Sound Businesses: Profiles of companies and business models we are keeping an eye on.

David Llorente wasn’t happy. Day in and out his writers were grinding out repetitive online gaming news in their Tallinn, Estonia, office and no time was left for his main interest — measuring traffic to the website. There had to be a better way.

As managing director, he told his colleagues to stop writing and hammer out a template for the daily, run-of-the mill stories that could be automatically filled in by a computer with access to a stream of raw text and data.

They succeeded, and that template became the genesis of Narrativa, which claims to be the only company providing automated Arabic news using artificial intelligence.

Natural language generation, or NLG, is a computerized process of turning chunks of data like weather reports, financial reports and sports statistics into stories that sound like they have been written by humans. News agencies around the world including Associated Press, Bloomberg News and Dow Jones are publishing basic news stories created by programs, using templates and algorithms developed by humans.

Narrativa clients include Microsoft’s MSN news and a handful of Spanish-language news and sports sites like El Periodico and Sport. After several years of losses, the U.A.E.- and Spain-based startup turned its first profit this year as it expanded into government and finance four months ago.

Its revenue reached $1 million in 2018, with roughly 40% from government contracts, 30% from the financial industry and the rest from media and e-commerce, Llorente, Narrativa’s CEO, told Karma. Although media was the early adopter of NLG, it’s less profitable than the finance and government contracts. Llorente attributes this to media being slow and cautious with respect to implementing the technology. News organizations have been reluctant to replace people with machines, despite the cost-savings of Narrativa’s algorithms in data discovery and image recognition.  

“We generate over 220,000 pieces of news every week, so the journalist in EU feels very threatened by this,” said Llorente, adding that it only costs $8 to produce 1 million pieces of news. “They don’t go all in.”

Entering the U.S.

Llorente is building the business by focusing on Arabic language customers in the Persian Gulf Area.  In the Middle East, Narrativa’s customers include the Saudi Football Championship’s official streaming application, Dawri Plus, and the Middle East e-commerce website, Noon.

Llorente is building the business by focusing on Arabic language customers in the Persian Gulf Area.  In the Middle East, Narrativa’s customers include the Saudi Football Championship’s official streaming application, Dawri Plus, and the Middle East e-commerce website, Noon.

As the company builds in this less-competitive region, it’s gaining momentum for the U.S. market, which is Narrativa’s top priority this year.  

“We tried to look for a market that will help us grow to a position to enter the U.S.,” said Llorente, adding that financial industry and government contracts will be their main focus in the  the country.

While Llorente claims Narrativa to be the largest NLG vendor in terms of content volume, he says the firm is still in its growth stage.

Narrativa is considering an offer to be bought by a U.S. company for $35 million, Llorente said, as well a $5 million investment from a New York investor. He declined to name either the company or the investor.

The buyout or investment will help the company with its goal to reach and acquire U.S. clients.  He will decide in three-to-four months on a direction.

Llorente sees more opportunities in the U.S. than anywhere else in the world. First, both American corporations and governments are more eager to implement the technology, while European companies, he believes, are resistant. Second, unlike the fragmented European market, the U.S. market is more profitable as English and Spanish are the most-spoken languages.

NLG Gaining Acceptance

NLG’s acceptance has been growing over the past decade on promises of newsroom automation. Sellers and buyers see it as a way to help companies control costs, write stories faster with fewer errors than humans, and free up reporters from mundane stories like fast earnings reports and box scores so they can write in-depth journalism.

There were fewer than 100 people in the world working on the technology 10 years ago, said Ehud Reiter, NLG professor from University of Aberdeen in Scotland and chief scientist of NLG company Arria. Now a handful of mid-sized companies are pitching their wares globally.

Established firms like Automated Insights, Arria and AX Semantics provide NLG software, and customers need to manually prepare templates in order to generate narratives. The technologies behind those programs range from word templates to more advanced systems that dynamically create sentences.

Narrativa and its Chicago-based rival Narrative Science use algorithms to provide fully automated narratives, said Llorente. He says the algorithm approach is more scalable than older approaches such as templates or rules. Llorente said Narrativa’s technology is able to develop narratives for a new language in one week, while it takes up to three months using old approaches. In fact, Narrativa plans to begin providing French, Italian and Portuguese content in August.

“We decided to go with this approach, which was very risky, but now it’s paying off because we are able to generate new type of news maybe in three or four days,” said Llorente.

Improving Local News through NLG

NLG has the potential to improve local news, Arria’s Reiter said. Local news has been dwindling globally as local newspapers  suffer from falling advertising revenue. UK news agency Press Association was able to ramp up local news to 30,000 pieces every month by using Arria’s NLG technology. BBC now generates 100 stories on local hospitals every month by using advanced templates to process data released from the National Health Service.

Still, NLG faces a few brave new world challenges, the primary one being the accidental creation of fake stories and having systems to prevent that. Llonrente and Reiter agree that if there’s an error in raw data, NLG-powered media is in danger of generating false stories. Narrativa said it prevents this by being a  gatekeeper and only generating news based on data from reliable sources such as the government or established research groups.

Equally important, companies like Narrativa have to be sure AI serves humans with a clear and unbiased purpose, said Llorente. With the ability to easily generate huge amount of news, even if they are all true, NLG could easily distort reality by over-representing certain topic, he said.

“If (NLG) only focuses on positive news, even if it’s real, the picture you get is very ugly,” said Llorente.