Is Product Personalisation An Illusion?

Nadia Udalova at UXDX EMEA. Video: https://www.youtube.com/watch?v=SZpg0zLhip8

Readable transcript: edited from the recording's captions for readability (fillers and false starts removed, punctuation and section headings added). Wording is the speaker's own. Timestamps are positions in the video. Names marked [?] could not be verified against the audio.

Is personalisation an illusion?

[00:00:00] Hello, UXDX EMEA. It is a pleasure to be speaking here again. My name is Nadia Udalova and I'm Head of Product Design at Moonpig Group. It is a bit more than a year ago since I was presenting at UXDX Netherlands, and I'm so happy to be back.

[00:00:18] Today I would like to challenge everyone listening to me here and ask, do you believe in true product personalisation for customers? How personalised is the content that some e-commerce shops are offering to us and proposing to purchase? Is it based solely on your preferences and interests as a user, or is it some kind of an illusion that today's e-commerce platforms are trying to create?

[00:00:48] During the next 25 minutes I'll make you familiar with Moonpig as a business, I'll walk you through some of the things that we've done and I will talk a little bit about our product range. We will look into some data and wrangle about product personalisation, discussing Moonpig's approach to this. And at the end I would like to leave you with some of the biggest learnings that we concluded to, while trying to create the optimal personalised experience for our customers.

[00:01:19] I appreciate those listening to me today in the United Kingdom or even Ireland who may be already familiar with the Moonpig brand. For those who haven't heard of us yet, we are all about celebrating those great life moments: the big birthdays, the new arrivals, those surprise engagements and just-becauses. We are helping our users to connect to the people that they care most about. We are one of the biggest online shops that offer personalised and non-editable[?] greeting cards of all sizes, from standard to giant, as well as a wide range of flowers and gifts to make a great pair and a great gift with them.

[00:02:02] At Moonpig, our mission, our aim is to create those better, more personal connections between people who care about each other, and we are doing this by making the gifting experience effortless. Imagine choosing your own card, adding the personal picture of your loved one, typing their name and adding any message you want to the card. We're turning our brands into the ultimate gifting companion. This means having the perfect gift range and leveraging all of the data and all the technology that we have to be as thoughtful as possible for every gifting occasion throughout the year.

How often gifting goes wrong

[00:02:39] But before moving forward I would like to walk you through some data to prove how hard it can be to select the right gift today, and how often we humans fail with this mission. There was a survey done by Worst Gift Giver[?] in August 2019 with almost a thousand adults, and it has revealed some very interesting things. Just listen to this: 45% of respondents said that the person who gave them their worst gift did so more than once in their lives. So this means more than one celebrating occasion.

[00:03:13] The average estimated price of a worst gift was about $20. Just now imagine how much money is wasted on bad gifts. Relatives seem to be the worst gift givers, more than anyone else. This is very surprising to myself, because I would expect that my family and my close ones would know me the best.

[00:03:45] Also 16% of respondents said that receiving their worst gift ever worsened their relationship with the gift giver. So all of this data just indicates that all people are different in our tastes and our preferences, and we may need some help in surprising our loved ones. Also, as we will see later in my presentation, it may not always be the fault of the person who is choosing the present when they give something that isn't perceived positively.

Behavioural segmentation at Moonpig

[00:04:17] So, in order to understand the intentions and preferences of our customers better, at Moonpig we have done quite some extensive research to segment the type of people who are buying with us, in order to fulfil their needs better. We used behavioural segmentation to get a better understanding about those people, not by just who they are but by what they do, using the insights derived from customers' actions.

[00:04:46] Behavioural segmentation is based on patterns of behaviour displayed as they interact with a company or brand or make a purchasing decision. It allows businesses to divide customers into several groups according to their knowledge or attitudes towards the product, or their response towards the product, service or brand. So our objectives here were to identify the customer segments and to understand how to address the particular needs or desires of that specific group, discover the opportunities to optimise their customer journeys, and quantify their potential value to our business.

[00:05:30] Unfortunately, here, due to the confidentiality of data, I cannot show you the exact information and data we have come up with. We have tons of data, and by using the prioritisation metrics we have identified the top customer segments that we would be focusing on with all our efforts in the future, as we believe we can get the best value out of focusing on those customers.

[00:06:02] Also, we quickly understood that, with all of those people's specialties and personalities, personalisation is going to be a big help and will be fundamental to providing the optimal customer shopping experience for our users. It is also super beneficial for businesses to generate more sales and increase customer retention.

Why conventional personalisation engines are not personal

[00:06:19] However, there was also one quite big problem that we came across when trying to build for the best customer experience in terms of personalisation. Conventional personalisation engines take certain information about each shopper, such as their age, their location, gender, and then they run this data through statistical algorithms that have been developed based on the data of thousands of other customers. It runs this data through the data that was produced by shopper actions, like clicking on an item, liking an item, adding the item to the cart. It runs this data through the algorithms and surfaces back items for the customer.

[00:07:09] So for example, if you are a 33 year old woman and you live in London and you're searching for a floral dress to add to your basket, you found it, you clicked it, it's in your cart right now, this action will make you see, and you will get as a result, more products shown that were liked or purchased by other British women in their thirties who also bought that dress. So here's the problem: sequencing algorithms that base recommendations on the prior actions of thousands of other customers, they are not personal at all. They are predictions based on statistics, but not an individual trait. So does this mean that personalisation is indeed an illusion?

[00:08:00] Decades ago, before the internet, the most successful shopkeepers were those who kept a list of their products that their best customers loved, and they called those customers when they got the new stock. This type of personalisation kept customer satisfaction really high and it cemented a lifelong loyalty. I was born and raised in a city in Ukraine, and my mum, still many, many years after I have left the country, she is still living there and buying her favourite cottage cheese from the same lady at the local market. This lady, the shopkeeper, calls my mum on her phone every time when she gets the stock of the cottage cheese in, and my mum always runs to her and returns to buying with this lady, because it's simply convenient and it meets all of her needs.

[00:08:49] Today, as technology evolved and e-commerce was born, artificial intelligence promised us to enable a much more sophisticated playbook, not only for providing those real-time recommendations but also for elevating engagement of the customer and better communicating with them. However, the real truth is that personalisation is still hard, and this is why it's so easy to get it wrong.

[00:09:17] Here are a couple of examples where other companies haven't done such a great job about this. Unfortunately, many of you have been the victims of the first name fail marketing model, and it probably happened quite a lot to you. Have you ever received those emails that would say "their first name" instead of your personal name, or "F name", or having weird characters? This is occurring when something goes wrong during the collecting of your data from multiple services or resources. Or cases like I've shown on the right: this is an unfortunate situation for Pinterest, when they have accidentally congratulated hundreds of single women on getting married, and those women had been simply browsing for wedding related content. This is a particular example of getting user interactions wrong. We don't want to be there.

Starting small with cross-sell

[00:10:11] Therefore, at Moonpig our approach was to start from a very basic level and take an incremental approach to personalisation. The idea was to start looking at one particular area and focus on it until we get it right. So we decided to give our attention to the cross-sell experience and make it better than it is today.

[00:10:35] We had some specific challenges due to the nature of the business. We are not a typical e-commerce business, because for us it's all about that relationship between the users who are sending each other cards and gifts, and not the users themselves who are purchasing the card. So for us it's harder to collect and analyse the data. It is also taking us longer to learn about the user. Our average order frequency is only three times a year, so we are not collecting tons of data as often as other big e-commerce businesses do. The cold start problem is also a real challenge for us. This is about making a recommendation to those customers who are first timers with us, and it's there due to all the above reasons that I just mentioned.

[00:11:25] So, in order to understand the different goals of our customers, we also have done lots of user research to define and outline the core job to be done of our customers and those that are related to it. This helped us to realise how many different variables there are impacting the success of the right gift selection. To give you a concrete example, the core job to be done that we have identified was of course giving a card or a gift to celebrate an occasion with a loved one. And we have found more than 50 related jobs to be done. For example, finding a suitable card, or a willingness to be thought of positively by the recipient when they get the card. Also finding the right words to write in the card or to say with a gift, getting the correct address of the recipient and making sure it arrives to the right place and at the right time, and many, many others.

[00:12:26] To show you a little bit more detail on our Moonpig journey of developing a personalised experience, here's how we started with enhancing our cross-sell three and a half years ago. And by the way, a little disclaimer here, everything that I'm going to show in the next couple of slides is only a part of the work that we've done in this time. It's not everything.

[00:12:50] So, three and a half years ago we had started from a very limiting situation. Our cross-sell experience would look like the following. After a customer would add a card to the basket, he would be introduced and invited to the cross-sell step that would be showing 16 products as a proposal for a gift in a static list. Those 16 products would be our best sellers. As I said, they would be shown to the customer in a list of static items. So for example, if you were choosing a card for your kid's birthday, we would show you the bestselling 16 gifts that can go in a pair with this card that you selected.

Unique pairings, wider ranges and carousels

[00:13:38] So, a kind of okay-ish cross-sell experience, but pretty static, isn't it? We wanted to improve. So we looked at proposing a unique pairing using the lift algorithm. What that means is that when the customer would be choosing the card, we would show her gifts that were previously bought uniquely with this same card. Some customers don't always know the perfect gift for the recipient, so we decided to leverage the data from similar customers who previously shopped successfully for the same type of recipient and occasion. Essentially it would mean that our customers who are more confident shoppers are helping those customers with less confidence.

[00:14:22] So for example, if someone chooses a traditional card, these cards are usually more likely to be bought with flowers, and what we would do, we would match these cards with a couple of nice bouquets as a gift. On the other hand, humorous cards, or those cards about wine drinking, they tend to match a drink gift. So we would follow this pattern. This was a very simple improvement and it worked great for our customers. The first test that we've done to validate this, we saw a big win, 10% uplift. So we're just trying to give a little bit more relevant gifts to our customers, and we already saw a big increase in metrics.

[00:15:10] After this we wanted to attack our biggest challenge, biggest issue, the cold start problem. So this is when one of the first-time customers comes to our website. Remember the 16 gifts as a recommendation after adding a card to your basket that I told you about earlier. So we have tried expanding the range from 16 gifts to 160, in order to provide at least a wider selection of gifts, your recommendation to buy with the same card.

[00:15:44] How did we do it? We worked with grouping of cards and missions to ensure that we have 160 products to recommend with each card. So if you bought a card congratulating someone for moving to a new house, you would get house decor gifts as a recommendation. For every card we would have a mission group level recommendation, and also, as we would see your orders as a customer coming in, we would be checking if we can recommend a better product based on what you have bought before, and we will rely less and less on those mission group level recommendations. Somehow, unfortunately, this wasn't a massive success for us, but this was something very useful for the next step that we undertook.

[00:16:29] So the next thing here was that we introduced product carousels. We have moved away from those 160 products in a random list to making a few groups of gifts in a few categories of products. This is what we called product carousels. Previously our customers found it hard to browse such a long static list of products, but yet we knew that they wanted to have more choice. And this meant that we had to group products to make the browsing of gifts easier for them.

[00:17:03] From the UX perspective this was a big improvement. Just to show you how it looked: on the left we see that old cross-sell experience, just a static list of gifts, and on the right we see the product carousels, you see a couple of products grouped together by the product type. So this allowed us to add more products to show to the customer and provided variety. Users got to see a set of two to five carousels, and one of them could be flowers and another could be chocolate, et cetera. This provided a great browsing experience and it was a big win for us. It brought us 10% of uplift.

[00:17:49] One of the most recent things here was an idea that relied on people giving us information themselves. We are currently working on a mechanism that would help us read the card text that a user would type into the card, and that would help us identify the mission from there. So for example, if you were choosing a birthday card and writing on the card something like "for my dear dad", we would know that you're buying a card for a birthday, for a male and for your dad. And we would turn this text into the recommendations to serve as the best gift for you as a customer. This is still coming in the future, so it's not currently available yet.

Lessons learned

[00:18:32] So, based on our three and a half years journey, I want to leave you with our most important lessons that we learned through our experience. Start small and iterate to improve. As I said earlier, take one area and focus on it to get it right, target all your efforts around it and find a way to collect the data and learn. Don't scatter your attention. You saw how we took a main area to focus on around our cross-sell experience and tried little steps around it.

[00:19:05] Capture the context and data everywhere where possible in order to learn faster. Any little piece of information you can learn about what your customers are doing or are willing to do will be helpful for you. Ask your data scientists for support, or hire an agency to help you. This will give you so much more confidence in what you're doing and that you are doing it right.

[00:19:30] The next point I want to focus on is to say that you should be looking at what your users find the biggest struggle and focusing your attention over that. So spend some time learning about the biggest jobs to be done of your customer. Pick something where you can genuinely anticipate your users' needs better than they can. For us this was proposing the best gift to go with a card, because we knew it was a struggle for them. And remember, not every step needs to be personalised at the individual level. You can also look at what other users are doing in the same situation.

[00:20:10] Use the wisdom of the crowds to surface the right item back to your customer. So, using the behavioural data from the customers who are buying for the same mission and the occasion to help future customers have the same type of recipient. For us this was a big help and a big win. Remember, those more experienced customers can help those who are the first timers at your shop and who have less experience using the product and selecting the gifts.

[00:20:43] Also, using product metadata to suggest similar products to each other will be a big help. For example, like "recommended for you" or "people who bought these bought also that". Also, we were not looking at the customer only himself, but also what many people have done in the same situation. That helped us surface the results to the people and make the best recommendation.

[00:21:07] And last but not least here, relevance to the users should go over everything else. For us at Moonpig, designing a personalisation strategy around our customer was the biggest goal. Of course there'd been some parts where business strategy played into that, but our customer needs were always in first place. We built something where we could genuinely anticipate our users and their needs better than they can, and we help them on.

Failures, and the answer to the question

[00:21:40] So, with all of these great stories, did we always win during this journey? And the answer is, of course not. We have failed quite a bit, and here are some of the most tremendous failures of ours, just to showcase the examples. We were spotted offering alcoholic beverages as a gift to the people who were buying cards to congratulate their loved ones with the start of Ramadan. This was a huge miscalculation for our Muslim audience. For those who might not be aware, Ramadan is a month of the Islamic calendar and this is when people fast between sunrise and sunset, consuming no food or water. I'm not even saying that alcohol is prohibited at all.

[00:22:25] And another example, at times we could end up in a situation like this: proposing the Wisdom of the Husband[?] book to go as a gift pair with a five years birthday card for a little girl. I'm not quite sure that they need this type of information at five years yet.

[00:22:46] So, to round up everything that I told you right now, I would like to go back to my initial question of this presentation: is personalisation an illusion? And today I believe the answer to this question is yes. Personalisation seems to be just an extreme segmentation of data. Good personalisation can be done due to collecting mountains of personal information about your customer, and due to lack of this data, that one-on-one personalisation is challenging for a large majority of companies to do well today. Over time I think we will get so good at segmentation and how we action around it that it may eventually become that one-on-one personalisation that we are all striving for. Maybe we will get so good as that offline shopkeeper who knew his customers so well and could anticipate their needs at any time.

[00:23:40] So, just to summarise my main learnings from this presentation to leave you with, and they sound as the following. Start small and iterate to improve, with that one area to focus your intentions around making the personalisation experience. Focus on what users find the biggest struggle and help them around it. Don't forget to use the wisdom of the crowds to surface the right item back to your customer. Relevance to the users should go over everything else, your users will appreciate this. And don't be afraid to fail, you have to learn from it just like we've done.

[00:24:20] If you have any questions about my presentation, I will try to answer them during the live panel with the name Progressive Personalization: Designing a Better Experience. This will happen on 6th October on the main stage at 2:20 PM. I will see you all there. Thank you for your attention.