case studies
Performance – people, augmented by technology
Efficiency and automation technologies, including Artificial Intelligence (AI) and Machine Learning (ML), serve as productivity tools, similar to how a power tool assists a woodworker or how a robot adds efficiency to an assembly line process. These technologies enable individuals and teams to achieve more, work faster, and improve accuracy. We strategically implement these technologies to bolster the capabilities of our human teams, making them more effective and efficient.
The development of decision models, like AI and ML tools, is progressing rapidly, with capabilities improving at an accelerated pace. Many tasks can be highly automated, while others may require more intervention. However, like any process, automated workflows require monitoring to ensure quality delivery. We employ scoring systems and confidence intervals to identify potential issues, flagging them for human review before finalization. This approach allows us to benefit from the efficiencies gained through AI and ML while maintaining high-quality standards. Any discrepancies found during our reviews are used to further train our AI and ML models, continuously enhancing their performance on your tasks.
For example, when using AI to read digitized images of documents, classify them, and extract and normalize data, we create a model that generates a statistical confidence score for each document type and each piece of extracted data. If this score falls outside the desired confidence interval, the document or data item is automatically escalated to a subject matter expert for review. The expert then either corrects or confirms the data, ensuring quality assurance. The results of this review are fed back into the model to improve its performance for future tasks.
Digital mailroom
Problem Analysis:
Large mailrooms require significant labor and time to sort and route incoming mail to the intended recipients, leading to high costs, inefficiency, and delays.
Solution
Instead of manually sorting and routing mail, a process is created where incoming mail is opened and scanned mechanically. A list of document types is generated, along with routing rules and the data fields to be extracted for each. A proprietary AI model is trained using actual sample documents from the client, ensuring both speed and optimal results tailored to the client’s needs. As each document is scanned and classified it is scored. If the score is within the desired confidence interval, it is automatically forwarded to the designated individual's digital mailbox. If the AI model does not recognize a document or the score falls outside of the desired confidence level, it is flagged for review by a human agent. The agent classifies the document, and the system assigns it to the appropriate individual. Additionally, any information gained from this human review is incorporated into the document routing rules, enabling the AI model to recognize the document in future encounters.
Results
Sorting and re-routing time is reduced from hours or days to mere seconds. This improvement also generates downstream efficiency benefits beyond the mailroom. Digitized images and extracted data facilitate automated tracking, reporting, and even automated actions.