<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-68358952-3"></script>
<script>
  window.dataLayer = window.dataLayer || [];
  function gtag(){dataLayer.push(arguments);}
  gtag('js', new Date());

  gtag('config', 'UA-68358952-3');
</script>

 

womenInai.co
  • White LinkedIn Icon
  • White Twitter Icon
  • White Facebook Icon
Contact Us

©2020 Women in AI. All rights reserved.

WaiTALK: "AI IN CUSTOMER JOURNEY" in BERLIN

Updated: Sep 10, 2019

[WOMEN IN AI LAUNCH IN BERLIN, GERMANY]

"I want the best, near me, right now." Today's customers expect fast and personalized online experience. Technology companies like Google, Facebook, or Amazon change the way customers communicate, search, find, or shop through applied machine learning & deep learning.


For the launch of our first WaiTALK in Berlin, on September 3rd, 2019, our Ambassador Tina Nord has invited five experts with different backgrounds and asked them to explain how companies can use technology to meet the changing customer expectations.

"How AI & Data change customer care" by SARAH AL-HUSSAINI, COO AT ULTIMATE.AI



Sarah Al-Hussaini, Chief Operating Officer from ultimate.ai, kicked off the event with a strong presentation about the challenges and learnings of her AI 1st company. Sarah and her Co-Founders build their state of the art natural language processing technology from scratch, starting with one of the most challenging languages in the world - Finnish. Focussing on Scandinavian languages made them enter an AI niche that has been neglected by other tech giants so far. Ultimate.ai was, therefore, a pioneer in bringing AI and the nordic market together.


As a representative of a company that aims to automate and augment the customer service industry, ultimate.ai had to deliver a flawless performance, straight out of the box. 4% of all workers worldwide are customer service agents that have to conduct mainly repetitive tasks all day. The agent's main challenge is to handle large amounts of information at a fast pace to ensure an excellent service. Data and repetition make customer service an ideal area to apply deep learning technologies. As AI quality depends primarily on data, the team around ultimate.ai had to find a solution for handling the high amount of unstructured information, coming in on a high frequency. Therefore, they started to group data automatically. This process also allowed them to make their products available for their customers in a short time frame.


But launching a product fast for different customers did not remain the only challenge. An AI 1st service means to iterate continuously - and ideally automatically. However, the customer care agent needs to be part of the picture. New technologies should enhance the work of the existing staff and are not meant to replace them. Understanding their needs requires a user-centric focus and a constant exchange of feedback. The agent happiness is, therefore, one of ultimate.ai's core KPI's. Today, Sarah's company offers its services in all languages for all countries around the globe.

"How we created personalized outfits for each of 27M customers" by Vilma Sirainen & Marta Skassa, Product Design Lead & Senior Product Manager at Zalando SE



Marta Skassa and Vilma Sirainen from Zalando SE followed a similar theme: the customer-first approach in their work as a product manager and product designer at Europe's biggest fashion retailer, Zalando. The e-commerce site is live in 17 markets and serves 28 million customers with 400.00 articles from 2.000 brands.

Developing an algorithm for this purpose also meant to ensure the right quality of the outfits. Which leads to the philosophical question: What is the right quality mean in this specific case? The answer was simple: it means that the recommendations made by a machine should be as good as human outfit recommendations.


Combining fashion items into an outfit appeared to be one of the significant user challenges. The product experts teamed up and designed and launched a first test solution for this problem, based on a large Excel sheet. This Minimum Viable Product enabled them to add quantitative data (user behavior) to the qualitative data (user interviews) and proved that an automated outfit combination makes sense for the customer and increased the click-through rate. This was enough evidence to get the resources it needed to build the machine learning-based solution to scale the answer for millions of customers.


However, building an intelligent, automated solution required several research steps in itself. To ensure a natural dialogue between men and machines, which was comparable to a sales clerk in an offline shop, meant to understand fashion and outfits at its core. It also implies, to identify the exact intent in that very moment, how to handle fashion trends and to discern customer taste. Developing an algorithm for this purpose also meant to ensure algorithmic quality. Which leads to the philosophical question: What does algorithmic quality mean in this specific case? The answer was simple: Algorithmic quality means that the recommendations made by a machine should be as good as human outfit recommendations.


Inspired by Alan Turing's Turing test, Vilma and Marta approached the challenge similarly. The team developed two outfit algorithms and tested them versus outfits created by a professional stylist. A large group of test users was asked to rate those outfits without knowing which one was created by a machine and which one was combined by a human being. It turned out that the ideal solution was a combination of one algorithm and human advice. The generated learnings were used to improve the quality of the winning algorithm. Today, Vilma and Marta are already working on the next level, which is even better personalization and enabling customers to give feedback in real-time.



"The future of personalization and automation: The Content-Lake!" by Norman Nielsen, Director of Growth at Omio/GoEuro



Norman Nielsen continued to emphasize the previously mentioned changing customer expectation and the resulting need for more personalized experiences on the web. He agreed on that point with Google, who introduced a dedicated persona - the demanding customer - on their blog "thinkwithgoogle.com". Norman also highlighted that personalization is already part of many marketing activities and detailed data about customers readily available on various sources.


Being one of the leading SEO experts in Europe, the Director of Organic Growth handles large amounts of performance-oriented content of major online websites like Omio/GoEuro with millions of users. Personalization, therefore, is a specific challenge due to the number of text snippets needed to serve individual user needs. Automating this text production was, therefore, a reasonable test case, while not knowing if the current state of technology is already capable of producing high-quality content. After reviewing similar providers like OpenAI or TalkToTransformer, he decided to run a test with AXSemantics and use a content lake (similar to the data lake) containing text snippets and data about travel destinations to produce category page copy. He and his team were positively surprised about the quality and continue now to automate significant parts of the text production for performance purposes. However, Norman admitted that the final copy still needs minor human adjustments.


Finally, he mentioned voice search as one of his passions and main AI-related SEO challenges to date. Norman underlined since voice assistants are part of the customer journey that there is no classic funnel anymore. Instead, customers jump between intents like search, buy, book, inform, as well as all kinds of devices and decide ad hoc. This change will remain the future challenge for all marketers, directly influenced by AI.

"Why did the AI ditch the API to improve the customer journey" by Isabel Schwende, Co-Founder at Mobius Labs



As the Data Protection Officer and Co-Founder of Berlin-based AI Start-up Mobius Labs, Isabel Schwende gave some insights into the technical challenges an applied AI company has to accomplish. Mobius Labs works with visual data and computer vision.


Their technology tags images and identifies objects or even emotions and similarities.

Isabel's customers are usually facing a similar challenge, as mentioned in the first presentation by Sarah Al-Hussaini: They are working with large amounts of unstructured data. Especially images are not easily machine-interpretable, which forces companies to select an intelligent technology or build a dedicated team to fix that and label or structure the data manually. In the first case, the smart (or AI-driven) technology connects to the IT system of a customer via an application programming interface (API). It is hosted on one or multiple servers that process requests from a publicly available web-based API to output results. This solution comes along with four significant benefits: There is no need to install software on a server, and it is easy to send data to the API. Also, software patches are applied immediately. Furthermore, most experts still believe that cloud services are state-of-the-art, so you should use it.


However, Mobius Labs decided to offer a Software Development Kit (SDK) based solution for their Mobius Vision technology. As a significant benefit, Isabel's team takes over the complicated technical integration as the so-called "delivery manager". She and her colleagues believe in SDK because of the simple installation in a Docker container. It is also even simpler to send data, as it stays in the same IT-system. Furthermore, it offers a customizable solution for human-machine cooperation and provides privacy by design.


As Isabel concluded in her presentation: "AI is far from going crazy on a remote server, like in the movie i-Robot." With singularity not being an immediate threat, it became clear in all presentations, that technology works at its best when humans and AI collaborate and support each other. Artificial intelligence does not provide ideal results without human feedback loops and support. However, it does offer significant benefits for companies that want to satisfy the growing and continuously changing user demands during the customer journey. Through structuring and combining large amounts of data, AI-powered technology builds the foundation of personalized user experiences.



Women in AI is looking forward to the next WaiTALK in Berlin. Special thanks goes out to Facebook, the company that hosted the event in their Berlin office.

ANY QUESTIONS?

Connect with our Ambassador for Women in AI in Germany, Tina Nord

And don’t forget to subscribe on womeninai.co and join our Germany channel.


The WAI Team

womeninai.co