Speedchain
Not specified
Expense Management
Amazon Textract, Amazon Bedrock, Amazon Comprehend, Amazon DynamoDB, Amazon S3, AWS Lambda, AWS Step Functions, Amazon API Gateway, Amazon SNS, AWS IAM
Speedchain
Not specified
Expense Management
Amazon Textract, Amazon Bedrock, Amazon Comprehend, Amazon DynamoDB, Amazon S3, AWS Lambda, AWS Step Functions, Amazon API Gateway, Amazon SNS, AWS IAM
Speedchain provides expense management for field teams, helping companies code transactions, verify receipts, and manage multi-level approvals. Manual receipt item categorization created friction for cardholders and slowed downstream review and approval. Avahi delivered an Al-augmented receipt categorization workflow that digitizes receipts, classifies items into ranked categories, and routes recommendations to users through an SMS conversational interface for approval. The solution reduced manual data entry while preserving human oversight and accuracy in transaction coding.
Speedchain is an expense management platform built for field teams in industries such as construction, where employees frequently purchase supplies and materials on the job. The platform supports three roles, cardholders, group admins, and admins, and manages transaction coding, receipt verification, and multi-step approval workflows to keep spend controlled and auditable.
Field team purchases often come in as unstructured receipts, with multiple items that must be categorized correctly for cost tracking and compliance. Speedchain’s existing workflow required cardholders to manually categorize each receipt line item, adding time and frustration at the point of purchase submission.
This friction created delays in the approval chain, since incomplete or inconsistent categorization could slow reviews by group admins and admins. Speedchain needed an approach that could extract receipt data automatically, recommend the right categories for each item, and still keep a human in control of the final coding decisions to maintain accuracy and trust.
AWS provided purpose-built services to digitize receipts, apply Al classification, and orchestrate a secure, scalable workflow. Amazon Textract enabled reliable extraction of receipt text and structure, while Amazon Bedrock provided flexible model access for product classification using modern prompting approaches.
AWS also made it straightforward to implement a serverless, event-driven pipeline with durable storage, state tracking, and SMS delivery. This allowed Speedchain to introduce automation without adding operational burden or compromising security and governance.
Speedchain chose Avahi for experience building practical Al workflows that combine automation with human approval loops. Avahi brought a production-minded approach to classification, including ranked suggestions and metadata tracking, ensuring the workflow remained auditable and aligned to expense management requirements.
Avahi also designed the user experience around field realities, delivering an SMS-based conversational flow that meets cardholders where they already are, on mobile, while keeping the approval step fast and lightweight.
Avahi implemented an Al-augmented receipt categorization workflow with human-in-the-lood approval delivered via SMS. The pipeline begins with Amazon Textract digitizing uploaded receipts and extracting text required for downstream item detection and categorization. Raw receipt text and processing artifacts are stored in Amazon S3 to preserve traceability and support repeatable processing.
For classification, Avahi used Amazon Bedrock to categorize detected products using zero-shot and few-shot prompting. The system produces a ranked set of category outputs, including a primary recommendation plus alternative suggestions, giving users and reviewers options when receipts are ambiguous. Amazon Comprehend provided additional classification support where needed to strengthen labeling and confidence.
To keep humans in control, the workflow sends item and category recommendations to cardholders via SMS using Amazon SNS. Cardholders can approve or adjust categories through the conversational interface, enabling fast confirmation without requiring logins or multi-step UI navigation.
The dialogue flow and processing steps are orchestrated through Amazon API Gateway, AWS Step Functions, and AWS Lambda. DynamoDB stores transaction records and maintains state, including pending approval status and reviewer metadata, so group admins and admins can track what was recommended, what was approved, and when. AWS IAM governs access across services and enforces least-privilege controls.
Speedchain gained an automated receipt processing workflow that reduces manual effort for field teams while keeping transaction coding under human control. The system accelerates categorization by turning unstructured receipts into ranked category recommendations and capturing approval through a simple SMS flow, improving speed and usability without sacrificing oversight.
Let’s explore your high-impact AI opportunities together in a complimentary session
Let’s explore your high-impact AI opportunities together in a complimentary session
Let’s explore your high-impact AI opportunities together in a complimentary half-day session
