Tierney - stock.adobe.com

How AntWorks is addressing the pitfalls of RPA

The Singapore-based startup hopes to lower the cost of deploying and maintaining robotic process automation technology with a platform that combines machine learning and data processing capabilities

Robotic process automation (RPA) should be part of a larger strategy to drive automation across the enterprise, and those that just focus on RPA tools are setting themselves up for failure, according to a top executive at Singapore-based AntWorks.

Speaking to Computer Weekly in an exclusive interview, Asheesh Mehra, co-founder and group CEO of AntWorks, said most RPA initiatives today are focused on automating tasks rather than business processes, making it hard for enterprises to scale their deployments across an organisation.

Citing a KPMG study which found that most automation initiatives have not gone beyond 10 bots, Mehra said RPA tools need to be augmented with deeper machine learning (ML) capabilities in order to be effective.

That is the premise behind AntWorks, a single “integrated automation platform” that offers ML-infused RPA capabilities to help enterprises automate entire business processes rather than tasks.

A key aspect of the AntWorks platform is its ability to understand and process a large variety of data, which Mehra said is the most critical component in automating a back office process.

While other RPA suppliers might use optical character recognition (OCR) technology to extract structured and semi-structured data, AntWorks built its own data ingestion engine as it wanted high precision rates to avoid the need to handle exceptions in data processing.

“The cleaner the data, or the more curated your data is, the less likely you will get an exception downstream. That’s when true cost benefits start kicking in for enterprises,” said Mehra.

Cognitive machine reading

Using what Mehra dubbed as “cognitive machine reading”, AntWorks’ data ingestion engine can read structured, unstructured, images and inferred data, the last of which refers to derived data about relationships between entities that have not have been codified in existing datasets.

“If you have to deliver a business process straight through, you need to be able to read all kinds of data, which RPA cannot do,” said Mehra, adding that after the data is cleaned and extracted, it is indexed and digitised.

Stressing that digitisation is in fact the third step in data ingestion – and not the first step in the case of RPA tools that use OCR – Mehra said the digitised data is subsequently enriched to achieve a recall rate of 75% to 80%, along with a near-perfect precision rate.

“And we’ll colour-code for you – what we can't understand will be red, what I'm doubtful about will be yellow. And what's green is a 99.99% precision.”

Tweaking automated processes

But the work does not end there. Mehra and his co-founder Govind Sandhu, armed with years of experience in business process outsourcing, saw that software intelligence was necessary to tweak automated processes quickly according to the needs of the business – and to minimise deployment costs.

Noting that the total cost of ownership of RPA tools remains high because specialised skills are needed to deploy and maintain bots using scripts and syntax prescribed by RPA suppliers, Mehra said the business case of RPA initiatives often does not stack up.

“The cost to build and maintain a bot is high because the minute the state of your target system changes, you're going to need to rewrite code,” said Mehra. “And the minute an exception comes in, the bot fails because it doesn't have the cognitive ability to understand that something is a purchase order and not an invoice.”

AntWorks’ platform uses fractal science to make such distinctions quickly using a small amount of data as opposed to neural science that requires far more training data and computing power.

Read more about robotic process automation in APAC

While an all-in-one platform that brings together ML, data processing and automation capabilities helps to ease integration challenges often associated with point solutions, Mehra said enterprises with existing investments in RPA or OCR, for example, can continue using them with AntWorks.

“Our platform is built no differently than Lego – you can take the blocks you need, piece them together and pull out those you don’t need,” said Mehra, adding that even business users can build bots on their own using drag-and-drop functions.

Since its inception in 2015, AntWorks has expanded its footprint to 11 markets. The company is looking to open 12 more offices around the globe, including in locations such as Israel and Vietnam where ML talent is more abundant.

In July 2018, AntWorks, which employs about 350 people worldwide, secured $15m in Series A funding from Japan’s SBI Investment. It’s eyeing another round of funding this year.

Read more on Business process management (BPM)

Data Center
Data Management