On June 22, 2016, machine translation provider, Lingua Custodia, announced a capital increase of EUR 0.625m. Slator spoke to Lingua Custodia CEO and co-founder Olivier Debeugny for details.
According to Debeugny, the company, which currently employs eight staff in its Paris office, plans to use the funds raised for “the acceleration of our sales operations in France, in Europe and then in Asia.” The company’s product, called Verto, is a customized machine translation tool based on open-source Moses technology.
Debeugny disclosed Lingua Custodia has a total of 50 shareholders to date, with the founders and their families remaining as the largest investors. He declined to disclose revenues but said, “we anticipate to reach profitability in early 2018.”
Citing their 2015 goal of reaching the threshold of 50 major clients by 2019, Debeugny said they are on track with their target and “should have established contracts with 10 major clients by the end of 2016.”
With Linguagloss and TMS-platform Wordbee, Lingua Custodia has signed up two clients that Debeugny also considers important technical and commercial partners. “Our engines are connected to Wordbee via our API and accessible to all Wordbee clients,” he pointed out. For Linguagloss, Lingua Custodia customized an MT engine for German-English financial reporting, a project that led to them to start working on a joint post-editing solution for the LSP market.
When asked about their pricing model, Debeugny said they have volume-based pricing with a price per word and “depending on the nature of the request, a setup fee or a minimum monthly fee.” Debeugny claimed they also regularly retrain their clients’ customized engines “even as they use it.”
Lingua Custodia was founded in September 2011 by Olivier Debeugny and Nicholas Jeans who, from their experience as finance professionals, both saw a market for enabling quick turnaround of financial document translations.
Asked if that market was big enough to make a compelling business case, Debeugny replied confidently saying the space is “fairly large in our opinion.” He went on to list the various units, departments, and teams within banks, brokers, law firms, insurance providers, and other finance-related companies that deal with financial documents.
A highly regulated and sensitive space such as finance, insurance, and law might not be the obvious market for a machine translation developer to target. But it is in the sensitivity and confidentiality of documents that Debeugny sees an opportunity. When a document is deemed too sensitive to send an LSP, he sees an opening for machine translation.
Debeugny also pitches time savings, highlighting the fact that finance professionals typically review what comes back from LSPs anyway. So rather than waiting for the LSP, they have MT do the translation, which then gets post-edited. As for types of documents that lend itself to this approach, Debeugny lists urgent communication, RFPs, and client presentations, among others.
Debeugny estimates the market size for his MT solution to be around 5% of all financial translation work that is outsourced to external LSPs.
“We anticipate to reach profitability in early 2018”—Olivier Debeugny, CEO of Lingua Custodia
Automation and time savings should sound promising to a 25-year-old investment bank associate working overnight to finish an M&A pitch book or a senior business developer at an asset management firm submitting a 200-page RFP response to a major foreign pension fund.
What will be hard is to penetrate the multilayered decision-making structures of global finance and introduce automation in a very conservative sector.
As for their take on the recent buzz around neural machine translation, Debeugny said they look at all new technological development as opportunities instead of threats. “We are currently looking at strengthening our relationship with a CNRS Laboratory (French National Centre for Scientific Research) to investigate, among other things, neural machine translation,” he said.
But Debeugny maintained Lingua Custodia’s solution is more about the quality of translation than anything else. According to him, training an old technology engine with higher-quality data will deliver better results than a newer technology engine with average quality data.
Note: An earlier version of this story rounded the funding to EUR 0.6m. The exact amount is EUR 0.625m