Rise of AI Conference 2018 · Berlin

Who Likes Pink Shirts? Leveraging Customer Feedback in Personalized Commerce

Dominique Ziegelmayer · Director Product & Technology at Trusted Shops

At Rise of AI 2018 in Berlin, Dominique Ziegelmayer, Director Product & Technology at Trusted Shops, argues today's personalized shopping is mostly superficial, using an online shirt purchase where a retailer asked for gender but kept showing colors already rejected by clicks. Ziegelmayer proposes three hypotheses about retail's future: stores will know shoppers' personal details before they arrive despite privacy laws, since retailers already try to infer demographics from behavior; machines will generate products tailored to individual preference rather than just help customers search for one; and human vendors will be replaced by AI that has read every product sheet and review and can ask a few clarifying questions instead. Present-day evidence includes Google Duplex, an early kiosk piloted at Media Markt, and facial recognition linked to emotion detection seen at trade shows. Ziegelmayer then presents Trusted Shops' own approach: deriving approximate demographic signals from customer feedback text through author profiling rather than mandatory profile questions, classifying review text to learn which product aspects customers like or dislike, and using progressive chatbot-style follow-up questions to turn vague feedback like "everything was perfect" into specific, actionable detail. The core argument is that structured feedback, not just click data, makes real personalization possible.

Key takeaways

  • Ziegelmayer argues real personalization requires acting on customer feedback, not just click or order data, citing Alexa ignoring a negative product review.
  • A German retail group's checkout cameras that inferred customer age and gender for advertising drew heavy criticism, but showed the underlying capability already exists.
  • Trusted Shops derives approximate demographic signals from customer feedback text through author profiling rather than requiring customers to state age or gender directly.
  • Progressive chatbot-style follow-up questions can turn vague feedback like "everything was perfect" into specific, actionable detail about delivery or product quality.
  • Ziegelmayer predicts human vendors will be replaced by AI that has read every product specification and review and can ask a few targeted clarifying questions.

Detailed notes

How personalized is online shopping today?

Ziegelmayer opens with a personal shopping story: a leading online retailer asked whether the shopper was male or female, correctly inferring male from past purchase history, but after the shopper clicked only white shirts, the product list kept showing pink and blue shirts regardless. The conclusion drawn is that personalization today works only partially, since retailers rarely act on the feedback a customer already gives through clicks and choices.

Three hypotheses about the future of retail

Ziegelmayer offers three predictions. First, stores, whether online, offline or conversational, will know a shopper's personal details before they arrive, despite data privacy laws such as GDPR; as evidence, a German retail group had installed checkout cameras to infer customers' age and gender for personalized advertising, which drew heavy media criticism, showing the underlying capability already exists. Second, machines will move from helping customers find an existing product to generating the product itself, illustrated by a Nike shoe configurator where a machine that already knew a customer's preferences could skip manual configuration entirely. Third, human vendors, who today consult customers by reading product sheets and offering advice, will be replaced by AI that can read every product specification and review in existence and ask a few targeted questions, for example about budget, to narrow down a recommendation.

Early signs: Duplex, Media Markt and emotion-aware kiosks

Ziegelmayer points to Google Duplex, shown at a recent Google keynote, as evidence that natural, human-like conversations with machines are becoming possible, and describes piloting an early interactive kiosk at Media Markt, calling it a limited first step rather than the vision being described. The imagined future store has a showroom that renders or projects only products likely to fit the shopper, a conversational robot clerk that opens by reading mood from facial expression, technology Ziegelmayer says already exists at trade shows, and narrows down preferences through a short series of questions before a drone delivers the chosen product.

Why click data alone is not enough: the Alexa and ketchup examples

Ziegelmayer describes asking Amazon Alexa to reorder razor blades that had been tried once and reviewed negatively for a slimy strip on the blade; Alexa offered the same product based on order history alone, ignoring the written feedback, and the reorder was declined. A separate example, Heinz's purple and green ketchup, which customers disliked, is used to argue that companies fail when they act without first understanding customer needs and expectations through feedback.

Trusted Shops' approach: feedback-driven personalization

Ziegelmayer presents Trusted Shops' own work using natural language processing to derive approximate demographic signals, age, gender, potentially income, from customer-written feedback text through author identification or profiling, avoiding the legal problem in Germany of requiring customers to state age or gender directly; accuracy is not 80 to 90%, but is enough for a first step into personalization. A second technique uses multi-class text classification on reviews to identify which specific aspects of a product category customers like or dislike, which Ziegelmayer says could eventually be used to design an "ideal" product from aggregated preferences. A third technique uses chatbot-style progressive questioning to convert generic feedback, such as German customers typically writing "everything was perfect," into specific detail about delivery timing or product quality by asking targeted follow-up questions, and the same questioning approach could guide a customer to the best-fitting product across millions of existing reviews within a few exchanges. Ziegelmayer closes urging companies to start collecting feedback broadly, including non-textual signals like which products a shopper physically picks up and puts back, as the foundation for genuine personalization.

Quotes

Feedback is the key to understand your customers.

Dominique Ziegelmayer

If you don't understand your customer, you will never be able to do something personalized.

Dominique Ziegelmayer

Vendors will be replaced by AI.

Dominique Ziegelmayer

Questions answered in this talk

Why isn't online shopping personalization as good as it could be?

According to Dominique Ziegelmayer at Rise of AI 2018, retailers often collect signals like clicks but fail to act on them, illustrated by a shirt retailer that kept showing colors a shopper had already ignored, and by Amazon Alexa recommending a reorder despite a negative written review of the same product.

How can companies infer customer demographics without asking directly?

Dominique Ziegelmayer explains at Rise of AI 2018 that Trusted Shops uses author identification, a natural language processing technique, to derive approximate age and gender signals from customer feedback text, avoiding the legal issue of mandatorily asking for such data.

How can businesses get more useful customer feedback than generic praise?

Dominique Ziegelmayer describes at Rise of AI 2018 using chatbot-style progressive follow-up questions to turn generic feedback such as "everything was perfect" into specific detail about aspects like delivery timing or product quality.

Mentioned

personalized commerce, conversational commerce, author identification, text classification, GDPR, Google Duplex, product review analysis

Keep exploring

More from Rise of AI Conference 2018

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 3,302 words

Show the full transcript

I would like you to recall your last online shopping experience and I want you to think about whether it was a personal shopping experience oh it was a generic one which means did you have the impression that anyone that the online shopper you were buying something knew anything about you that helps you find your product that helps you make it you're buying choice did anyone of you have the impression we calling your last online shopping experience that the shop knew who you were all right 1 2 3 4 which is very good because I expected no one would raise the hand my last experience was I was looking for a shirt and that was it was not my last experience saw it one of my last experiences was I was looking for a shirt and I visited an online retailer which I would consider one of the market leaders in terms of technology and the first question I was asked when I open the West web page was are you male or female and I

said well yeah I'm male and I thought about this just um could the retailer have known that I'm male and I think so I think that the question the answers yes because I've bought numerous times in this at this retailer and my wife didn't buy anything at this retailer using my computer at least but you can think about well the gender is is it's a difficult question because if you are wrong if you are offering skirts on the front page to me as a male buyer I will probably close the page so I'm sorry so the next thing I experienced is I have to find my product myself I go through to a list of shirts and I click on three shirts white shirts and what happens in the list well I still see pink shirts blue shirts and so on and I think if I think about a personalized shopping experience it would consider my feedback by at least when when I click on the pages on the product pages and maybe adopt adapt

what I see and this didn't happen for me so my personal experience is personalized shopping works a little bit but not perfect now I haven't talked about today I want you to think about the future when you think about the future of retail the future of shopping ten years from now maybe 20 how will that evolve well it's will it be personal and I've come up with some ideas of how I see the future and this is not a visionary talk I will try to turn it around in the end but I will first talk about the future to show you the power you're working on our presence and I've come up with three hypotheses that up I think very straightforward but which have big implications on how we how we perceive or how the future may evolve and the first one I see is that basically before in the future before you set foot into a store and it may be online and may be offline and may be conversational the store knows basically

all your personal details and you can say well I don't think so because we have GDP l and everything data privacy laws going on and it's basically impossible to derive or to to to to make a profile or something but if you look at then now the the past you can see that and I don't know I don't know if you remember this last I think it was last year this at the beginning of this year the retailing group VR came into the media quite bad because they have installed cameras at the checkout and they filmed their customers in the checkout area and they were deriving demographic properties such as the age the gender maybe more I don't know and what was this used for I think it was for personalized advertising as you can see this was a big problem for the I think at least the news cabbage wasn't good but I'm sure if if you think about the future in ten years even though we have privacy laws and everything going on

there will be a possibility that we find ways to at least know which persona we are dealing with and actually online we have a lot of possibilities already the second hypothesis is we are always talking about finding a product when we are looking on a webpage or we talking to a vendor we are looking for for a product that fits us but I think in the future the machine will create the product we are looking for especially when you think about the first hypothesis that I just said that the machine knows everything about us if the machine knows all our personal details and knows our preferences and knows that I have given maybe feedback maybe just by clicking maybe by giving a product review and were appstate that I don't like pink shirts it will never create a pink shirt this is this is very straight forward but it is if it will find aspects of things I like and I dislike it can do this and again if you look at today I don't

know if you know this this is the possibility and it's just one example we have many experiments this is a Nike page where you can configure your shoes you can say I want this color I want a big spot a small swash I want this in yellow whatever and if you imagine a machine that knows our preferences I don't have to configure anymore I maybe can change something but the first thing I see when I when I go on this page is probably my shoe the shoe that fits me if it's pink or white okay and the third one is the hypothesis that I'm almost ashamed to present because you will hear it on every talk you will read it in every newspaper but it's very important to have it to get a full picture which is that vendors will be replaced by AI and I think also this is very straightforward what is the vendor good for and then consults me on a product I'm looking for a new phone and I go into

a store and the vendor will help me find my phone well the difference when we give this to an AI is that the AI can read every product sheet every product specification sheet it can read any product reveiw out in the world it can read form entries and it can really consult me on what this product does and with the second hypothesis it can ask me a few questions to find out a little bit more I mean even if it knows all the personal details it might still have some choice on on my budget if I'm willing to even if I can't spend 500 EUR am I willing to spend 500 euro today so basically the vendor can ask me a few questions or the AI and either produced my product or consult me on an existing one and as I did before you can see that machines are already finding their ways in our daily daily daily shopping experience this is at Media Markt I think it was something that I piloted last year

and yes this is not the machine that I'm I'm envisioning for the future it's not I don't want to talk to a moving trashcan I don't want to this machine will probably not scoop through millions of product reviews it will not have all the information about every product available but this is starting and I think maybe you have seen the keynote from Google a few few days ago I think well we have Google duplex which shows in an in an impressive way how we can communicate with machines and if you imagine a machine looking a little bit more pleasant than this one and talking to you just like a human but with the knowledge of everything this could be a revolution for for retail and you can easily transform this to a webpage a webpage today is something where I need to find my product where I need to read a lot of information if there is a next iteration it might be a bot system that I talked to and even another iteration which

we are almost experienced today is conversational commerce I talk I already talked to a machine sometimes if I want to buy stuff and you can get all these hypotheses you can just apply to these other areas okay if I put everything together and I envision the future it is something like I set foot into a store and when I come when I go inside I see a little showroom which renders products prints them or projects them somehow that I that I might like okay so I go into a store and it's not like today there is a shelf full of clothing no there's a few a few spots where clothing appears that may fit my needs this is my personalized idea and I go up to a clerk to a robot which talks to me in a very natural way and he asked me how are you doing today and the funny thing is he already knows that's that's why he asked this question he knows that it's socially very good to do it but

also he knows that I'm in a good mood if I'm in a bad mood he will probably say hey you look you look you don't look so happy today how can I help you and you can see this technology right there on the exposition up there I've seen a company which does facial facial recognition and assigns the facial recognition to emotions I mean also from from a research perspective this is old okay and then I talk to the Machine and with some progressive question the Machine narrows down the search of Who I am and what I like and offers me a customized product that the showroom the the the spots in the showroom will change and the products that I might like will appear and finally I can just take it or there's a drone delivering it to me that's something we will see okay but is that this is an apply talk and I don't want to I'm not on the futurist stage I'm not a futurist so let's take a step back in

two what we have today and I think everyone knows this Amazon Alexa a digital assistant where I can play my music using my voice I can do a lot of stuff I can turn my lights on and often but I can also say Alexa please buy something and this is something I tried the first time about a year and a half ago I said Alexa please buy me some razor blades and Alexa said I have found these in these ways of date do you want to buy him and I said no and why because I bought these razor blades a few weeks ago the first time and I tried them out and I didn't like them because I have a slimy strip on them and I left it in the water and I didn't like it at all I didn't know why they have to add a slimy strip on a razor blade and I wrote a product review stating that I don't like it but obviously what did Alexa do it looked to my

to my order history which is good but it didn't take into consideration what I like and what I don't like and this is why it didn't make the sales so basically what I'm saying is it's not enough to look at this at the data it's a click data or something it is important to understand your customers needs and expectations and there's another thing I don't know if you can see it right and I don't know if you know the history behind this Heinz ketchup has come up with a purple and green ketchup thinking mm and what do you think did people like it or not people hated it and I think this is a this is a me as a as a product manager or the product manager inside me would say yeah well I have totally failed because I didn't I didn't understand my customers I didn't ask my customers about their needs and their expectations if I would have done this I would have would have done this they probably would have wouldn't

have come up out with up with green ketchup they would have said do a biological variant or something okay but I said again this is an applied to talk and I want to show you a little bit we are working with feedback to move a little bit on in this direction we asked I'm working for trusted shops we as trusted shops we are at the beginning of the AI evolution I would say but still we have found many use cases where AI or natural language processing can help us a lot and this is one of the ideas we are we have come up with with all the discussion about privacy and data going on we have thought about how can you derive personal the more or less personal demographic data without violating laws without building profiles and everything because basically if we have the email address we are not allowed to ask you mandatory how old you are if you're male or female and stuff like this this is in Germany this is it this

is creating a profile and which is somehow I'm not the legal guy but if I listen to our legal department which is some our problem if you are familiar with text classification on natural language processing you might also know about author identification or author profiling we can derive a lot of data from text from written text from spoken text we can we can derive the age the gender maybe even the income and this has limits and this is not 80% accuracy or 90 but it suffices for the for our first step into personalization and I think this strongly supports the hypothesis that my personal details will be known because the first time I write a feedback I can at least the second time I come into the store the first the second time I come into a store I know who you are how old you are if you are male or female and so on just from the stuff you said okay the second is very straightforward when you have text the problem is

always understanding what people wrote because if I want to act upon and if I want to to make a personalized recommendation I have to understand what people like and don't like and basically what we do is we do a multi-class classification on our texts on our texts and then we find out which aspects people liked and didn't like about a product category or product and if we are taking this a step further we can start creating products because when you have the product category mobile phones and you know what people like and don't like about existing phones they tell you what they like and what they don't like you can build the ideal phone right so this is the first step into this direction and the last one is quite connected to this it is understanding text what we have experienced when you ask people for their for the full rating regarding a company or something they will often write at least the Germans thank you everything was perfect when they give five stars does

it help the vendor does he know this works very good this aspect works very good this you could improve this no he doesn't so basically and you know both technology I don't need to go into the details but you can use a bot like technology to talk to to your customer and what we need to do is to ask progressive questions that means the first thing is yeah he says I write down everything was perfect thank you very much the second thing the Machine will ask okay thank you very much for your feedback can you tell us a little bit more did the delivery work well this was the product quality okay and I think about it and if I have the time I will maybe revise my statement and I will say okay the delivery if you're saying this the delivery wasn't perfect it was it was predicted for Monday but the product arrived on Tuesday and then I can ask another question do you think this was related to the dealers process or

do you think this was due to hammers or anything and given this this production is these progressive questions we can narrow down the search on what happened here and we can gather feedback and if you sling this a step further you can use the exact same technology to find out which product you need to consult on if I say well I want a cheap product and the machine can go through the millions of product reviews and look which aspects can divide this a little bit further and it can ask the next question so I think within two or three questions I could a machine could consult you on every existing product and could point you out which one is the best for you okay there are many many more use cases and I'm happy to discuss with you ah sorry the last one as before this supports the hypothesis that vendors will be replaced by AI because this technology can help narrow down the search it can help narrow down whatever you need to know

and this is the first step if you combine this with technology like Google duplex we are there I think ok there are many many more use cases and I'm happy if I'm running around until 6 o'clock this evening so if you have any questions if you would like to discuss what else you can do just find me and talk to me what I can tell you is that I indeed believe that feedback is the key to understand your customers and then given the feedback of your customers that you can enable truly personalized commerce and just with this because if you don't understand your customer you will never be able to do something personalised and my strong recommendation is to if you don't if you don't collect feedback yet and feedback is not just textual feedback its clicks on something it's everything that is connected you can even think about in a in a supermarket someone taking out a muesli muesli box and looking on it this is feedback does he put it back or does

it put it in his anus in his so just start collecting feedback and start you start your ideas what can you do with a eye on this ok the end of my talk have a good lunch if you have any questions I think you can find me afterwards so everyone can have his whatever is available thank you very much

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.