Get best of the week

Apply personalization in E-commerce

At a certain stage in business development, personalization may seem like a good step to improve conversion rate optimization (CRO). Companies often try to use ready-made algorithms, relying on AI and machine learning to create personalized experiences.

Also, many enterprises begin to work with marketing tools for collecting data with built-in AI, which leads to fragmentation of experience for the client and suboptimal results for the company.

Automatic personalization must be approached carefully. Tool-dependent strategies are some of the worst - they can seriously harm your business.
Personalization is not only data and algorithms. To achieve it, you must:

  • Streamline business processes;
  • Hire enough people to create relevant content for all page options;
  • Systems that can serve content.

Full personalization requires a serious investment. And the larger the investment, the higher the risk.

Thus, it makes no sense to start with full personalization (unless you have a bunch of free time and extra money). As with A / B testing, you can start small and gradually move on to a more comprehensive approach if your initial attempts are successful.

Of course, it is good if the results appear at an early stage - this will help to quickly get the resources necessary for scaling and automation of the personalization process. This can be achieved using the same sequential steps that are used to optimize the conversion rate:

  1. Study;
  2. Hypothesis;
  3. Test preparation;
  4. Conducting a test;
  5. Rating.

The purpose of this article is to show, in a hypothetical example, a pragmatic way of transitioning to full personalization. So we will demonstrate the various stages of the process and what to expect at each of them. It will also simplify some personalization tasks.
I have already used a similar approach for several e-commerce companies. It all starts with segmentation - a process that you are probably familiar with.

Start with segmentation


Let's introduce the fictional company, The Garden Company. She threaded through various garden tools and tools, ranging from shovels to sets for relaxation. She has a large product portfolio and spends a lot of money on marketing. Also, this company is constantly optimizing its site using A / B testing.

Personalization can be a logical step for the development of The Garden Company. Many things can be personalized - for example, which channels to target, for which users, in what order, at what time, with what content, what products, etc.
In this hypothetical case, we will try to solve the often encountered problem of user interaction on large e-commerce sites - the availability of goods.

Do research to prioritize potential segments


As with A / B testing, the study identifies potentially valuable segments.

If you are involved in CRO, then many consumer data is already available to you. All employees, from product experts to analysts and employees from the procurement department, put forward hypotheses about the customer base, which are based on data.

The available information is already sufficient to determine the possible attributes for segmentation. For example, the following hypotheses are true for The Garden Company:

  • Clients can be classified as "lovers of relaxation" or "active gardeners";
  • Some customers like discounts more than others;
  • Some customers prefer expensive brands and are willing to pay more for them.

Highlighting broad groups among your audience based on data is not the same as quick selection and personalization for one customer segment. The latter approach often fails.

The method of personalization by segment is dangerous because it limits the sample size too soon. Quality personalization begins with as wide a group as possible, and then narrows down to those customers who are of the greatest value.

Prioritizing segments is best based on their expected value and ease of personalization:

  1. Create a personalization plan that lists all customer segments and the experience you want to give them.
  2. , , , .
  3. , — .




Match the proposed segments with the available data. To do this, it is not necessary to create a complete portrait of the client. Shopping statistics and some analytic information about your consumers should be enough.

“The Garden Company” uses order data for a cluster analysis of the categories of purchased products among the “leisure enthusiasts” and “active gardeners” segments. Surprisingly, the result was revealed not in two, but in three groups: “lovers of relaxation”, “lovers of decorative gardening” and “lovers of fruit growing”.



When analyzing segments, you will be surprised again and again.

Targeting objects are often not obvious. Sooner or later you will come to this conclusion. I dealt with hundreds of different companies that tried to create a personalized experience. For all this time, the original ideas for targeting have never been good.
Cluster analysis of brands shows that some customers really prefer expensive brands, while others prefer cheaper counterparts. To determine the best segments for discounts, you can set a percentage discount on all products.

Estimated segments that do not show much sales growth should be ignored.

Based on the results, The Garden Company will first compare holiday makers, gardening enthusiasts and horticulture enthusiasts. For such testing, users are classified based on their purchasing behavior and the products they view on the Internet. Using this classification, A / B testing can be configured for one of the segments.

Create a test


Select a channel to prepare a test for segmentation. The easiest way depends on the technologies and tools used. For example, if a site stores a lot of information in the cache, you may need to fix this by involving web developers.
However, everything is much simpler if you use good merchandising tools (such as Bloomreach, Algolia or Magento) that allow you to create custom segments. E-mail can be the first good channel, since it does not suffer from caching problems, users logging out of the system, and other nuances that make it difficult to analyze the test results.

At The Garden Company, testing is done by email. All classified users are included in one of the segments. Since resources for creating content are limited, at first the team creates only one version of the message for the email newsletter in addition to the main one - it is aimed at “leisure lovers”. During the first test, 50% of them will receive a regular letter, and 50% will receive its target version.

You must evaluate the effectiveness of personalized experience for a specific audience. There is only one way to do this. You need to check the content created for a specific segment on clients that are not part of it - and vice versa.

In our hypothetical case, the target version gives an impressive increase - 15% more income from “holiday lovers”. With these numbers, the project team will be ready to test other segments and identify more people to create personalized emails.



After additional resources are allocated, all three customer segments are tested by e-mail with three different letters. Thanks to this, The Garden Company will get a general idea of ​​how segmentation affects the distribution results.

Further actions


After determining the segments in the selected channel, you can move forward by choosing one of three possible methods:

  • Repeat the initial test on other channels. This is usually the easiest solution if the same customer segments can be identified in other channels.
  • . , .
  • . , .

It doesn't matter in what order these steps are performed. Again, it's best to start with the simplest.

1. Repeat the initial test on other channels.

As always, if testing yields results, take what works and zoom in. Use segments and just test them in other channels. For example, The Garden Company may use the same classification to display ads by testing its effectiveness.

Another good solution is sorting the product on the website by segments. Since emails have proven effective, it is worth highlighting a group of developers who will help create this feature.



2. Combine segments

After identifying the segments, start combining them to make the interaction with the client more personalized. You can create effective combinations of various product categories and brands (for example, "lovers of relaxation" who prefer expensive brands).

The combination of segments will complicate the process of creating content and managing tests. The number of page options will begin to grow rapidly. Even initial testing of narrower segments will require a lot of resources.

Although targeting can be of great benefit, many have not fully realized that it is always associated with organizational difficulties. Therefore, it requires large investments. Difficulty is the flip side of targeting.

At some point, having a scalable content hub becomes a prerequisite for content scaling and personalization. When content can be easily reused and combined into multiple templates, creating a personalized experience is noticeably easier.



Many tools support this (for example, Bynder and Nuxeo), but they themselves are not solutions. Most likely, you will have to change the way you create content so that you can reuse it.

In any case, segmentation requires resources to configure and manage tests. You must try hard to get them.

3. Add behavior triggers

Based on what you already know about segmentation, it will be useful to consider data on consumer behavior. "Recreation lovers" who are looking for a new grill should be targeted specifically for this product, and not for the general category created for this segment of customers.



By combining segments, you can make reasonable assumptions about which behavior when viewing your online store makes sense and which doesn't. If a “lover of relaxation” chooses between four grills, two of which are premium, and two are counterparts from cheaper brands, which products should be promoted as the main ones?

The answer depends on the user segment. If a customer prefers premium brands, he should promote premium grills and accessories. Of course, such targeting can be applied using several channels (as described earlier).

Personalization enhancement


Combining segments and using them sequentially across multiple channels using behavioral triggers is already an impressive job. At this point, segmentation begins to intersect with personalization.

However, with the help of personalization algorithms, it can be taken to a new level.

Start by adding personalized recommendations.


Based on the segments, you can configure the recommendation algorithm. The Garden Company decided to offer customers a list of recommended products based on views from other customers in the same segment.

After that, "gardening enthusiasts" began to receive more detailed recommendations in letters from the segmented mailing list. This small change within the segment makes it easy to determine the added value of personalization algorithms. Moreover, all recommendations will be relevant, since the products from them were of interest to customers from the corresponding segment.

With each new test, The Garden Company improves the algorithms based on which recommendations are made and applies them to other channels and segments.

Start with orchestration


Since feeds now display consistent content, you can start testing which one is best for each user. This can be done by contacting customers who often click on links from banners.

This is an additional form of segmentation used at a different level. With its help, you can determine the optimal amount of time after which you should send a follow-up letter after showing ads with retargeting.

Transition from segmentation to personalization


Managing all segments, triggers, channels, and orchestration will soon become too difficult. In the example of The Garden Company, three segments of product categories, two segments of brands and, for example, five segments of sensitivity to discounts were distinguished - in total 30 combinations of segments are obtained.

In combination with channel segments and orchestration stages, it becomes quite difficult to regulate the entire system. However, during the tests a lot of data is generated - you will know how users react to certain channels, order, content and time frame.

In addition, since the content is segmented, many marketing efforts are classified in one or more segments (for example, a decorative garden, a premium brand, a discounted customer who has switched from email). The combination of all this data and content classifications is ideal for machine learning.

Conclusion


The final step — AI-driven personalization — is unattainable without going through the previous steps.

Everything we talked about earlier applies to machine learning. Start small and keep exploring which approaches work best. Machine learning is just one more tool that allows you to improve on previous developments created in the process of segmentation and personalization.

More articles:


All Articles