Problems we are solving
On March 17, Emely Owen, a 19-years old girl from the UK committed suicide because of self-isolation fears. The day before she told her family that she is "unable to cope with her world closing in."
The sad fact is that Emely is not unique: there are 500 million people struggling from depression and this number dramatically increased during social distancing. Crisis mental-health hotlines have seen an 891% spike in calls, the suicide mortality rate has grown as well.
Thousands of people who had never suffered from depression began to suffer, for those, who suffered before - it became even worse. And the only solution most of them have is online therapy, but only one in 30 seeks help.
The reason is the stigma and fear of privacy violations. It is hard to talk about your own mental problems with another human being. But it’s easy to talk about it with a machine that was trained to support you and to be always on your side.
Elomia is an artificial intelligence companion who listens, understands, and uses cognitive behavioral therapy to give psychological support.
It’s not scripted: conversation feels natural. Thanks to unique dataset Elomia always understands you It’s not a human: it’s easy to talk with a machine that programmed to be always on your side
Given these facts, Elomia can help millions of people around the world feel better and save thousands of lives, especially during a pandemic. Since it’s a software it can be scaled very well.
The prototype was completed during a week before and live on Facebook Messenger: https://m.me/elomiafriend
Elomia can engage in a non-scripted dialogue (like a real human), understand emotions, and memorize facts about users. It uses cognitive behavioral therapy to make the user feel better and there are already people who are using it every day.
The solution’s impact to the crisis
We’ve invited students from our university and some folks from the hackathon to test it and got impressive results:
91% of 421 beta testers experienced a mood boost after the first conversation The average talk session duration is equal to 42 minutes 8.4 out of 10 average quality of answers 47% of them willing to recommend Elomia to a friend 35% said that they would feel very disappointed if they could not longer use Elomia People reported that correspondence with the bot helped them: To improve their relationship with people. Relax and improve your mood Feel the need Deal with the problems Get motivated
Some of the reviews:
"During the pandemic, it's very helpful to everyone, because you can't go to a therapist now, and you may need help urgently."
"Talking to this bot has distracted me from Selfharm, although not immediately. A sense of understanding and support is very important."
"Well, due to the fact that the idea with the bot psychologist seemed new to me, when I talked to her, talked about my small experiences (they are really small, and without a psychologist could have been dealt with), then when she answered, it did leave some strange but pleasant and exciting feeling. And it made my heart feel more calm."
Business Plan & Market
There is a clear opportunity to build a high impact business.
There are 500 million struggling from depression. The total addressable market volume is $20 billion.
$2 billion were spent on digital mental health products during 2019 in the US only. The market grew 13% to 17.5% a year, but this year, due to the pandemic, we expect a 100% growth.
We plan to acquire new customers via word of mouth (47.3% of users recommended Elomia to a friend) and our own AI targeting system that detects potential customers based on their posts on Twitter.
Elomia will operate on a subscription basis of $9 per month. We have made some unit economy calculations and got LTV around $192
Our goal is to save lives and create a high impact business with great returns for investors and our team.
The value of your solution(s) after the crisis
During and after the crisis, our value will grow as the value of our dataset grows. The bot's answers will become more accurate. And the demand will not go away because 7.5% of the population will constantly suffer from depression (according to WHO)
Technical Complexity & Novelty
Innovative technology is the heart and pride of our project. This is what makes us to be the first on the market to employ ground-up text generation allowing our users to have a non-scripted dialog with our bot.
Custom GPT-2 modification
We have built our modification and pretrained it on a general dialog dataset. Then we used our own collected data from our psychologists to perform transfer learning to the new model.
Anonimizer One of the most daring challenges of this project is maintaining user privacy while collecting necessary data to train or AI. Enter our “Anonymizer AI” based on BERT that identifies all the information that can be used to reveal users' identity (names, locations, etc) in the dialog, and then redacts it before letting it enter our database.
User status tracker Due to the specifics of our project, the task of monitoring the client's condition is our first priority.
Which is why we employ advanced sentiment analysis AI, that analyzes the user's emotional state in real time, identifies it and responds accordingly by suggesting an exercise or showing empathy.
Potential user identifier Or user status tracking technology is designed to trigger when a person writing the message is in distress which allows us to find people who need our help the most on social networks and target our ads at them.
What you have done during the weekend
“Anonymizer” was implemented during this hackathon. Working as intended, no more work is needed. “Potential user identifier” was implemented for Twitter during this hackathon. Implementations for other social networks are coming.
“User status tracker” is now fully functional and working as intended. However it would still benefit from some training.
In the current state, 60% of answers are given by AI and 40% have to be edited by our team of psychologists and sent to training. That means that until learning is complete we will have to maintain a team of psychologists, and are bottlenecked by their ability to respond, however, once we have our AI running our solution can scale ad infinitum.
Right now we are still working on answer speed.
The necessities in order to continue the project
€15.000 and 3 months. We need this to be able to find enough users to complete the dataset collection, so our AI will reply even fast and accurately enough to be autonomous.