A sad person will complain and, at most, will not shop a second time. When someone is angry – they can move to action: discourage others or go to court. So Damian Grimling’s company taught artificial intelligence to define real emotions. Now it is helping corporations recognize customers’ feelings, and stock market players how to invest…read more in Forbes 05-2023.
Translation
Sentimenti. When a customer gets into a frenzy
Publication date: 27.04.2023, 08:06
NATALIA CHUDZYŃSKA-STĘPIEŃ
The algorithm helps corporations recognize customers’ feelings and stock market players tell them how to invest. Damian Grimling’s company has taught artificial intelligence to define real emotions, and has plenty of ideas on how to use this sensitivity of intelligence.
He once had a long-distance relationship. When instant messaging caused more misunderstandings, a business idea was born. That’s how Damian Grimling founded Sentimenti, which deals with reading emotions in texts.
Catherine the Machine, team member. “A hard head, an analyst by vocation, a researcher of human emotions, she weighs, measures, prompts.” Damian Grimling, founder of Sentimenti, still refers to it as “artificial intelligence with real emotions.” It’s a technology designed to examine texts for the feelings latent in authors. It has already been used by telecoms, banks, media houses, advertising agencies and even Allegro, Żabka or IKEA, as well as government offices and political parties.
– For business, we primarily study emotions in customer comments. So far, the tools available on the market are based only on sentiment research: whether it is positive or negative. This is a very shallow measure. We go a step further, as we identify eight emotions and the so-called emotional arousal. In doing so, we don’t give zero-one results, but indicate the percentage intensity,” explains Damian Grimling, CEO of Sentimenti.
This is an important difference. Because, for example, both anger and sadness fall under negative sentiment. But while a sad person will complain and, at most, won’t shop a second time, use a service, come to the office or vote for a politician – an angry person can move to action: promote offensive opinions, discourage others, file a complaint or go to court. Distinguishing between these emotions tells which customer is worth dealing with first. The same is true of positive emotions, such as trust and joy. It is worth focusing on joy to turn it into trust, the most desirable feeling in marketing.
Sentimenti makes money by selling reports and access to a tool into which marketers enter content themselves to check. Its financial performance is still unstable, but diversification of revenue is expected to change that. The company’s best year was 2021, when it recorded more than USD 160 thousands in profit on USD 200 thousands in revenue. Last year closed without a profit, and revenues fell to USD 100 thousands.
A few months ago, the company launched an algorithm that makes suggestions on how to play the stock market based on investors’ emotions. Customers get access to it by paying a monthly subscription of about USD 49. For now, private investors are the clients, but talks are underway with institutions about launching an algorithmic investment fund, or so-called quant fund. Sentimenti made it to the list of nominees for this year’s TradingTech Insight Awards USA in the categories: “Best Matching Engine for Cryptocurrency Trading Venue” and “Best Trading Analytics Platform.” Previous winners include Bloomberg and Moody’s.
– Catherine the Machine can measure the level of emotion to the fifth decimal place. We decided to use this in such a precise industry as investments. We started with Bitcoin, as the least anchored in the real economy, and strongly susceptible to emotions,” says Damian Grimling.
Sentimenti extracted mentions of bitcoin from Twitter for two years – more than 14 million posts came out. In addition, bitcoin quotes from the Binance exchange – also a powerful database. Artificial intelligence has shown that it is possible to find a strong correlation between the emotions contained in tweets and the quotation of the cryptocurrency: the effectiveness of the prediction reaches 80 percent. In 2022, the bot launched by the company earned about 40 percent, while bitcoin lost more than 60 percent. If the bot had run in 2021, it would have been more than 600 percent in profit with about 60 percent increase in the quotation of the cryptocurrency. In the future, the results could be even better. For now, the algorithm, based on Twitter and Reddit posts, says that in the next hour the price will change by a given value on average, and the bot makes a final decision to buy or sell based on that. Ultimately, this prediction period is to be shorter, which will create more chances to win over the market, and in addition, industry online forums are to be plugged into the analysis.
The Sentimenti tool allows you to identify not only sentiment, but also eight exact emotions in comments. This allows the company to catch the most annoyed customers and deal with them first.
The Sentimenti tool allows you to identify not only sentiment, but also eight exact emotions in comments. This allows the company to catch the most annoyed customers and deal with them first.
Sentimenti’s programmers have also done analysis for the forex market and for Amazon, Apple stocks and dozens of companies on the Warsaw Stock Exchange – and “the model is highly effective.” The next step is to launch a bot for the most popular stocks and currency pairs.
– Interest in our tool for marketing tasks is growing, but not by leaps and bounds for now. Since our technology can have a very broad spectrum of applications, we are eager to enter new areas,” comments Damian Grimling.
He has become a strategic partner of Statista, one of the world’s largest statistical research agencies, for which he stands for so-called emotive data. It recently began working with “one of the world’s largest” platforms providing chatbots for customer service. During a conversation between an operator and a customer, the emotions of both sides of the conversation are monitored in real time. On the one hand, the operator can react more effectively in the conversation and take appropriate action when he gets a message, for example: “the customer’s anger level is deviating from average, agitation is increasing.” On the other – the operator’s supervisor can assess how he has handled the customer’s emotions and his own.
– Live chat is a direction we definitely want to develop. Likewise, we plan to offer HR departments a tool that will allow us to survey employee satisfaction levels while doing remote work, when we are not able to see what we would see in a live contact in the office, the CEO announces. – Here, of course, there are a lot of sensitive privacy issues that need to be worked out, but we are able to monitor any texts that are sent from a given computer via mail or instant messaging. Without going into the content of the texts themselves, we can identify the mood of the employee,” he explains.
To implement the new plans, he will soon apply for funding from the NCBiR from a new round of the so-called fast track. In the previous round, in 2018, he secured USD 1 million. He then added his USD 360 thousands and set out to build the model. Along the way, it was subsidized to the tune of USD 240 thousands by the SpeedUp fund. Today, the company employs nearly 20 people.
The idea for Sentimenti came from… long-distance love. When Damian Grimling and his then-partner often “talked” via instant messaging several years ago, there were times when misunderstandings arose from misreading intentions and emotions. And this despite the fact that the partner was a psychotherapist and psychiatrist, and Damian Grimling – who had previously graduated from a merchant school in Germany – is a certified coach. It was then that the idea for a tool that prompts what the person on the other side of the chat is feeling was born.
He was well versed in getting EU grants, because that was what the company he previously ran specialized in. He quickly raised capital and began working with two Polish universities, which conducted advanced research for three years. A sample of 22,000 people was extracted to analyze the content provided (a total of 30,000 texts of various lengths) and identify specific emotions in them. The data went to an artificial intelligence and machine learning process, which dispersed it throughout Polish and then into 17 other languages.
– This is no longer done today: you don’t conduct research on huge groups of people, you count on models. Our investment has paid off: the model reads emotions with the accuracy of an average person,” Damian Grimling argues.
There are quite a few companies in the Polish market and globally that monitor online mentions of brands, but still determine their sentiment zero-one. Among the largest in our country, for example, is Brand24, and globally, for example, Meltwater. A tool developed by IBM’s Dr. Watson supercomputer focuses on specific emotions.
– The quality is clearly inferior to ours. IBM breaks down a sentence into individual words and analyzes them for emotion, then extracts an average for the sentence. Our AI understands context, irony, sarcasm, goes deep into the text comprehension analysis and only then reads and measures emotions,” comments Damian Grimling.
A recent Deloitte study, “Digital Customer Service Excellence 2022,” shows that 83 percent of companies measure customer satisfaction, with 49 percent underutilizing the results to drive change and improvement. Knowing customer emotions can make a big difference in better prioritizing such initiatives.
– Regardless of what metric we use to study customer satisfaction, only effective use of feedback for changes within the organization can positively affect the satisfaction and loyalty of our customers, and knowing in what context positive or negative emotions occur is invaluable in this process,” says Alexander Pruzinski, partner, Head of Customer Service Excellence at Deloitte Digital CE. – Above all, customer emotion analysis needs to be approached thoughtfully, starting with opportunities for feedback analysis, quality control or proactive monitoring of customer satisfaction trends. One of the most important challenges facing companies implementing such solutions today is not the technology, but the preparation of processes in the organization. That’s why it’s worth thinking about this in parallel with technology implementation,” he points out.
Testing the level of specific emotions is not even offered by Amazon or Google in their cloud-based solutions for business. The question is whether they are convinced that it doesn’t make sense because business will stay with simple solutions, or whether they haven’t “figured it out” yet. Sentimenti made an appearance at the recent Google TechTalk. Who knows, he may have inspired positive sentiment or even confidence in the biggest Big Techs.
This text is from the Forbes 5/2023 issue