Customer communication using AI: parlamind performs better than Google Dialogflow, Rasa and IBM Watson


AI Made in Germany achieves a higher hit rate than the global big players (PHOTO)

Berlin (ots) talking, Expert in customer communication using AI, is proud of what has been achieved: In the test, your AI does better than the AI ​​on which Google Dialogflow ES, Rasa and IBM Watson Assistant based. Expressed in numbers, this means: with an accuracy of 0.88 as with parlamind, only 12% of the incoming inquiries cannot be answered, with IBM Watson it is 16%, with Rasa and Google Dialogflow 18% (1).

Dr. Tina Klüwer, managing CTO and founder of parlamind translates the result into practice: “With 100,000 customer inquiries that reach a company via email or chat, our AI can process around 5,000 more correctly on average than the next best – Google Dialogflow and Rasa. This number can also be expressed in terms of time – that is valuable working time that customer service saves and can use for the benefit of inquiries for which human understanding and empathy are still required. “

© The  test results based on the F1 score are statistically more precise: parlamind’s F1 score is 0.92, followed by Google Dialogflow and Rasa with 0.87 and IBM Watson with 0.78 (1). So parlamind is only 0.08 points away from fully recognizing all existing contact reasons, while Google is 0.13 (2).

Test scenario: How well does Artificial Intelligence understand customer communication?

For the comparison, all systems were trained with the same learning examples and test data. © The  same queries were then sent to the systems and the results compared. Over five years of development and AI training finally pay off: parlamind achieves a better hit rate than the three global companies IBM Watson, Rasa and Google Dialogflow. More detailed information on the procedure and the results can be found in the “KI report: Intelligent contact reason recognition – quality analysis and comparison of providers 2021”.


 The  aim of the test was to find out how well all the systems in the test understand the content of customer queries correctly and assign them to one or none of 10 defined contact reasons,” explains Klüwer.


 The  contact reasons used by parlamind in the test are those inquiries that the expert for automated customer communication has identified as the top 10 concerns in customer service based on years of experience. To ensure comparability, the tested inquiries always contained only one or none of the 10 contact reasons. parlamind has been using this test method for years to improve its own AI.


 The  following contact reasons in customer service were tested:

  • Send the return slip
  • cancel order
  • What are the bank details?
  • Confirmation email is missing
  • When will my delivery take place?
  • Send invoice
  • Extend the return deadline
  • Bill is missing
  • Package not received
  • Unsubscribe from the newsletter / No contact

Open source model developed for evaluation

To carry out its test series, parlamind has developed its own evaluation model UnifyEval, which makes it easier to compare different NLP pipelines for machine learning with different backends. ©

 The  powerful tool is primarily used wherever the actual performance of an AI for processing a specific task – be it text classification or language modeling – needs to be evaluated quickly and reliably. In this way, developers and providers of AI are able to safely keep pace with the high dynamics in the field of research and development in machine learning.


 The re are several reasons for parlamind’s convincing performance compared to the competition: On the one hand, parlamind has a demonstrably better command of the German language than the competition and can also filter less relevant information such as polite phrases in the course of the analysis. Another reason is the natural nature of parlamind as a customer service specialist: the AI ​​solution already has a large number of frequent contact reasons in its knowledge database that apply to all customers.

(1) Status: October 2020

(2) ©

 The  maximum F1 score value of 1.00 that can be achieved here means that all customer concerns examined are 100% correctly understood, an F1 score of 0.00 means that everything was recognized incorrectly.

About parlamind

parlamind is a Berlin start-up for artificial intelligence in customer service. Parlamind’s AI understands incoming messages from customers on a semantic level and thus determines the issue, the specific content and the mood of the customer in the message. Equipped with these capabilities, the AI ​​enters into dialog with the customer independently and quickly executes the associated processes automatically. Current research results from the areas of computational linguistics and machine learning form the basis for the continuous further development of parlamind AI technology. Parlamind has been part of the 4TechnologyGroup since July 2018. ©

 The  4TechnologyGroup is an amalgamation of German technology companies from the communications and AI sector. All over the world, companies in the areas of e-commerce, information and communication technology, energy, logistics and finance use parlamind’s AI solutions. Over 20 employees work continuously on the further development and marketing of this unique technology “Made in Germany”.

Inquiries & contact:

Katja Gaesing
T: +49 89 5000 1340

[ source link ]

Customer communication parlamind performs Google Dialogflow Rasa IBM Watson


Please enter your comment!
Please enter your name here