Efficiency drives modern business operations, yet many organizations remain burdened by cumbersome, manual processes that slow productivity and increase costs. WorkDone, an artificial intelligence-driven SaaS company specializing in business process automation, addresses this challenge through its flagship program, Corporate Memory.
Designed to observe and optimize enterprise workflows by continuously learning from user interactions, Corporate Memory analyzes employee activities across multiple sources, from enterprise resource planning and inventory management systems to websites, email attachments, personal spreadsheets and more.
In order for the solution to scale effectively and deliver maximum value, WorkDone needed a more sophisticated method for extracting structured data from unstructured sources without relying on costly human intervention. That’s where Cardinal Peak came in.
The Challenge: Using Video & Gen AI for Workflow Complexities
WorkDone’s mission? To develop a solution that could peek behind the curtain of complex business processes and reveal optimization opportunities.
Across manufacturing operations, inefficiency goes beyond a mere nuisance to pose a serious bottom-line threat. For many manufacturing enterprises, procurement processes involve a tangled web of enterprise resource planning (ERP) systems, spreadsheets, webpages, and emails, forcing employees to manually copy and paste purchase order details between platforms.
Adding to the challenge, legacy ERP systems without modern APIs created data silos. Procurement specialists found themselves trapped in a manual maze: checking inventory levels, comparing vendor pricing, inputting purchase orders and tracking progress across multiple disconnected systems. Fragmented workflows open the door to harmful inefficiencies, such as duplicate orders consuming unnecessary resources, excess inventory tying up capital, supply chain bottlenecks and reactive rush purchases at premium prices.
Understanding the need for more than a conventional software solution to extract meaningful insights from employee workflows without time-intensive manual intervention, WorkDone tapped Cardinal Peak’s multifaceted product design services. The goal was to streamline procurement processes and eliminate redundant manual tasks.
The initial attempt — feeding raw screen recordings directly into a large language model (LLM) — proved financially unfeasible. At more than $1,800 per seat monthly, the approach was untenable. Our internal team knew WorkDone needed a more intelligent strategy.
Cardinal Peak’s engineers implemented intelligent filtering of the data before AI processing, dropping the LLM workload by 99% and reducing monthly costs per user by more than 99.55%.
Our Solution: Engineering a Smarter AI Workflow Automation Engine
Cardinal Peak collaborated with WorkDone to design an AI-powered video review system that integrates seamlessly with WorkDone’s Corporate Memory product. The key was creating an intelligent preprocessing layer that could dramatically reduce unnecessary data loads and computational costs while maintaining high-accuracy insights.
Optimizing the Preprocessing Layer
Combining AI with traditional computer vision techniques, Cardinal Peak engineered a highly efficient preprocessing pipeline that selects only the most relevant video frames for AI analysis, not consecutive hours of screen recordings. From there, Corporate Memory intelligently extracts and analyzes both text and visual process data, enabling real-time insights without excessive computational overhead.
Our approach to enhancing Corporate Memory employed:
Scene change detection and optical character recognition (OCR)
Identifies meaningful changes on the screen and extracts relevant text and visual information from application interfaces.
Contextual process mapping
Recognizes when a user performs specific tasks — such as entering purchase orders — by correlating extracted text and visual data with application behavior.
Smart frame selection
Filters redundant or useless frames using a sophisticated algorithm, ensuring the AI model only processes critical workflow steps.
By leveraging Anthropic Claude (3, 3.5 Sonnet and Haiku) and other LLMs, as well as Amazon Bedrock and other AWS services — including AWS Lambda, Amazon RDS, Amazon API Gateway and AWS Fargate — our team developed a highly scalable, serverless architecture that extracts structured data from unstructured sources like screen recordings, keystrokes and spreadsheets with minimal overhead.
For example, the system can distinguish between a purchase order entered in Excel versus an ERP system, providing nuanced workflow understanding that goes beyond simple data capture.
Visual Workflow Efficiency Video & GenAI Solution Architecture
Refining the LLM for Continuous Improvement
We implemented human-in-the-loop validation to ensure exceptional accuracy and enhance the algorithm’s performance. By manually tagging data to establish performance benchmarks, WorkDone employees helped create a robust feedback mechanism for continuous model refinement.
Our self-improving prompt tuning mechanism allowed the AI to use human-tagged data and past results to progressively enhance its processing logic. This approach ensured increasing precision without escalating costs — a critical consideration for WorkDone’s business model.
Recognizing the sensitive nature of workflow monitoring, we also prioritized employee consent and data privacy. The Cardinal Peak-built solution includes robust security measures like Amazon API Gateway authentication and transparent monitoring protocols.
Reimagining how AI models process and interpret workflow data helped us optimize costs and transform how WorkDone’s technology empowers businesses to operate more efficiently.
The Results: Workflow Efficiency System Leveraging Video & GenAI
The results of Cardinal Peak’s AI-driven workflow optimization were transformative:
- Significantly reduced costs and 99% reduction in the LLM workload: By intelligently filtering data before AI processing, cloud costs dropped considerably.
- Seamless scalability: The fully automated, serverless infrastructure scales cost-effectively and dynamically based on workload demands during business hours and inactive periods.
- Enhanced accuracy and completeness over time: A built-in feedback loop refines LLM prompt tuning, ensuring continuous improvement in process automation while minimizing manual oversight.
- Optimized procurement processes: Automated workflow analysis boosts operational efficiencies by eliminating redundant orders, enhancing purchasing efficiency and minimizing operational risk.
- Improved onboarding operations: Onboarding is a significant cost to businesses, but it helps new employees understand their roles, company culture and goals. Corporate Memory makes getting new employees up to speed faster than ever.
With these advancements, WorkDone’s Corporate Memory platform now drives enterprisewide efficiency at a fraction of the cost, modernizing different organizations’ procurement workflows with confidence.
Why Cardinal Peak?
As WorkDone continues to expand its Corporate Memory product, Cardinal Peak remains a trusted technical partner. We don’t just solve problems but reimagine possibilities. By combining deep technical expertise with creative problem-solving, we transformed WorkDone’s vision into a market-ready solution that unlocks new dimensions of business intelligence.
Whether optimizing business processes, refining AI models or designing scalable cloud solutions, we help our clients confidently innovate. Ready to streamline your workflows with AI-driven efficiency? Contact us today to see how Cardinal Peak can help your business scale smarter.
Expertise
Partnerships
Tools
- Anthropic Claude (3, 3.5 Sonnet and Haiku)
- Amazon Bedrock
- Amazon Lambda
- Amazon RDS
- Amazon API Gateway
- AWS Fargate
- Tesseract OCR