SINGAPORE (Jan 22): Sinuhe Arroyo is excited about having a new office to decorate. Arroyo, who is from Spain, founded his artificial intelligence (AI) start-up Taiger while living on the second storey of a farmhouse in the Australian Alps, surrounded by cows. Now, he would prefer to be surrounded by some graffiti-covered walls instead.
Arroyo’s posher surroundings — he now has an office in Singapore’s Chinatown — are a reflection of the progress his company has made. Taiger has developed three AI-powered software programs that can interpret and process text and speech, helping organisations cut down on cost and staff hours spent on customer service and document handling.
Taiger’s semantic machines are used in various industries and companies, including a major bank in the US, government agencies in Singapore and a global insurance company. But as Taiger grows, Arroyo says he wants to focus on the banking industry as it faces pressure to go digital. “We feel the biggest demand [is coming] from this industry because they see the value in what we do,” Arroyo says. “It makes sense to go for the lowest-hanging fruit first.”
Major financial firms are in fact leading the charge in the adoption of AI, for fear of losing their customers to fintech start-ups. Three in four business leaders in the industry believe AI will give them an edge over competitors, according to PwC’s 2017 Digital IQ Survey. About 66% of them also expect to make substantial investments in AI in the next three years.
Oversea-Chinese Banking Corp worked with start-up CogniCor to create a virtual assistant called Emma, which has helped the bank sign $70 million in home loans. On Jan 9, Canada’s largest bank, Toronto-Dominion Bank, acquired AI start-up Layer 6, which is capable of analysing data to predict customers’ needs.
Anticipating that demand will grow, Arroyo is expanding aggressively. Last month, he and his team of more than 50 moved from the start-up incubation park in One North into a 400 sq m space on Temple Street. Taiger also has offices in Madrid, Hong Kong and San Francisco. Over the next 12 months, Arroyo plans to open offices in New York, London and Sao Paulo, as well as a software factory in Vietnam. He projects Taiger’s headcount to double in less than a year.
“Revenue has been growing rapidly,” he says. “We are not yet profitable, but we do not want to be. If you are [profitable at such an early stage], then you are likely not growing fast enough, or you do not have a market.”
According to Taiger’s website, the company has more than 20 clients. Contract sizes vary between US$100,000 ($132,285) and a few million US dollars. Customers pay based on the number of tasks needed. Last year, the nine-year-old start-up secured US$5.8 million in a Series A funding round led by Tembusu ICT Fund I and SGInnovate. Five years before that, Arroyo raised €400,000 from friends and family.
What’s in a word?
Taiger’s technology is based partly on the use of natural language processing technology, a form of AI that interprets the structure of words, as well as knowledge representation, a field of AI that has never been successfully implemented until recently. “I don’t need a lot of data [unlike machine learning-based technology],” Arroyo says. “I need knowledge of how a particular task is performed and how a document is read, and I teach my system to do it.”
The company’s earliest product was iSearch, a search engine that does not rely on keywords. Arroyo explains that different people may use different terms to search for the same material. “If you search for documents based on keywords, you get a lot of noise. Our technology [churns out] matches based on the meaning of the word, which produces more accurate results.”
These same principles power Taiger’s iConverse software, which is a chatbot that works on major messaging platforms such as WhatsApp and WeChat. Singapore’s HDB uses iConverse to answer queries on new home ownership and parking issues. While Arroyo refuses to disclose how the machine works, he has previously said that iConverse passed the Turing test — a process designed 64 years ago to evaluate if a machine can convince enough people it is human.
A third product, iMatch, was launched about three years ago. It interprets and processes documents. JPMorgan Chase rolled out a similar program last year that can process commercial loan agreements, saving the bank 360,000 hours of work each year. Arroyo says iMatch has helped a European bank cut its costs by 85%, shortening processing times from weeks to a matter of minutes.
“The minimum key performance indicator we give clients is 80% accuracy [for our software],” Arroyo says. “Typically, clients would state their expectations to read a certain set of documents, and how many data points they want the machine to process. Then, I’ll teach my system to do so.” iMatch is currently employed for Know Your Customer procedures, credit card assessments, loan on-boarding and anti-money laundering checks.
Arroyo says Taiger tweaks its software to fit each client’s requirement. It can take the technology up to five months to learn a new language while a sale process may take up to nine months.
Learning journey
Arroyo was formerly the corporate development manager of Austrian computer firm Phion, which was acquired by California’s Barracuda Networks in 2010. When he left his job to “do his own thing”, his mind went back to his doctoral research at the University of Innsbruck, Austria. “Semantic technology was my strength, so it [only made sense] that I [tried] to build a company around that,” Arroyo says.
When Arroyo founded Taiger in 2009, he thought he had a jaw-dropping technology and would have no trouble selling it. He was wrong. “We didn’t know how to engage [with customers],” he says.
It was only in Taiger’s third year that Arroyo began to see some success. “We learned. Instead of talking about how brilliant the technology was, we started talking about KPIs [key performance indicators],” he says. The start-up got its first big break when it clinched a deal with media company Sony.
As Taiger grows, it may begin to hit the limitations of AI and natural language processing technology. Gao Cong, associate professor at Nanyang Technological University’s school of computer science and engineering, says it is difficult to program machines to understand natural language.
“Relatively simple tasks such as text classification, which do not need an in-depth understanding of the text, can be handled well. But complicated tasks such as full comprehension of articles or open-domain [queries] are still beyond the current state of [AI technology capability],” he says.
However, Taiger’s investors seem confident. “Many AI companies are built only on machine learning, which is a data-heavy statistical methodology and does not achieve good results for language-related AI solutions,” says Brijesh Pande, founder and managing partner of Tembusu ICT Fund I. “Taiger is quite unique in that they have expertise in multiple branches of AI: [areas such as] knowledge representation and reasoning.”
Arroyo, who is already decorating his office walls in his mind’s eye, certainly thinks so too.
This article appeared in Issue 814 (Jan 22) of The Edge Singapore.
Subscribe to The Edge at https://www.theedgesingapore.com/subscribe
