TL;DR: An effective personalization strategy should be treated as a design problem with a structured framework of ideation, implementation, and evaluation. Co-creation workshops are used to brainstorm audiences and personalized content based on credible hypotheses, which are then implemented and continuously monitored to assess the performance of the personalization strategy and individual campaigns. A good strategy focuses on UX while helping with achieving business goals.
Originally published on konversionsKRAFT.de. This article is a follow-up to a webinar held by konversionsKRAFT in cooperation with Dynamic Yield.
With current advances in machine learning and an ever-growing amount of shops and products that are available online, personalization will be one of the main drivers of successful e-commerce businesses in the coming years, and will, therefore, play a huge role in providing optimized user experiences. Even though many users are rather wary of being tracked and analyzed online, from surveys, we (at C&A) know that to a certain degree they still expect to be provided with personal recommendations and the best-fitting products for their needs. Or, “users’ general attitudes about tracking are often at odds with their comfort in specific situations”, as Melicher et al. (2016) put it. Besides, there exists a plethora of analyses showing the positive effects of personalization in e-commerce. I’ll leave the googling up to you.
So, personalization is really important for your online shop, but getting started with it can feel a little overwhelming. I know this first-hand. What are the specific audiences you should personalize for? Which content should (or could) you provide for them? How do you get all your stakeholders on board? Should you implement everything yourself or just go with one of the many available personalization platforms out there? I can’t really answer these questions for you, but with this little guide, I hope I can at least provide you with a framework for answering them yourself. And luckily, I don’t even have to invent such a framework myself because after all, personalization is yet another design problem …
Step 1: Ideation
Goal: Identify interesting audiences and personalized content for them, and generate buy-in.
Personalization campaigns comprise two core components:
- well-defined audiences and
- exclusive, personalized content for those.
Let’s assume you’re an e-commerce manager who is operationally in charge of creating and implementing a personalization strategy. One way to tackle this would be to lock yourself in in peace and quiet, brainstorm different audiences and corresponding content, and implement and test all of those. Now, this example is a little exaggerated, but it highlights two important issues: Firstly, you’re most probably not a subject matter expert (SME) for potentially interesting and worthwhile audiences; and secondly, you would lack buy-in from the rest of your team and department, which is crucial for Step 2, Implementation.
Therefore, what we do at C&A are, among other things, co-creation workshops—e.g., in the form of Design Jams—that mitigate those issues. Participants of such a workshop could (or should) include:
- SMEs for audiences (e.g., our merchandise team), who already have deep knowledge about which customer groups are more valuable than others, e.g., based on order volumes or life-time value;
- SMEs for personalization campaign implementation (e.g., our UX team); and
- SMEs for providing personalized content (e.g., our marketing communications team).
The important thing here is not the specific teams, but the presence of the different SMEs, so that everything is represented: audiences, implementation, and content. In this context, personalized content doesn’t always have to be images and copy. It can also be different prices for different audiences, exclusive services, vouchers, CO2 compensation, or other benefits.
As for the co-creation workshops themselves, we usually like to split them into two phases: audiences and content.
For phase one, we form (cross-functional) teams and have them come up with audiences that might be worth to personalize for, without paying attention to the specific content yet. Audiences are first specified in a high-level way and then based on specific signals that can be tracked in the online shop. To give a non-fashion-retail example, an audience “users that are interested in indie rock music” could be defined as “all users that have browsed the category ‘indie rock music’ at least 3 times in the past 30 days”. Very roughly speaking, there are two types of audiences that are worth looking at from a monetary perspective:
- Small segment size, but above-average CR and/or AOV: In this case, personalization can be employed to push more users into the audience since the audience is already relatively valuable.
- Big segment size, but below-average CR and/or AOV: In this case, personalization can be employed to increase purchasing motivation as well as up- and cross-selling.
For phase two, the cross-functional teams remain the same and everyone presents their audiences, how they are defined, and why they would be worth personalizing for. Subsequently, the audiences are randomly assigned to different teams who then ideate corresponding personalized content, based on credible hypotheses. Optimally, those hypotheses are supported by evidence such as analytics figures, existing research insights, heatmaps, etc.
Et voilà, you come out of the co-creation workshop with a well-founded set of audiences and content that can be prioritized and selected using whatever method you like. If time permits, we usually like to do some dot voting or “buy a feature” at the end of the workshop (cf. 5 Design Methods I’ve Successfully Applied as a UX Manager at C&A).
Step 2: Implementation
Goal: Create and provide personalized content for an audience and implement it as a personalization campaign.
Step 2 is probably the most work-intensive, but also the one I can write the least about. Generally, it comprises two sub-steps:
- Provision of personalized content.
- Technical implementation of the personalization campaign itself.
However, both of those are largely dependent on your individual setup. Regarding the provision of personalized content, in the optimal case, one or more responsible SMEs were already part of the co-creation workshop to ensure the ideas being implemented are actually viable. For instance, if audience-based pricing is not feasible with your technical setup, it shouldn’t be an outcome of Step 1. If a personalization campaign is viable, the team(s) in charge of providing the necessary content are already part of the process and (in a perfect world) don’t require additional briefings.
There are two possible relationships between campaigns and content:
- Campaign follows Content: In this case, personalization campaigns are defined based on the content that’s already available (e.g., simply hiding certain images or copy for a given audience). This is good for quickly getting started with personalization, but has deficits in the long term since existing content usually doesn’t address more specific or unusual audiences that can have huge potential.
- Content follows Campaign: This is the optimal case in which potentially valuable audiences are defined first and the personalized content is provided specifically for them, for an improved user experience.
Regarding the actual technical implementation of the personalization campaign, anything is possible—from coding everything yourself to simply buying a ready-to-use solution—as long as you can ensure to correctly track the appropriate signals to identify your audiences and then provide the corresponding content exclusively to them. Signals might include add-to-cart events, newsletter subscriptions, visits to specific categories, usage of a specific filter, etc. We at C&A are working with Dynamic Yield as the underlying platform (hence their participation in the webinar) while our UX team is responsible for setting up audiences and personalization campaigns.
Step 3: Evaluation (& Iteration)
Goal: Evaluate the success of your campaign and your overall personalization strategy.
Once you’ve implemented your personalization campaign, the really interesting and fun part starts: Does your idea actually have a positive impact on users and their behavior? We monitor campaigns on two levels:
- through a local control group and
- through a global control group.
The local monitoring works like with any A/B test. Often, we implement several variations for a personalization campaign (e.g., different mixes of copy and images, different touch points) and always include a control group to be able to draw a statistically valid conclusion about whether the campaign works or not, based on a primary metric like add-to-cart events, conversion rate, or average order value. Here, it is advisable to also perform “audience drill-downs” and make sure already well-performing audiences (big segment size, above-average CR/AOV) are nor negatively impacted.
On the global level, we have implemented another control group (called the “global control”) which doesn’t get served any personalization campaigns whatsoever. Its purpose is to estimate the combined effect of all active campaigns based on one or more KPIs. In this sense, it’s set up like a classic A/B test, with the combination of all campaigns being the variation. This was done based on advice by Dynamic Yield (and specifically their Personalization Strategy Lead, Kamal Karim) and allows us to evaluate the performance of our personalization strategy as a whole.
Moreover, our monitoring and the fact that we almost always test different variations per personalization campaign have a very important side-effect: The teams in our e-commerce department gain valuable insights into which (types of) content resonate well with which audiences, so that they can learn and iterate. Those insights are also fed back into co-creation workshops to ideate even better campaigns in the next round (see Fig. 1).
As I said at the beginning of this article, personalization is already becoming one of the main drivers of successful e-commerce businesses. To prevent fishing in muddy waters and enabling an efficient setup, it is necessary to treat personalization as a design problem with a structured framework of ideation, implementation, and evaluation & iteration. This is what we have done at C&A, where I have led the efforts to build and implement a personalization strategy from scratch. Moreover, personalization must provide users with recognizable added value to increase acceptance for data collection.
However, it happens all too often (and too easily) that personalization solely focuses on monetary aspects and “classic” e-commerce KPIs such as conversion rate or average order value. Yet, its primary concern should be to contribute to a user experience that’s as optimal as possible. Therefore, while those traditional e-commerce KPIs are often available out-of-the-box, we have started to look into possibilities to also evaluate our personalization strategy and individual campaigns using more UX- and usability-centered metrics (such as SUS, Inuit, or AttrakDiff). For instance, first steps have been taken to measure the NPS (which is a reasonable proxy for usability) of different variations of personalization campaigns, in order to focus on longer-term retention effects in addition to short-term revenue improvements.
To conclude, a good personalization strategy should strongly focus on your users’ experiences while supporting you in achieving your business goals.
Calls to Action
- Read Sleeknote’s Complete Guide to E-Commerce Personalization—I think it’s a good follow-up with some more specific ideas for campaigns.
- Take a closer look at your favorite website and think about how you would personalize it to improve your personal experience. Write down answers to the following two questions:
(a) What is a specific audience you’re a part of?
(b) What could a piece of personalized content for this audience look like?
- No matter if you’re already engaged in personalization or not, hold a co-creation workshop. It’s always good for discovering untapped potential.
William Melicher, Mahmood Sharif, Joshua Tan, Lujo Bauer, Mihai Christodorescu, and Pedro Giovanni Leon (2016). (Do Not) Track Me Sometimes: Users’ Contextual Preferences for Web Tracking. In: Proceedings on Privacy Enhancing Technologies.
Thanks to Kamal Karim for proofreading. 🙂