JP Morgan Investment Insulates Limeglass Start-Up Amid Coronavirus

Research start-up Limeglass is riding out the coronavirus storm, thanks to investment from JP Morgan.

Limeglass announced late last year that the bank had acquired an undisclosed stake, and the start-up has since used the funding to hire more people and improve its product, which filters financial research documents to find the most relevant information.

“The investment was a game-changer for us. Clearly, it is helping us through [the coronavirus pandemic], but even before that, we were able to double our company size. And we are planning more investment in technologies and staff and expertise over time,” says Simon Gregory, co-founder and CTO at Limeglass. “JP Morgan thinks of us as a strategic investment and wants to help us bring something to market.”

Gregory and Rowland Park, the company’s CEO, founded Limeglass in 2016. They received early stage funding from angel investors to develop the platform for financial research, as well as a grant from InnovateUK. In 2018, the company joined JP Morgan’s In-Residence fintech incubator program, and launched the platform in 2019. In late 2019, it announced the investment from the bank.

Limeglass has used the funds to grow its team from seven direct employees to 17, plus additional contracted service staff and advisers hired on an ad hoc basis. JP Morgan is already a client, as is a global forecasting organization, and the company is currently speaking to a number of firms on the buy side and sell side to begin using the platform, Gregory says.

He says he has not noticed a diminution of interest from these prospective clients since the coronavirus pandemic began; indeed, he says, people at the larger banks especially generally have fewer demands on their time now, so he is able to schedule calls more easily.  

Bulky Research

Limeglass is the third company Park has co-founded. He and Gregory worked together at another of his start-ups, 4CAST, which produced short research snippets, distributed intra-day, for trading rooms at banks. 4CAST’s clients would tell Gregory that they read only a fraction of the research their own organizations produced, because there was so much of it. Research reports could be four pages, or 200 pages; they could be full of graphs and tables, and delivered via email as a PDF.

When 4CAST was sold in 2010 to a private equity firm, Gregory and Park began to think about how the information in these bulky research documents could be better distributed. Would it be possible to extract just one sentence from a document and deliver it in a curated fashion to individual users? The idea was sound, but they decided that a sentence was too short.

“We thought, well, inside these documents, how can we surface the bits that are relevant, and the most interesting to the people who use them? And we thought that maybe we could take sentences; but we realized that actually, a sentence out of context isn’t very useful, while paragraphs are a useful unit,” Gregory says.

“But still, if you want to understand the document, you need to retain some kind of context from the document—you can’t just rip out the paragraphs. There might be titles, there might be subtitles under the titles, and other things around the paragraph. And that is where we came in and built our idea,” he says.

The software-as-a-service platform Limeglass then built breaks down documents into constituent parts and tags each paragraph with relevant concepts. This works a bit like a Google search, in that if a user types in a search term, the solution would return results not only for the term searched, but also related concepts. Users can feed their own research reports into Limeglass, or install its APIs so that their own systems can run the process. The solution can track which parts of the reports are most useful. And users can receive email alerts as new reports are published.  

This is intended to provide a macro view of a market, and a granular level of detail for a topic, and to deliver it to individuals in a personalized way. The platform uses machine learning to keep improving.

The comparison to Google, however, and the artificial intelligence technology, only go so far, Gregory says. A crucial human element is also needed.

“If you take the example of Google, people are used to that interaction. You search for something, you might not get back what you are after, and you would accept that,” he says. “But when you are highlighting a paragraph and saying, This is what you are looking for, you get judged on very different criteria of accuracy. … And that is why we have integrated human input into the process.”

The human component is six research analysts who continually update the Limeglass taxonomy and ontology. They are always “looking for new words that come up, to build our taxonomy. We cover things like macroeconomics, which is very difficult because there are no tickers, no currency code or anything. It’s concepts like ‘inflation,’ which includes many different words. Some of these concepts could have 40 or 50 different synonyms that we have to include in the taxonomy. So we are constantly growing and building it,” Gregory says.