Re:infer


Founded 2015 | HQ London | 11-50 employees | <$5M annual revenues

Determining the intent of communications and messages has immense potential to add a missing piece to the enterprise automation jigsaw puzzle. Re:infer’s AI platform can convert unstructured communications into structured data for the analysis of intent.


The Company

Re:infer is an AI start-up that was founded in 2015 by three colleagues from the famed University College of London (UCL) Centre for AI Research. UCL provided seed money to help launch the company. By 2019 Re:infer was recognized as the UK’s AI Innovator of the Year by the UK government, and in the same year it beat IBM Watson in a head-to-head comparison.

Re:infer is building a market category that it calls “Conversational Intelligence as a Service,” described as helping enterprises to create business value from business conversations. In simpler terms, the company’s AI platform can convert any unstructured text into structured data for action and analysis.

The company competes in the conversational AI space, using machine learning (ML) and natural language processing (NLP) technology to read and interpret the text found in emails, text messages, and chats. Conversational AI technology is best-known for powering those ubiquitous chatbots, the ones that triage incoming customer service requests to classify, prioritize, and route each request with little or no human intervention.

Pushing beyond that use case, Re:infer is finding success with other use cases such as insurance underwriting and post-trade operations at investment companies. The company leverages its ability to provide in-depth analytics on vast amounts of unstructured text and help its customers to discover process inefficiencies and opportunities in the data.

Re:infer has raised $8 million to date, with the last round closed in 2020.

The Technology

Re:infer created a cloud-based offering that ingests text from various enterprise sources such as email servers, chat servers, file shares, or CRM systems into a data store where proprietary ML algorithms segment and clean up the data in preparation for the next step. The system connects to existing applications and data stacks via an API. Python, JavaScript, Bash, and Response are all currently available. The system also complies with ISO 27001:2013 (security management).

Deep learning sentence models crawl through the data store to extract semantics. Meanwhile, new incoming data is constantly searched for new intents by Re:infer’s unsupervised learning models, which analyze the data and identify root causes of inefficiencies or problems.

The results are presented to subject matter experts in a visual, no-code user interface. The active learning engine then trains the supervised models, which will be deployed in production after being validated through a closed feedback loop that retrains the models. At the end of the process, the results are shared with other enterprise systems such as RPA and information managers. (Figure 1 illustrates the process.)

To break this down further, the Re:infer system is on the lookout for the “intent” of a communication; or to put it another way, the “why” in a conversation. Is this a complaint? Is it request for a service? And so on. The focus on extracting intent is the most important thing to understand about Re:infer. There are countless alternative NLP-based systems available that can tell you “what” something is, but few are able to take the cognitive leap to the “why,” the intent. Similarly in the burgeoning world of task and process mining, we have advanced tools that can detail what and how a process or work task is undertaken, but the intent, the why, is left for us to figure out.

Re:infer does all this through mining and correlating data sources and using advanced AI techniques. However, finding intent is remarkably difficult to do. Re:infer has built in some sophisticated Human in the Loop (HITL) capabilities to both monitor and continuously train the system and to balance both supervised and unsupervised learning in the system. Architecturally there are echoes of modern enterprise search here, but rather than indexing, Re:infer parses the incoming data at an earlier stage.

In practical terms Re:infer can be used in many different scenarios, but the simplest to grasp is for automating transactions. In this example, one could use Re:infer to analyze past emails, messages, or other communications related to a specific process or set of processes, such as onboarding new customers. The results of Re:infer’s analysis could then be used to improve or radically rethink the existing process steps and to automate multiple activities that are currently done manually by humans.

Figure 1
Conversational Intelligence as a Service

Our Opinion

Automating the analysis of the intent of communications and messages has immense potential to add a missing piece to the enterprise automation jigsaw puzzle. As already stated, mining human communications and recognizing the intent is awfully hard. As such, even though Re:infer’s tech is very sophisticated, each organization using it will need to commit to building its own specific taxonomies, map data flows and applications, and then not only train the system, but continuously monitor it. In our analysis, that is a good thing: AI is too often overpromised and as such underdelivers. Re:infer is realistic about the work needed to optimize its system, but if that work is done it can deliver a lot of value, and quickly. The company is not the first to try this approach, but in our analysis it appears to be setting a new benchmark.

Advice to Buyers

Any company with substantial amounts of conversational and textual data should consider adding Re:infer to its shortlist of vendors to evaluate, particularly if the business process involves customer interactions. Furthermore, for those embarking on genuine work transformation projects, combining Re:infer with task and process mining tools could well be worth investigating as it holds the potential to radically improve and accelerate both the initial business analysis and the ongoing monitoring of reengineered activities. Selecting a relatively new company to store and analyze one’s mission-critical data can be a risky proposition for an enterprise. However, the backing of UCL and other investors signifies that Re:infer should have the resources to grow and serve its customer base well into
the future.


SOAR Analysis

Strengths

  • Ability to extract intent from communications
  • Deep NLP and machine learning expertise

Aspirations

  • Become the bridge to convert all messages into actionable structured data
  • Create a global platform for textual data understanding

Opportunities

  • Build out more industry-specific solutions
  • Acquisition candidate for an enterprise software company

Results

  • Blue-chip customer base
  • $8 million raised to date

Attribution-NonCommercial-NoDerivatives 4.0 International
CC BY-NC-ND 4.0 license