Dasha.AI — robot-caller

JoshPitek • 7 November 2019

Dasha.AI is a call center automation service that simulates human speech using artificial intelligence.

To create a lively conversation, the robot uses parasitic words, interrupts the interlocutor, assent and, if necessary, is distracted from the main topic of the conversation.

If Dasha is interrupted in response, the robot will listen to the interlocutor and build a further dialogue based on the context.

The company was founded in 2018 by cousins ​​Vladislav Chernyshov and Alexander Zaitsev. Until the spring of 2019, they developed the service at their own expense, and in March, the American RTP Ventures fund and the European RTP Global (associated with billionaire Leonid Boguslavsky) invested $ 2 million in the company.

The startup is planning to spend money on expanding the team - now there are 50 people in it, including Alexander Dyakonov, doctor of physical and mathematical sciences and the winner of the competition for AI specialists from the Kaggle community.

In addition, the company intends to increase its presence in the US market - now several American companies are testing Dasha.AI technology. Chernyshov does not disclose their names, but notes that they are on the Fortune 1000 list.

Today the company has about 18 customers. In Russia, Dasha.AI technologies used 2GIS, but later developed a similar solution, noted the head of the cartographic service Alexander Sysoev.

And the operating director of Modulbank Anton Lipatov told Kommersant that in the first month of operation, the Dasha.AI robot interviewed four times as many customers as compared to call center operators. The bank introduced the technology in the summer of 2018.

In addition, the team is working on a platform for self-modeling of conversations, with which ordinary users can create virtual assistants like Google Duplex (for now, Dasha.AI works individually with each client).

The company wants user robots to not only perform the same tasks, like most virtual assistants, but to choose the most optimal behavior in a given situation.


Why is Dasha “smarter” than other voice assistants?
Dasha.AI employees manually prescribe a scenario in which the robot will communicate with customers.

For scoring, the technology of hybrid speech synthesis is used - first, the announcer records phrases, phrases, words or syllables, which are then “glued” into sentences with suitable intonation using digital signal processing algorithms.

During a conversation, Dasha recognizes the meaning of what was said using the machine learning algorithm and “understands” which branch of the dialogue to build the conversation on.

The robot can inform the customer of the online store that his order has arrived at the post office, and if he is interested in which one, Dasha will prompt his number or send an SMS with the address.

If the interlocutor says things incomprehensible to the robot, Dasha will transfer the conversation to a call center employee. If the robot understands that there is nobody to connect to the conversation, then it apologizes for poor communication and hangs up. Also, the robot can call back after the agreed time.

For voice recognition, the APIs from Google and Yandex are used, as well as the own developments of the Dasha.AI team. The technology of the company is based on the dissertation of Chernyshov and the technical director of the company, Ilya Stupakov.

One of the problems they solved was the delay during the conversation.

The robot maintains a lively dialogue: interrupts if necessary, can reformulate the same question if a person does not understand it, and uses parasitic words like “um”, “aha” and others


Story of creating Dasha.AI

In 2007, Chernyshov took part in the first competition for Android developers from Google, and also participated in the ACM ICPC programming world championship. But he did not become a programmer and went into business - he opened his own construction company.

He does not disclose her turnover, but notes that by 2013 more than 150 employees worked in it. However, the 2014 crisis did not survive the business and went bankrupt.

However, this experience helped Chernyshov to see the problem: it was difficult for construction contractors to plan a budget. Then the entrepreneur decided to create "Uber for builders" - an application with which the user could draw up a cost plan and immediately get an estimate.

Chernyshov assembled the first prototype in 2016. The iOS version was written in Swift, for Android - in Java, and the backend - in Node.js and C #. He worked on the application with his cousin Alexander Zaitsev.

Chernyshov wanted to make the service more convenient: so that the user could list the amount of work by voice (for example: “change two doors, replace the mixer and lay a 3 m² of laminate in the corridor”) and find out its cost. To do this, he built in natural language processing (NLP) technology, and the application learned to synthesize speech.

Despite its functionality, the application did not become popular. But Chernyshov liked the recognition and speech synthesis function so much that he made a new product based on it - Dasha.AI

Команда Dasha.AI


The robot sells stairs

Alexander Zaitsev helped his father develop a family company that was engaged in woodworking in Novosibirsk, in particular - the production of stairs for cottages.

By that time, the local market seemed to Zaitsev cramped, and he wanted to find dealers in other regions. However, he did not have the resources to increase the staff of sales managers.

Then in November 2016, he asked Chernyshov to "teach" the application cold calls in order to look for builders and contractors for cooperation. Chernyshov spent three months creating the program, and in January 2017 the service made the first call.

The phone numbers of the masters who were engaged in the installation of stairs were loaded into the database, and the robot phoned them. He made a presentation of the company and talked about the conditions of cooperation.

If the interlocutor was interested, the robot asked him to send an SMS with a phone number and e-mail, and in response sent an agreement and information on cooperation. Then, managers contacted the dealer to close the deal.


First client in USA

After a successful prototype test, Chernyshov and Zaitsev decided that the service should be further developed, and in May 2017 they submitted an application to the Y Combinator accelerator. They were invited to an in-person interview in the United States, but they did not have money to travel. Nevertheless, the brothers managed to borrow about $ 5,000 from acquaintances.

The prototype was in Russian, so I had to translate it into English already on the spot. It seemed to the entrepreneurs that the interview was successful, but they were refused - Y Combinator manager Michael Sable and accelerator partners did not understand how Dasha.AI could make money in the future.

Despite the failure, the brothers were in no hurry to return home. Chernyshov sent a prototype of Dasha.AI to friends from the United States and asked them to tell friends about the service. The idea was a success, and the next day, entrepreneurs signed the first contract with a network of dental offices in San Francisco. The amount of the contract at that time did not exceed $ 5000.

In early June 2017, the brothers returned to Russia and began to look for customers, one of the first was the 2GIS service. According to the entrepreneur, many customers came thanks to word of mouth.


Dasha.AI monetization and startup development today

Today, a startup is implementing pilot projects for large customers. According to Chernyshov, this helps not only to “cones”, but also to accumulate data - conversation records - for training neural networks.

Startup works with companies from banking, finance, insurance and technology. Among the main offers of Dasha.AI:

  • NPS polls.
  • Delivery verification.
  • Additional sales.
  • Appointment, cancellation and rescheduling of doctor's appointments.

Replacing first line operators on incoming calls.
According to Chernyshov, all clients expand the number of cases within three months. The company’s record for today is eight cases per client.

After starting the project, the client pays a monthly amount according to the subscription model. The price is calculated individually and depends on the options chosen by the client.

  • The first software installation and customization for business processes costs from 500 thousand to 1 million rubles.
  • Paying minutes for calls. Depending on the volume of calls, a minute costs from 4.17 rubles to 6.67 rubles.
  • There is a one-time payment that the client pays for additional configuration.

Before starting the robot, Dasha.AI and the client select targets. One of the most frequent is the percentage of unsuccessfully completed conversations (conversation error rate - CER), when the robot either answers out of place or gives itself away. According to Chernyshov, a satisfactory result is no more than 5% failures when talking to the robot, and the team record is 0.8% –0.5% failures.

The Plan

Chernyshov wants to turn Dasha.AI into a platform with which users can create virtual assistants and any other conversational voice models.

For example, a part buyer can automate his work: an assistant will call, ask for prices, specifications and buy suitable items.

And if there will also be a robot on the supplier’s side, then the work will be fully automated. Chernyshov calls such communication natural language.

The developer says that using speech synthesis, you can automate many routine things. For example, add a voice interface to Amazon’s pedestals so that users do not have to type a password each time. In this case, the user will be able to ask the device to give the package, and the pedestal will identify it by voice and determine which cell the order is in - and open it.


Result: raised: $2000000

Теги: AI Sales Startups Business



Too few case studies?

Sign up to all case studies.

Are you love case studies?