NLP and RPA: UiPath gets access to Re:infer

The Romanian-born publisher, based in New York, relies more not only on robotic process automation, but also on artificial intelligence. In 2020, he introduced AI Fabric, which became AI Center in April 2021.

In addition to the compute instance hosted by Kubernetes, AI Center is a platform for implementing ready-to-use machine learning and deep learning models offered by UiPath or imported by data scientists. These can be NLP, computer vision or OCR models (via the Document Understanding tool), as well as more traditional statistical models.

In this regard, UiPath announced the acquisition of Re:infer for an undisclosed amount. This British startup is a spin-off from the AI ​​research lab at University College London. Since then, the company has raised over $7.6 million (about $11 million according to Crunchbase) to develop its natural language processing (NLP) technology. According to information available on LinkedIn, it has about 40 employees.

Since 2015, Re:infer has been developing a platform for analytics and automation. In particular, the startup intends to analyze the communication between the company and its customers in near real time, and then automate the responses to their requests.

For example, insurers implementing Re:infer use it to review claims and automatically triage claims through various communication channels with the account manager. Some e-commerce companies rely on this solution to identify (bad) experiences that worsen customer retention.

Re:infer, competitor to IBM Watson

Hosted on Google Cloud, this SaaS platform is based on a no-code approach and AutoML techniques.

This allows you to extract semi-structured data from email, messages, CRM, call transcripts, or other documents. The vendor documentation clarifies that a model can only be trained from one type of data source (e.g. emails, transcripts, etc.).

It then uses Re:infer, its own deep learning algorithm, to clean up the data. The startup then relies on unsupervised learning techniques to recognize “trends.” In other words, the algorithm determines the most common terms and intents in the dataset and creates “message clusters”.

According to Re:infer, you don’t need to enter large amounts of data into it to get convincing results.

Data scientists or analyst teams can then annotate these datasets according to the problem they are trying to solve and then train the model, this time using a supervised learning technique.

To do this, the startup has developed two types of taxonomy. One contains labels describing the role of the message as a whole: an order, a clause, a statement, a comment, etc. The second is used to report the entities in the message: names, dates, addresses, values, etc.

The platform, based on the “transform” algorithm developed by Re:infer, gives indications of patterns that it considers to be the most appropriate for an annotation and those where the labeling seems out of place.

The business intelligence and analytics features allow you to learn the first lessons, such as deciding whether forecasts are good or bad.

The model can continue to learn as annotators enrich their taxonomy. The model is automatically retrained and recalculated. Users can also study the behavior of a model on the same dataset where at least two different annotations have been applied. This allows us to see how label distribution affects the performance of the NLP model.

After performance verification, the model is re-analyzed and trained on previously obtained data. Finally, the model can be deployed to top-down systems (CRM, CMS, MIS) and RPA tools.

Indeed, objects and labels are available on the platform for analytical purposes or via API to download the automation process. Templates can also include triggers to initiate actions, create a case in the CASE Management platform, or receive alerts.

For connoisseurs, Re:Infer is very similar to IBM Watson Natural Language Understanding, a solution that the startup claims in 2019. His platform was supposedly 30% more accurate and 200 times faster than the IBM solution.

The London-based startup has already worked with RPA players including Blue Prism, Automation Anywhere, Microsoft (Power Automate, Logic Apps) and UiPath.

It is with the latter that Re:infer has the closest cooperation. The fledgling publisher has several clients in common with UiPath, including specialist insurance company Hiscox. UiPath also invested in a startup.

“By bringing together UiPath and Re:infer, we have achieved significant business success through automation, enabling automated cataloging, sorting, and responding to tens of thousands of email requests per month,” said Marco Rodriguez, automation manager at Hiscox, in a press release. release. . “Thanks to UiPath and Re:infer, our insurance brokers receive instant automated responses to email inquiries. Our SLAs have been reduced from days to hours […] “.

Sealing holes in the racket of automation

Re:infer also convinced UBS, Deutsche Bank, Expedia and

UiPath customers can already access the NLP platform from a private preview. The RPA specialist does not explain whether he intends to optimize the integration of Re:infer with his own tools. For now, he intends to look at the most common use cases for his clients, namely email sorting, sentiment analysis from comments, and automatic case attribution.

“Our customers are inundated with documents, messages and data that they need to understand and process efficiently,” Ted Kummert, executive vice president of products and engineering for UiPath, said in a press release. “Combining NLP Re:infer technology with our document understanding and AI products expands the range of our current automation capabilities and opens up new automation opportunities for our customers.”

From a commercial standpoint, it’s about bringing in new logos and a greater combination of RPA and AI. UiPath already has over 10,000 clients. UiPath plans to detail its roadmap at the end of September next year during the annual FORWARD 5 conference.

In early June, the company released financial results for the first fiscal quarter of 2023 ending April 30. During this period, UiPath’s turnover was $245.1 million, up 32% from last year.

Its net loss reached $122.5 million compared to $239.6 million in the first fiscal quarter of 2022. However, its share price on the NYSE fell from $74.84 on April 23, 2021 to $18.33 on April 29. July 2022.

On June 27, UiPath announced that it was laying off 5% of its employees, approximately 210 positions out of 4,200.

Back to top button

Adblock Detected

Please consider supporting us by disabling your ad blocker.