Automationscribe.com
  • Home
  • AI Scribe
  • AI Tools
  • Artificial Intelligence
  • Contact Us
No Result
View All Result
Automation Scribe
  • Home
  • AI Scribe
  • AI Tools
  • Artificial Intelligence
  • Contact Us
No Result
View All Result
Automationscribe.com
No Result
View All Result

Scaling agentic workflows with native case administration in Amazon Fast Automate

admin by admin
July 12, 2026
in Artificial Intelligence
0
Scaling agentic workflows with native case administration in Amazon Fast Automate
399
SHARES
2.3k
VIEWS
Share on FacebookShare on Twitter


A synthetic intelligence (AI) agent can course of an bill, assist adjudicate a declare, or classify a help ticket in a proof of idea. However operating these brokers throughout 1000’s and even thousands and thousands of labor objects in a manufacturing atmosphere introduces a completely totally different set of challenges. At enterprise scale, success depends upon rather more than the agent itself. Organizations want to trace the state of every work merchandise because it strikes by a number of brokers and programs, floor precisely the place failures happen and why, enable a human to step in when wanted, and scale the infrastructure dynamically primarily based on demand. Amazon Fast Automate addresses these operational challenges by case administration. Each work merchandise is represented as a case that persists all through its lifecycle, offering step-by-step visibility into workflow state, supporting human-in-the-loop processing the place judgment is required, and enabling parallel execution so groups can orchestrate brokers to automate enterprise processes at scale.

On this submit, we present you easy methods to mix case administration with agentic automation capabilities in Fast Automate. We introduce case administration and discover the lifecycle of circumstances in an agentic workflow from case creation by processing to decision. We cowl easy methods to create and handle single or a number of circumstances, mechanically monitor and replace standing, deal with exceptions, and incorporate Human-in-the-loop (HITL) steps inside workflows. We additionally present the case creator-processor sample that allows dynamic scaling. Lastly, we stroll by easy methods to construction case administration for enterprise processes, together with HITL and case monitoring, by a real-life use case.

Fast Automate combines AI brokers and workflow orchestration inside Amazon Fast to automate complicated, end-to-end enterprise processes throughout functions, UIs, and APIs. By natively combining agentic and deterministic automation capabilities with enterprise workflow orchestration, Fast Automate brings construction, visibility, and management to agent-driven enterprise processes. It offers native case administration with trackable work objects together with controls like granular entry administration, exercise logging, model management, exception dealing with, and HITL capabilities. The result’s that processes transfer reliably from initiation to decision at enterprise scale.

Case administration in Fast Automate delivers a number of key advantages. It offers improved visibility as workflows progress by clearly outlined levels with real-time monitoring. Course of house owners can see precisely the place every work merchandise stands, whereas managers can monitor throughput, floor bottlenecks early, and intervene earlier than delays impression enterprise outcomes. Parallel processing permits a number of circumstances to be executed concurrently, growing throughput and serving to organizations meet enterprise service degree agreements (SLAs). Compliance, auditability, and governance are enhanced as a result of each motion, choice, and state transition is recorded as a part of the case historical past. Stakeholder engagement additionally will increase by real-time standing updates that hold operators, managers, and enterprise stakeholders aligned. Communication occurs inside the context of every case, changing fragmented electronic mail threads with centralized collaboration, clear possession, and fewer misplaced handoffs.

Stipulations

Earlier than implementing case administration workflows in Fast Automate, confirm you have got entry to Amazon Fast and that you’re working within the right AWS Area in your group, as a result of workflows inside Fast Automate are region-specific. Your person account will need to have an Enterprise license to writer workflows in Fast Automate.

Case lifecycle and standing

Case administration in Fast Automate transforms complicated, long-running workflows into discrete, trackable work models that may be processed in parallel. Every case capabilities as a discrete work merchandise that progresses by outlined lifecycle levels. A case comprises a case sort, which is a user-defined container for grouping associated work akin to invoices or claims. It additionally features a reference title that serves as a singular ID for monitoring, and customized information fields saved as key-value pairs that maintain the enterprise info wanted for processing. The system mechanically manages further metadata together with the standing of the case within the workflow, exception particulars if a failure happens, and execution logs pertaining to the case. The next desk describes the lifecycle statuses a case can have.

Standing Description How case enters this Standing What occurs subsequent
Prepared Workflow generates new circumstances obtainable for processing Workflow generates a brand new case Workflow consumes the case and begins execution
In Progress Workflow actively processing the case Workflow begins processing a Prepared case Case strikes to Profitable, Failed, or Pending Decision
Profitable Case has accomplished every step with out errors Every step inside Course of case completed with out exceptions Terminal state. No additional processing happens.
Failed Case completed with a enterprise or system exception An exception is raised throughout processing Terminal state – examine logs for particulars.
Pending Decision Case paused, ready for human job completion A job is created within the Activity Middle. Case processing is suspended whereas the system strikes on to the subsequent case. When the duty is accomplished within the Activity Middle, the case mechanically transitions again to the Prepared state with the related human enter.

Case administration automation greatest practices

Case administration in Fast Automate is built-in into the usual automation lifecycle. The next subsections stroll by every section, from preliminary design by monitoring in manufacturing.

Designing an automation

When designing your workflow, begin by figuring out what constitutes a case in your automation. A case represents a single unit of labor, akin to an bill, declare, ticket, or order, that strikes by the workflow from initiation to decision.

Subsequent, resolve easy methods to create circumstances. Circumstances might be created individually as occasions happen or generated in batches from sources akin to recordsdata, databases, or exterior programs. Then decide easy methods to course of circumstances. Circumstances might be processed sequentially or in parallel. Though a single automation can run your entire workflow, groups typically separate obligations into two automations to allow parallel processing and clearer possession. The Case Creator ingests information from sources akin to Excel recordsdata, databases, and internet functions, and transforms every file right into a case representing a discrete work merchandise. The Case Processor consumes circumstances and performs the precise workflow steps, together with exception dealing with and HITL duties. A number of processors can run in parallel to extend throughput.

Lastly, decide the place and when human intervention is required and incorporate it as a part of the workflow. Groups can create human checkpoints that pause a case, current reviewers with related information and choice fields, and mechanically resume processing after the duty is accomplished.

Authoring an automation

Fast Automate offers builders with two actions for ingesting work.

The primary is Create New Case, used when your information supply is a person occasion akin to a person submitting a kind, an API name arriving, or a help ticket being raised. You present a singular reference title (for instance, Order-12345), choose a case sort that teams comparable circumstances (akin to SalesOrders), and fasten key-value pairs containing the case’s customized information. This motion creates precisely one case and instantly units it to Prepared.

The second is Create A number of Circumstances, used when your supply is a spreadsheet, CSV file, or desk of information akin to bulk imports or scheduled information hundreds. You specify the case sort, supply desk, and which column serves because the distinctive ID. The motion creates a separate case for every row, mechanically mapping every column to a customized information area.

Case information typically wants modification as workflows progress by totally different levels. With the replace circumstances motion, you possibly can add or modify present case information by offering a case ID and key-value pairs. Use this motion to file processing outcomes, seize audit timestamps, retailer calculated values, or preserve customized statuses. This motion is simply obtainable for In Progress circumstances. Entry up to date info utilizing updated_case["custom_data"]["key_name"] in your automation. If you want to retrieve or find present circumstances dynamically, use the Search Circumstances motion. With this motion, you possibly can question circumstances primarily based on filters akin to case sort, reference ID, standing, or different attributes. The returned outcomes can then be used to examine case particulars, set off further processing, or replace circumstances as a part of the workflow.

For Human within the loop (HITL) duties, builders can use the Create Consumer Activity motion to pause a case and current a human reviewer with related context and choice fields. Whereas the case waits in Pending Decision standing, different circumstances proceed to course of. When the reviewer completes the duty in Activity Middle, the case mechanically transitions again to Prepared standing with the related human enter, and processing resumes. As a result of the automation resumes from the start, it ought to examine the latest_task_resolution attribute to find out the subsequent department of execution.

Deploying an automation

After an automation is authored, you publish and configure it for manufacturing execution. You possibly can publish workflows and schedule them to run at outlined intervals. Throughout deployment, you configure two key elements. First, you set the variety of parallel cases to run concurrently when configuring the set off, primarily based on the anticipated workload. Case processor automations can run a number of cases in parallel to distribute the case quantity. Second, for HITL duties, you specify the customers or teams who can resolve duties in Activity Middle. Throughout deployment, you add people or teams to grant Fast customers entry to finish duties.

Monitoring an automation

Fast Automate offers real-time visibility into every case because it strikes by a workflow, permitting customers to trace standing throughout every processor occasion from a single dashboard. The monitoring view exhibits key metrics akin to profitable completion charges, enterprise exceptions (attributable to logical or data-related points akin to lacking required fields), system exceptions (attributable to technical failure), and incomplete circumstances. These metrics assist groups shortly determine bottlenecks and measure automation efficiency. Every case file contains its standing, metadata, and processing particulars, and customers can drill into circumstances to view full logs, customized information fields, exception messages, timestamps, and screenshots the place obtainable.

Use case: Automating multi-bank assertion processing with case administration

This use case demonstrates how Fast Automate reduces guide information entry throughout two programs utilizing case administration for parallel processing. We stroll by a real-world treasury consolidation state of affairs, from answer structure by implementation particulars, exhibiting how every idea launched within the earlier sections applies to manufacturing automation.

Firms banking with a number of establishments face a recurring operational problem. Every month, treasury analysts obtain statements from every financial institution’s portal and manually key lots of of transactions into the corporate’s central treasury administration system. A single information entry error for quantity can cascade into reconciliation points that take hours to hint, and Sarbanes-Oxley (SOX) compliance requires each entry to be correct and auditable. The guide course of is sluggish, error-prone, and delays month-end reconciliation.

Answer structure

The next diagram exhibits the end-to-end answer structure for the multi-bank assertion processing automation. The case creator and processor automations work collectively to allow parallel processing and scale back cycle instances at scale. The primary automation, case creator, retrieves the most recent financial institution assertion, extracts every transaction utilizing a customized AI agent, and hundreds them as particular person circumstances. The second automation, case processor, processes every case by navigating to the banking portal, choosing the proper account, and coming into the transaction. A number of processor cases can run concurrently, every selecting up and coming into transactions independently. Funds over $200 are flagged for human assessment, the place a treasury analyst verifies the payee, quantity, and authorization earlier than approving or discarding the transaction. Rejected transactions are captured as enterprise exceptions with the analyst’s motive code and feedback for the audit path.

Architecture diagram of the case creator and case processor automations for multi-bank statement processing

Determine 1: Answer structure for multi-bank assertion processing

Implementation: Creator-processor design

The case creator workflow (Treasury Financial institution Assertion Gathering) performs three key actions. It navigates to the financial institution assertion portal and generates the most recent assertion for obtain. It then makes use of a customized AI agent to parse the downloaded PDF and extract every transaction into structured information fields. Lastly, it batch-creates circumstances with one case per extracted transaction, every set to Prepared state for downstream processing. The next picture exhibits the case creator workflow as authored in Fast Automate. The workflow comprises two course of steps. The primary handles browser navigation to the financial institution portal and PDF obtain, and the second invokes the customized AI agent for extraction adopted by batch case creation.

Case creator workflow in Quick Automate with two process steps for bank portal navigation and AI agent extraction

Determine 2: Case creator workflow in Fast Automate

The Customized agent, which is a specialised job agent configured with actions and directions, is used natively inside the workflow. It receives the downloaded PDF and makes use of a Python code device to extract transaction line objects right into a structured desk. The agent’s output schema defines the fields wanted for every case. The next picture exhibits the agent configuration with its instruction and output schema.

Custom AI agent configuration showing the instruction and output schema for transaction extraction

Determine 3: Customized AI agent configuration

After the agent extracts all transactions, the automation makes use of the Create A number of Circumstances motion to batch-create circumstances. It specifies “Financial institution Transaction” because the case sort, maps the agent’s output desk because the customized information supply, and designates the Reference column because the distinctive ID for every case. The next picture exhibits the batch case creation configuration inside the workflow.

Create multiple cases configuration mapping the agent output table to case custom data fields

Determine 4: Create a number of circumstances

After it’s created, every case comprises the complete transaction information as customized attributes. A case file contains the case standing (Prepared), its reference title, the case sort (Financial institution Transaction), and customized information fields containing the financial institution title, account quantity, transaction date, description, sort, and quantity. The next picture exhibits the Monitor dashboard with the case desk and its key elements.

Monitor dashboard showing created cases with status, reference name, and custom data fields

Determine 5: Monitor dashboard with created circumstances

The case processor workflow (Treasury Financial institution Transaction Entry) retrieves circumstances in Prepared standing and processes every transaction independently. It first opens the banking portal and authenticates as an administrator. It then enters a processing loop that retrieves circumstances of sort “Financial institution Transaction” and handles them one by one. As a result of Fast Automate helps a number of concurrent processor cases, you possibly can scale this automation horizontally to match transaction quantity.

For every case, the processor first checks whether or not the case has already been by human assessment. If the case was beforehand routed to a treasury analyst and has returned with a decision, the processor checks the authorization final result. Accredited circumstances proceed on to transaction entry. Rejected circumstances are raised as enterprise exceptions with the analyst’s motive code and feedback, creating an entire audit file with out coming into the transaction into the portal.

For circumstances that haven’t but been reviewed, the processor evaluates the fee threshold. If the transaction is a fee (Credit score sort) and the quantity is $200 or extra, the case is routed to a treasury analyst for dual-authorization assessment by the Activity Middle. The HITL job contains transaction particulars, an authorization drop-down checklist with Accredited and Rejected choices, a motive code drop-down checklist for rejections, and a free-text feedback area. The case enters Pending Decision standing whereas awaiting the analyst’s choice, and the processor strikes on to the subsequent obtainable case with out ready.

The next picture exhibits the case processor workflow as authored in Fast Automate, together with the conditional logic for human assessment dealing with, fee threshold analysis, HITL job creation, and transaction entry.

Case processor workflow in Quick Automate with conditional logic for human review, payment threshold, and transaction entry

Determine 6: Case processor workflow in Fast Automate

For transactions that don’t require human assessment, or which have been authorized by an analyst, the processor makes use of a UI agent to enter the transaction into the banking portal. The agent navigates to the dashboard, provides a brand new transaction, selects the suitable sort and quantity together with different particulars, and submits the shape. The agent additionally validates that the transaction entry succeeded. The next picture exhibits the UI agent configuration with its pure language instruction for transaction entry and validation.

UI agent configuration with a natural language instruction for entering and validating transactions in the banking portal

Determine 7: Native UI agent and instruction

Monitoring case particulars and standing

The Monitor dashboard offers real-time visibility into case processing efficiency for the treasury consolidation automation. From this view, operators can monitor what number of circumstances have been created, what number of are presently in progress, and what number of have succeeded, failed, or are pending human decision. The dashboard contains charts exhibiting case creation traits and exception patterns over time, supporting each operational monitoring and compliance audit wants.

The next picture exhibits the Circumstances tab within the Monitor dashboard, displaying case statuses, processing instances, and exception particulars throughout all processor cases.

Cases tab in the Monitor dashboard showing case statuses, processing times, and exception details

Determine 8: Circumstances

Operators can even entry HITL duties by the Duties tab to view which transactions are awaiting analyst assessment, monitor SLA adherence, and see the decision historical past together with approval choices and rejection causes. The next picture exhibits the HITL job because it seems within the Activity Middle. The treasury analyst sees the complete transaction context together with the authorization kind fields for his or her assessment choice.

HITL task in the Task Center showing transaction context and authorization form fields for analyst review

Determine 9: Duties

Conclusion

Case administration in Fast Automate empowers you to deal with high-volume, complicated automations at scale. By changing your work into discrete, trackable circumstances with outlined lifecycles, you achieve full operational visibility into workflows. This method reduces cycle instances by parallel processing, improves choice high quality by routing essential circumstances to human reviewers, and builds operational confidence by real-time monitoring that surfaces exceptions earlier than they impression outcomes.

Subsequent steps

Begin implementing case administration in your workflows by reviewing the orchestration documentation for Fast Automate and attempting the automations as pre-built samples as described on this submit. Open Fast Automate, navigate to your challenge, select Begin constructing, and seek for “Treasury Financial institution Assertion Gathering” (case creator) and “Treasury Financial institution Transaction Entry” (case processor) within the pattern automation library. After you open each samples, you possibly can run the case creator first to generate transaction circumstances, then configure and run the case processor to start coming into transactions in parallel.

You possibly can adapt this sample to different automations like bill processing or claims adjudication by modifying the case creation logic and processor routing guidelines whereas sustaining the identical core orchestration construction. For extra assets, see the public documentation for Fast Automate.


Concerning the authors

Sumit Wasuja

Sumit Wasuja

Sumit is a product chief whose work has helped form the evolution of automation from rule-based course of orchestration to clever, agentic workflows. Because the founding product chief for Amazon Fast Automate at AWS, he leads the event of agentic AI programs that allow organizations to automate complicated end-to-end enterprise processes, from pure language descriptions to production-ready workflows that motive, act, and adapt autonomously. Over practically twenty years, Sumit has suggested and led transformations for among the world’s largest enterprises throughout monetary companies, healthcare, power, and expertise, serving to them reimagine how work will get completed by the progressive utility of automation, AI, and human-machine collaboration.

Kushal Bhattacharya

Kushal Bhattacharya

Kushal is on the Amazon Fast Automate product group at AWS, the place he works on constructing agentic AI merchandise that enable organizations to rework complicated enterprise processes into autonomous, self-adapting workflows. Over the previous decade, he has led clever automation and AI transformations for big enterprises throughout monetary companies, insurance coverage, healthcare, power, and expertise, architecting options that join rising AI capabilities to actual operational challenges. His work spans the complete arc from answer design to product technique, serving to groups undertake and scale AI-powered automations from proof of idea to manufacturing.

Srikanth Bharadwaz Samudrala

Srikanth Bharadwaz Samudrala

Srikanth is a Sr. Fast Builder within the ProServe group at AWS. He works straight with giant enterprise prospects throughout monetary companies, healthcare, pharma, and leisure to design and ship generative AI options utilizing Amazon Fast. He begins by understanding their most complicated workflows and figuring out the place agentic AI could make a fabric distinction. From there, he takes these concepts from use-case discovery and answer structure by constructing POCs and MVPs that show out the worth earlier than scaling towards manufacturing. His focus all through is ensuring the correct safety controls are in place, the answer suits inside the buyer’s present atmosphere, and measurable ROI is delivered at each stage.

Tags: agenticAmazonAutomateCaseManagementnativeQuickscalingWorkflows
Previous Post

That Is Embarrassing: Why Frontier AI Nonetheless Makes Issues Up, and What to Do About It

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Popular News

  • Greatest practices for Amazon SageMaker HyperPod activity governance

    Greatest practices for Amazon SageMaker HyperPod activity governance

    405 shares
    Share 162 Tweet 101
  • How Cursor Really Indexes Your Codebase

    405 shares
    Share 162 Tweet 101
  • Construct a serverless audio summarization resolution with Amazon Bedrock and Whisper

    404 shares
    Share 162 Tweet 101
  • Context Engineering — A Complete Fingers-On Tutorial with DSPy

    403 shares
    Share 161 Tweet 101
  • Speed up edge AI improvement with SiMa.ai Edgematic with a seamless AWS integration

    403 shares
    Share 161 Tweet 101

About Us

Automation Scribe is your go-to site for easy-to-understand Artificial Intelligence (AI) articles. Discover insights on AI tools, AI Scribe, and more. Stay updated with the latest advancements in AI technology. Dive into the world of automation with simplified explanations and informative content. Visit us today!

Category

  • AI Scribe
  • AI Tools
  • Artificial Intelligence

Recent Posts

  • Scaling agentic workflows with native case administration in Amazon Fast Automate
  • That Is Embarrassing: Why Frontier AI Nonetheless Makes Issues Up, and What to Do About It
  • Construct a semantic layer for agentic AI on AWS with Stardog and Amazon Bedrock AgentCore
  • Home
  • Contact Us
  • Disclaimer
  • Privacy Policy
  • Terms & Conditions

© 2024 automationscribe.com. All rights reserved.

No Result
View All Result
  • Home
  • AI Scribe
  • AI Tools
  • Artificial Intelligence
  • Contact Us

© 2024 automationscribe.com. All rights reserved.