3 ways APIs drive hyper-automation through text mining, machine learning and NLP

Application programming interfaces (APIs) have been around for a few years, but recently, their potential to transform businesses operations—and even mimic human language—has skyrocketed. Text mining solutions such as intelligent document processing (IDP) are no exception, and they are growing more sophisticated with each breakthrough in artificial intelligence (AI) and machine learning.

Text mining APIs allow business leaders to clear manual data entry bottlenecks, tap into state-of-the-art machine learning and AI resources and accelerate the myriad of data extraction tasks that drive back- and front-office business processes. The range of capabilities they bring organizations is expanding rapidly, unlocking exciting opportunities to drive automation and transformation.

The trick is choosing the right API building blocks to fuel business processes, stacking them strategically and mapping a holistic transformation strategy.

APIs act as the bridge that allows companies to quickly and securely carry information to and from the outside world—and as our world grows more interconnected, APIs are cropping up everywhere. For example, an API is what carries Google Maps functionality into a ridesharing app like Uber. In the same way, businesses can use text mining APIs like IDPs to automate data extraction and transform the way they process, organize and share documents and information.

Without IDPs, enterprises have to manually classify and extract data from thousands of complex, unstructured documents. This requires teams to clean or format data, prepare it for import or export and create a training and evaluation model to integrate into production. But with the right IDPs and APIs in place, teams can unleash a series of machine learning and AI technologies like computer vision and natural language processing (NLP) to organize, automate and amplify the speed and volume of the contracts, forms and other documents they consume.

To lay the first row of building blocks for IDPs, organizations can choose text mining APIs like Amazon Textract, Google's Document AI and Azure Form Recognizer. These tools extract text, handwriting and data from scanned documents like contracts, then pull out, classify and validate relevant information to create an easily digestible meta-database. Next in line are programs like Elasticsearch and Kibana, which take the metadata and make it searchable.

By building a stack of API-based services, a data extraction process that would otherwise take months to complete takes days. As a result, teams that would have spent hours processing data can instead easily search through thousands of contracts or documents in the blink of an eye, drawing insights they need to make business decisions faster.

While mapping your organization’s growing AI and API ecosystem, a holistic approach is critical. Take advantage of opportunities to maximize the value of each building block and create an effective automation loop. 

To choose the right combination of machine learning APIs and link them strategically for a range of purposes—like text classification, conversational AI, sentiment analysis, language translation, entity extraction, image classification, speech to text, and text to speech, for example—follow these steps:

1. Build an NLP AI stack. There are many commercial NLP APIs and open source libraries that contain data and entity extraction components. Companies like Google, Amazon and Microsoft provide powerful NL-based APIs for optical character recognition (OCR), named entity recognition (NER) and other powerful text mining capabilities. Use these tools to build your first row of APIs. Then search open source libraries for solutions to targeted problems. Adding these APIs to the top of your stack will help you round out your solution.
2.  Separate the big issues from the small ones. AI-powered APIs and open source libraries may not hold solutions to all your organization’s most cumbersome processes. Where use cases are narrower, use custom algorithms to solve specific domain problems. For example, consider developing custom models using services like AutoML to automate the selection, composition and parameterization of your machine learning models.
   
3. Frame your workflows around people. Despite efforts to automate bug fixes and security updates, achieving 100% accuracy is not feasible without some extra work. In some cases, human validation is a necessary step in the process. Creating a human-centric design for your document processing workflow ensures that with speed, you are also achieving high quality. Wherever confidence in extraction is low, redirect the validation to a team member.  

To find the right API stacks for your organization, a sound integration strategy is essential.

The ecosystem of AI-powered APIs is growing exponentially and the options vary in terms of performance and ease of use. For example, an API for intelligent character recognition (ICR) might work well for hand-written images, but not for digital forms. Other APIs may fit the bill just as well, depending on what types of documents you are processing. Moreover, the output of one API may be easier to integrate into your systems or API stack than another.

When mapping your API strategy, it’s important to weigh your options based on the way each interface works, what they can do and how they fit together. Stacking your APIs strategically can help you fuel automation and get the most value out of each building block.

As AI, machine learning and deep learning capabilities grow increasingly sophisticated and the ecosystem of cloud-based apps expands, APIs are indispensable to today’s business leaders because they allow organizations to harness automation and exchange volumes of information faster than ever. But as capabilities advance, the machine learning and AI space has become less centralized, requiring a new approach to process design.

Today, teams no longer need to build all AI components themselves because there are many APIs that can deliver the same functionality. The shift to decentralization is a modern trend that will likely continue, but it’s not new to the business world. When Ford’s Rouge automotive plant opened its doors in 1921, producing about 1 million Model Ts that year, it was the pinnacle of vertical integration. But since then, the vertically-integrated plant has disintegrated into a network of suppliers. Driven by new information networks and flexible manufacturing technologies, the relationship between suppliers and builders became diffused, and specialized sectors of the economy emerged to support the production of each part of the car.

The machine learning and AI space is going through a similar shift toward decentralization. For today’s business leaders, agility and growth hinge on a modern approach to collecting API building blocks, piecing them together strategically, and topping them off with the right customizations. The organizations that successfully apply these APIs to their business processes will be ready to transform product development, innovate and thrive in a dynamic market.