2017 in review — startups, bots & AI
After doing a review of 2016 I realised how helpful it was for myself as I saw how much we had archived and what we’ve learnt with chatShopper (fashion shopping bot on FB Messenger, Kik, Google Assistant, Alexa, ios/android…). I also found out that it’s sometimes very helpful to step back and take a different look at your journey and challenge yourself. Therefore I decided to do it again 🚀:
January 🤖🏆
The year started with a great success — we won the first Chatbottle award as worldwide best ecommerce bot 🤖👗😍. Looking back that was crazy as we won against big players like H&M. Probably people liked that our bot was one of the rare “conversational” bots in ecommerce and not a decision tree / button only approach. Our “success rate” (right understood and correct delivered messages) back then was only 46% 🙄 and we were not really satisfied with our bot. Those of you working on bots know how hard nlu (natural language understanding) is and that you need data, time and persistence to build a great bot. For us it was always clear that we’re not in this game for a short success but that we want to build something great.
February 🤖💄🔈
February was quite a busy month: it started with the launch of a valentines bot we built with Douglas (beauty retailer). I was quite simple but had different things you could do — find a present, get ideas for handmade gifts and send love letters to other Facebook Messenger users. It was only live for 2 weeks prior to valentines day and it was great to test our technical solution in another context.
We also got accepted into the VentureVilla Accelerator program in Hannover. We chose it mainly because of the great team and their ecommerce competence and network. Nadine (leading the program) was one of the founders of CouchCommerce (exit to Newstore) and we met a lot of awesome people there (thanks Finja, Sven, Alex, Kai, Thorsten & many more!). We could not have chosen better sparring partners and I’m grateful to have them among the people who are supporting us on our journey. When we let Alexa pitch our startup it was a great start into the program 😂🤖🔈.
March 🤖📱👩🏻💻
We finally launched our new checkout with Zalando (thanks to the great distributed commerce team!!) — without having to log into an account you could buy in the web-view of Messenger. But as we were still working on improving the nlu we weren’t marketing the bot (yet). I got a bit more into coding and built my first Alexa Skill and even got a hoodie from Amazon for doing so (if someone from Amazon is reading: please make female hoodies, they look horrible on women!).
April 🏢
We moved our office to the great Innovation Fabrik Berlin from Infosys — a great space, team (thanks Julien!) & lots of interesting startups (among them rasa.ai, we love the guys!).
We started working on a new backend for our search as we realised we need much more information about the products than the Zalando API has to offer and we also wanted to use a graph database (shoutout to Neo4j, great solution!) to represent the semantic connection of the products.
We learnt that new interfaces (chatbots, voice assistants) lead to a change in user behaviour towards conversations and natural language searches and the existing search approaches aren’t the best for this new area. My great cofounder and CTO Matthias Nannt did an awesome job designing and building it — I still think he is a genius 😍!
May ₿
As the guys were busy building our new backend I had some time exploring what’s going on in the tech scene Berlin and realised that the blockchain topic had matured and grown a lot since I looked into it (2012). As I was annoyed that I had missed the bitcoin wave (that’s what I thought when it crossed $300, $600, then $1000…) I got excited by the idea and potential impact of Ethereum and started investing a bit.
June 🤖🤖🎤
This was the month with the most conferences, I think I’ve never met so many people in one month which I had only “known” from online communities or Twitter. At the chatbot summit in Berlin it felt like a class reunion when meeting several people from the bot community, it was an amazing experience!
The K5 future retail conference was another awesome event where I pitched chatShopper and got great feedback from industry experts. While visiting other talks I realised how relevant our topic is already.
July 🤖🤓📦
I finally moved to Berlin! I wanted to this much earlier but finding a new place for me and my family (+dog) and selling a house was just not possible to do before.
We finished our new backend and also moved to our own nlu solution — since 2015 we had used wit.ai but there were ongoing issues (long API response times etc.) and we wanted to be independent from another player. We also realised that we enjoy working on challenging tech tasks and rather build our own solutions. Changing the nlu meant a lot of data labeling and therefore we changed the bots language to english only — which annoyed a lot of our german users. As a startup you sometimes have to make hard decisions but time and resources are very limited. Therefore we decided to focus on one language and enlarge it at a later stage.
August 👨🏻💻🤖
Finally Chris joined us full-time! He’s a great data scientist who had worked on deep learning for quite some time at Max Planck Institute during his studies. He was working with us part-time since the end of 2016 and having him on board now full-time meant we could go ahead much faster with the CNN’s to enrich the product data. By having deep learning models enriching the data we can have much better quality and quantitative data with low costs and therefore much better search results.
We started advertising the bot but the results weren’t convincing. One problem was that the ads were run as if it were an app and promoting a chatbot is a very different thing. We should have done the ads ourselves and keep our assume →test→analyse →learn →iterate approach, but we thought it would be better if people who’re working in advertising for longer do it. Lesson learnt: do everything yourself and if it’s not done the way you wanted it you can only blame yourself. We couldn’t get enough conversations started as we wanted to generate kpi’s which we need in order to compare them to the very promising numbers we had collected in our pre-bot phase (same idea but with human experts answering the requests) in 2015–2016. With these numbers we wanted to start rolling out our B2B white label solution for fashion retailers and brands. But we also realised we still needed to spend time on the nlu — the success rate was about 68% at that time and nlu and data quality were the main points why it was still too low. We set the goal to reach 80% by the end of October to do more ad tests afterwards ourselves. The new motto: no meetings, just work to reach that goal!
September 🤖🔈👖
We launched two new front ends, a bot on Kik and an action on Google Assistant. Our new backend API made it very easy and fast to build new chatbots or voice assistant and therefore I built the Google Assistant action myself (with some help from Matthias 😉). It was a great feeling to build something myself and to contribute to the tech side of our startup. It was hard to admit that to myself but I realised I’m not a business person, I love to code, to build and to learn (why the hell did I study business 🙄). As there was still plenty to do to get to 80% success rate I didn’t worry too much about this fact back then.
October 🤓🤒👩🏻💻
Working on bots means you’re working on the Facebook platform since for our target group FB Messenger has the most users and they’re pushing the bot topic a lot. Therefore I discovered the new initiative Developer Circles quite early and was asked if I wanted to become one of the leads for Berlin. As a former event manager I still love to organise stuff and I also love the Berlin tech community (I’ve been to over 100 tech meetups in the past years) I immediately said yes — thanks to Malte & Zane who’re doing a great job! Together with the other amazing leads (shoutout to Raphaël Gonçalves, Andreas, Stephan & Yulian Ustiyanovych) we organised the launch event which was a huge success and it’s great being a member of such a great community! If you’re based in Berlin => join our group 😍
The rest of the month was quite boring for me since I was ill for several weeks and the only thing I could do was labeling data 😔. While in bed I stumbled upon an article from Daniel Jeffries about “Gamifying the Delivery of Money” and this article really got me thinking about the future, tech and the possibilities decentralised systems and (!!!) organisations bring. I wrote my own “Masterplan” of steps which would be needed to archive this but I got terrified by the thought what could happen if this works. Sometimes ideas need their time to grow and mature.
The guys very very busy implementing the first deep neural net for data enrichment (where we reached 92,46% accuracy) which meant that if users searching for destroyed jeans we could find products without having these information in the Zalando database 👖🎉. This was only the beginning, our goal is to be nearly independent from product data in the future!
November 🏆📱🏷
We finally made it: over 80% success rate and over 90% with our new nlu 🎉🚀. This was a mayor milestone since we had worked very, very hard to get there. We also launched an ios & android app (thanks to reactjs it was easy to build for both platforms) in order to test and compare ad results for bots on platforms and within apps.
While preparing and labeling data (which is a mayor task if you want to work on nlu, image recognition or any other machine learning task) we started to build a labeling tool which we put into an app.
I can’t believe how fast we had become: from idea to an app for ios and android it took only a few weeks. Matthias had become an incredible programmer in the past 2 1/2 years, Chris no longer is “only” a data scientist but can handle code very well and I’m getting better with it as well. I’m very proud of all of us, I would have never thought that it’s possible to learn, build & iterate at such a speed 👍.
December 🎄❓
I prepared the launch of the B2B products for fashion retailers/brands but while building landing pages I realised there is too much and we need to focus on what we want to promote: a white label bot solution a solutions to enrich fashion data or both? When starting to speak to big corporates about our solutions we often got the questions if we could imagine to do an exit. Learning, working and handling a startup life at such a fast pace has it’s downside and we couldn’t make any decisions. We therefore decided to take some rest and make decisions in the new year. I spent some time with my family, sister, mum, dad and great friends. These are rare moments and I’m grateful for them to still support me and understand that I don’t have as much time as I should.
— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —
It has been an awesome 2017 and I’ve never learnt more about myself than in any other year. Thanks to Matthias & Chris, you’re the best team to work with! Thanks to the people who advised, helped and invested in us! Without all of you this journey wouldn’t be possible and I hope one day I can give back everything I received. I’m very grateful I met so many bold people in the last 2 1/2 years, it feels like I’ve just started to “see” the real world and got a glimpse of the future. I hope I can contribute to this new world and help that it becomes a more fair place for all of us.
As you can read 2018 will be a very exciting year for us and I’m looking forward to share more learnings with you in the future ❤️👩🏻💻.
