents from one language to another while preserving formatting and style.

• For domain-specific content (legal, medical, technical), provide glossaries or term lists if supported to improve translation consistency.
• Use translation primarily for content consumption. Run an additional human or specialized review step before using translations in legal or external-facing materials.
• Store both original and translated versions, and include metadata (language, translation date, workflow ID) for future reference and audits.

​

Handle Conversation

Route users to the correct conversation flow based on their specific query.

• Define clear routing criteria (intents, keywords, entities) and keep them mutually exclusive where possible to avoid ambiguous flows.
• Use a fallback or “unsure” route that gracefully handles queries that don’t match any defined conversation path.
• Continuously review real user queries and conversation logs to refine routing rules and add new intents over time.
• Pass along context (user attributes, previous messages, channel) when handing off to downstream flows so they don’t need to re-ask basic questions.
• Implement guardrails for high-risk topics (billing changes, privacy, security) by routing them to specialized flows or human agents.

​

Scheduled Execution Node

Execute a workflow based on a schedule.

• The Scheduled Execution node allows you to run a workflow automatically at a specific date and time. This is perfect for automating routine tasks without any manual intervention.
• Use it to schedule daily, weekly, or monthly tasks. For example, you can set a workflow to generate a weekly sales report every Friday at 9 AM, or to send out a daily email reminder.
• You can also schedule a workflow to run just once at a future date. This is useful for things like sending a timed reminder or launching a campaign on a specific day.
• The Scheduled Execution node acts as a trigger for your entire workflow. Once the scheduled time arrives, the workflow automatically starts and runs to completion, pulling in data, processing it, and generating an output.

​

How to Build a Workflow

​

Build a Linear Workflow

A linear workflow runs in a straight sequence to produce an output. It is a simple, straightforward sequence from start to finish where each step is completed one after another without any loops, branches, or conditional logic. The output of one step becomes the input for the next, creating a clear, predictable chain of events.

​

Build a Decision Workflow

A decision workflow allows you to make a choice from multiple options, guiding the workflow down a specific path based on certain conditions. This is more than just a simple “yes/no” process; it can present multiple options, evaluate different criteria, and then execute the appropriate next steps. At its core, a decision workflow uses a decision node to evaluate a condition. For example, the workflow might ask: “Is the invoice amount greater than $1,000?”

• If Yes: The workflow can be configured to automatically route the invoice to a manager for approval.
• If No: The workflow can send the invoice directly to the finance department for payment.

​

Schedule a Trigger to Run Workflow

To execute a workflow automatically at a specific date and time, you can schedule a trigger. This feature allows you to set up recurring tasks, such as generating a weekly report or sending a daily summary, without any manual intervention.

​

Edit a Workflow

A published workflow can be edited into new versions.

​

Workflow Templates

Workflow templates are pre-built workflows designed to help you get started quickly with common automation tasks. They provide a solid foundation and best practices, saving you the time and effort of building a workflow from scratch.

​

Prompt-to-Workflow

Workflow building can be initiated via Prompt-to-Workflow. Simply input the prompt with specific requirements to allow Shieldbase AI to draft an automated sequence, reducing the need for manual configuration.

​

Pro Tips

​

Best Practices

​

Common Use Cases

​

Data Processing Pipeline

Example Configuration

1. Upload Files: Receive CSV file
2. Process OCR: Extract text from images
3. Instruct: Clean and format data
4. Store Into a Dataset: Update database
5. Create a Report: Generate analytics dashboard
6. Send Email: Notify stakeholders with report

​

Approval Workflow

Example Configuration

1. Fill In Form: Employee submits request
2. Decision Node: Route based on request type/amount
3. Get Approval: Manager reviews request
4. Decision Node: Approve or deny path
5. Download Document: Generate decision letter
6. Send Email: Notify employee of decision
7. Store Into a Dataset: Log request in system

​

Scheduled Reporting

Example Configuration

1. Scheduled Execution: Weekly at 9 AM
2. Contextualize From Library: Pull latest data
3. Instruct: Analyze trends and patterns
4. Create a Report: Generate weekly metrics
5. Download Document: Create PDF report
6. Send Email: Distribute to team

​

When to Automate

​

Good Candidates for Automation

​

Troubleshooting

Workflow fails to execute

• Check all required fields are filled
• Verify data source connections
• Review error logs in Run mode
• Test each step individually
• Ensure proper data formatting

Scheduled workflow not running

• Verify schedule configuration
• Check timezone settings
• Ensure workflow is published (not in draft)
• Review system permissions
• Check for conflicting schedules

Performance issues

• Break complex workflows into smaller ones
• Optimize data queries in Contextualize From Library
• Reduce unnecessary steps
• Avoid processing large files in single steps
• Contact support@shieldbase.ai for optimization help

Form or approval not working

• Verify user permissions
• Check that all required fields are configured
• Test with different user roles
• Ensure email notifications are configured