Blog

In March, AI automation platform Nanonetsmade headlines after bagging $29 million in a Series B funding round led by Acceland existing investors Elevation Capital, Y Combinatorand others. 

Despite the increased competition in the workflow automation space, the San Francisco and Bengaluru- based startup has carved a space for itself by automating the most tedious parts of the job for highly skilled professionals such as finance, legal and procurement teams.

According to its founders Sarthak Jain and Prathamesh Juvatkar, eight-year-old Nanonets has been doubling its revenue year-on-year, with most of its revenue coming from the US market followed by Europe.

A major share of the company’s revenue is driven through automating finance processes like accounts payable and reconciliation. The company, however, declined to disclose specific numbers. 

The need for simplicity

Nanonets is Jain and Juvatkar’s second go in the startup space. The duo previously built content aggregation startup Cubeit.io, which was acquired by Myntra in 2012. 

However, at Myntra, they saw the pain points of implementing artificial intelligence, even with a skilled team. This led the duo to start Nanonets at a time when AI wasn’t as popular. as it is today. 

“When we started Nanonets in 2016, the application layer of AI (neural networks) models was just getting started. Early applications like content moderation and image tagging were being discovered and to deploy them at a company was extremely challenging—training, deploying, testing, etc. We knew AI would become a basic requirement at each company adopting technology, hence we started Nanonets,” Jain tells YourStory. 

Even today, many companies require employees to manually review complex documents, often entering data into systems or spreadsheets. More than being time-consuming and error-prone, the job is incredibly monotonous. 

Drawing from their extensive experience in AI and machine learning research labs, they decided to automate these tasks. 

The company enables customers to leverage machine learning tools for automating processes such as invoice processing, accounts reconciliation, and expenses management.

Automating processes 

Data forms the backbone of many businesses today. However, it also remains inaccessible when it’s trapped in unstructured formats such as emails, PDFs, and invoices.

Companies often need to extract data by fetching information from its source and arranging it in a structured layout. This structured data can then be easily used in other software or databases for thorough analysis. This can involve pulling out specific details like contact information or financial data of a company. 

Nanonets can accurately extract data from PDFs, documents, images, emails, scanned documents, and unstructured datasets with over 95% precision. While manual invoice processing typically takes 15 minutes, the startup accelerates this process to under a minute. 

The startup has a monthly processing volume in the millions, and has achieved a Straight Through Processing (STP) rate exceeding 90%, without any human intervention. 

Straight Through Processing (STP) rate refers to the percentage of transactions in a process that are completed without any manual or human intervention. 

Additionally, Natural Language Processing (NLP) enables their technology to understand the contextual meaning within documents, enhancing its capability beyond simple word recognition.

Using discriminative models

Jain says the primary challenge is improving the accuracy of AI models. Unlike generative models, Nanonets uses discriminative models that do not generate new data but find information based solely on provided data. 

“Instead of using generative models, we use discriminative models. These models, though large like generative AI models, don’t make things up. This distinction is crucial, especially in scenarios like a CFO closing monthly books where accuracy is important. What you want is a model finding the right result and running the correct output, as opposed to making something up,” he says. 

The platform is SOC 2 compliant, GDPR compliant, ensuring data is used only for intended purposes. 

To address security concerns, the startup has developed its own models to ensure data is not shared with third parties, ensuring confidentiality. 

Future plans

With its latest round, Nanonets has raised $40 million to date. In 2022, the startup secured $10 million in a Series A investment round led by Elevation Capital.

The company plans to deploy the capital on research and development. Its 110-member team has recorded a fourfold increase in its user base in the past 12 months, it said in a statement. 

Nanonets offers a freemium model to its customers, mainly to evaluate the efficiency of its automation solutions firsthand. Customers are charged based on their paperwork volume.  

At present, the startup has over 10,000 customers, including those using its developer offerings. Some of its major enterprise clients include Swiss pharmaceuticals giant Roche and AsianPaints, an Indian paint major. 

One of the major challenges the startup faces is automating workflows that involve unstructured data, which arises from the variety of document formats and types.

“For example, even for a single document type like invoices, there are thousands of possible formats, and your models need to be smart enough to work across all of them,” Jain says.    

Jain says that few companies offer a solution that has all three: AI-based without manual intervention, high accuracy, and full workflow integration. 

“This is our biggest strength on competitive deals today, and this is also what we consider our biggest challenge – to keep improving accuracy and the quality of our workflows,” he adds.

Some other players in the RPA (Robotic Process Automation) market include Uipath and Automation Anywhere, which tend to focus more on workflow automation rather than on data extraction.

The workflow automation market is expected to reach $34.18 billion by 2029, growing at a CAGR of 9.52% from 2024 to 2029, says a report by Mordor Intelligence. Some of it’s competitors include Docsumo, HyperVerge, Amazon Textract. 

“We grew 100% last year and are on track to grow again, another 100%. We largely sell into a global market. Majority of our customers are in the US, but we have customers all over Europe, Singapore, and Australia and other regions,” Jain adds.