In the evolving landscape of small business operations, technology continues to carve new pathways for efficiency, yet many traditional shops remain rooted in time-honored practices. This dichotomy is evident in the bustling markets and neighborhood convenience stores of the Philippines, where handwritten sales logs have long been the cornerstone of daily trade. Researchers from Ateneo de Manila University’s Business Insights Laboratory for Development (BUILD) have embarked on a groundbreaking journey to harmonize cutting-edge artificial intelligence (AI) with the analog simplicity cherished by small business owners. Their innovative AI system transforms the mundane, handwritten sales logbooks into powerful, actionable business intelligence, poised to revolutionize operations without overshadowing the invaluable role of human workers.
Small businesses, the lifeblood of many economies including the Philippines, often rely heavily on manual methods of record-keeping. The pen-and-paper sales log remains a ubiquitous tool due to its affordability, accessibility, and resilience in environments where electronic devices can be impractical. However, these handwritten ledgers, while reliable, pose significant challenges when it comes to data aggregation and analysis. Decoding and tabulating such records is a laborious process, often limiting owners’ ability to fully understand product performance, inventory flow, and sales trends. BUILD researchers recognized this gap and aimed to create a seamless bridge between manual logging and digital analytics.
Harnessing the power of advanced AI, particularly optical character recognition (OCR) combined with large language models (LLMs), the team has engineered a system that reads and interprets handwritten text with surprising accuracy. The use of OCR technology enables the extraction of raw data from photographed logbooks, converting it into digitized text. This forms the foundation upon which AI tools like Anthropic’s Claude 3 Haiku LLM operate, further refining the output by contextualizing entries, matching product names with prices, and summarizing sales figures. By melding these technologies, the researchers have enhanced the precision of data interpretation, even when faced with the idiosyncrasies of varied handwriting styles.
The conceptual framework of this AI system is strikingly human-centric. Rather than seeking to replace workers or force a steep technological learning curve, the researchers advocate a “copilot” model. Here, AI tools act as augmentative partners, enabling shopkeepers and stall operators—who may have limited digital literacy—to effortlessly glean insights from their own handwriting. This inclusive approach respects the nuances of small-scale business ecosystems and underscores a broader ethos: technology as a facilitator of human agency, not its usurper.
Extensive field testing was conducted at Ateneo’s Student Enterprise Center, where the prototype underwent rigorous trials in an actual food stall environment. Utilizing Python-based programming and cloud services from Amazon Web Services for the OCR layer, combined with Anthropic’s Claude 3 Haiku LLM for language understanding, the system processes photographed log pages swiftly. Its user-friendly interface requires minimal interaction, producing digestible summaries and visualizations that highlight key metrics such as bestselling items, inventory depletion rates, and price fluctuations. These insights empower vendors to make timely decisions aligned with consumer demand and operational capacity.
While the current accuracy rate of the system is moderate, the researchers emphasize its potential for continuous refinement through iterative training on diverse handwriting samples and shorthand variations. The adaptive quality of the AI is pivotal, as it must accommodate unique notations, regional dialects, and the natural inconsistency found in manual record-keeping. Over time, with expanded datasets and improved algorithms, the system’s reliability is expected to reach levels that transform it from a helpful assistant to an indispensable business tool.
Beyond sales logbooks, the research team envisions broader applications for their AI model. Handwritten documents such as inventory lists, supplier delivery records, and payroll ledgers also represent untapped troves of business data trapped in analog formats. By extending their system’s capabilities to encompass these documents, small enterprises can unlock comprehensive digital audit trails with minimal disruption. Such scalability underscores the versatility and pragmatic value of the technology.
A critical aspect distinguishing this research is its insistence on accessibility and affordability. The architecture of the AI tool leans on widely available cloud services and open-source programming languages, ensuring that costs remain manageable for small business stakeholders. By lowering the barrier to entry, the team paves the way for widespread adoption, particularly among micro and informal enterprises traditionally marginalized from sophisticated data analytics.
The intersection of AI and human-computer interaction explored in this research epitomizes a forward-thinking paradigm in technological development. Presented recently at the Artificial Intelligence in Human-Computer Interaction Conference 2025 in Sweden, the study has attracted attention for marrying practical utility with cutting-edge AI innovations. Its contribution lies not merely in technical prowess but in sensitivity to social context, economic realities, and the lived experiences of small business owners.
Furthermore, the project confronts common fears related to digital transformation—namely, concerns over job displacement and unfamiliar technology. By positioning AI as a supportive copilot, the researchers advocate a future where human expertise is enhanced rather than overshadowed by machines. This philosophy resonates deeply in communities where entrepreneurial spirit is intertwined with personal identity and cultural heritage.
In essence, this AI-driven system developed by BUILD researchers represents a significant stride toward democratizing business intelligence for traditional small enterprises. By synthesizing optical character recognition with sophisticated language models, they deliver a solution that is precise yet approachable, innovative yet grounded. As AI continues its relentless progress, initiatives like this ensure that its benefits extend inclusively to those who need it most, preserving the human touch that defines small business success.
The journey from scribbled sales notes to clear, actionable insights symbolizes the transformative potential at the nexus of technology and tradition. With this pioneering work, Ateneo de Manila University stands at the forefront of a movement that reimagines how small businesses can thrive in an increasingly digital world—preserving the past while embracing the future.
Subject of Research: Application of Optical Character Recognition and Large Language Models in Enhancing Manual Business Processes for Small Traditional Enterprises
Article Title: Applied Optical Character Recognition and Large Language Models in Augmenting Manual Business Processes for Data Analytics in Traditional Small Businesses with Minimal Digital Adoption
News Publication Date: 30-May-2025
Web References: http://dx.doi.org/10.1007/978-3-031-93429-2_18
Image Credits: Ccai Llamas / The Guidon
Keywords: Artificial Intelligence, Optical Character Recognition, Large Language Models, Small Business, Data Analytics, Digital Transformation, Handwritten Logs, AI Augmentation, Python, Amazon Web Services, Anthropic Claude 3 Haiku, Human-Computer Interaction
