Getting Started with Personalization at Your Enterprise

Gabriella Vas Gabriella Vas
7 min read | June 16, 2021

So your organization is considering implementing a personalization strategy (or updating the existing one). Maybe you’re grappling with the “build or buy” dilemma. Or perhaps you’ve chosen the latter option, that is, to outsource personalization to a provider. 

In which case, congratulations, what an exciting new prospect! But presumably, there are some burning questions on your mind. Is setting up a personalization system a long and convoluted process involving multiple stakeholders? Is it insanely expensive? Does our data and tech stack even meet system requirements? To put it bluntly: 

How much of a pain is this going to be?

Well, declaring that implementing personalization is a breeze would be a bit of an overstatement. But the good news is, there is a precise, multi-step choreography to this process, led by the personalization vendor of your choice.

Let us walk you through it, so you get a better idea of what to expect. 

Making a Pitch

Getting a feel for personalization begins with understanding what’s in it for you. When a potential client expresses interest in using a recommendation system, the Customer Success team at Yusp explains the added value of personalization and how it could best be leveraged by this client-to-be, based on a thorough understanding of and previous experience in the given domain.  

Meanwhile, we also make the case for choosing Yusp in particular, as a personalization vendor with a robust scientific background, serving enterprise clients in various industries around the world. 

Then we provide a so-called personalization roadmap that identifies personalization opportunities specific to the prospective client’s business or industry, using the client’s actual digital assets as examples. Decisionmakers can then pick and choose whichever of these personalization elements best fit their strategy. They can also add their own requests or suggestions.

Preparation and Assessment

Once we’ve got the bases covered, it’s time to define the project scope, set up teams in charge on both sides, and agree on the practical details of the collaboration. But in order to formulate what needs to be done, first we must have a clear view of business requirements – the client’s objectives and how we can serve them –; and of the “raw material” and the tools we can work with – that is, the client’s existing data and technical infrastructure. 

Let’s look at these two in more detail. 

Locking Business Requirements

This step involves both the personalization vendor (in this case, Yusp) and the client, because it needs expertise in personalization as well as a thorough knowledge of the client’s business domain. 

Together, we specify the business objectives of implementing a personalization system, and by what means we intend to meet these goals: 

  • Relevant personalization product modules, such as item recommendations, advanced search, on-site personalization, etc.;
  • Where the recommendations should be displayed – at what point in their user journey the consumers are to encounter them;
  • Algorithms to be used, such as collaborative filtering; rule based, popularity based or deep learning based recommendations, etc.;
  • Business rules overriding or complementing the recommendation logic for each placement; and
  • KPIs – the relevant metrics for tracking the performance of personalization.

The resulting, detailed document is called the Business Requirement Definition (BRD). Approved by the client, it will serve as the basis of the evaluation of the personalization engine. 

Data & Tech Stack Check

Done by the Yusp team, this is an assessment of the client’s technical infrastructure (their data and IT architecture) as is. Some of this information is obtained through a questionnaire, but the bulk of it comes from examining data about end users and about the products or content we can recommend to them.  

Potential clients often worry that their datasets cannot provide a viable base for personalization because they’re incomplete or unstructured. Rest assured: although the law of “garbage in, garbage out” applies to machine learning-based personalization as well, it’s usually pretty easy to start collecting the data needed for personalization in case you don’t have it or what you have is a mess. And algorithms can work even if you have just a few weeks’ worths of data, so the implementation won’t be delayed.

At the very least, we need the following two categories of data: 

Item Catalog 

This is a list of products or pieces of content (generally referred to as data synchronization items) available on the client’s site that can be recommended in a personalized way. 

These items are identified by a unique ID and they can have as many custom attributes as required, depending on the client’s industry. 

  • For instance, in e-commerce, some of the most common item attributes are Category, Title, Short Description, Price, and Image URL. 
  • In video streaming, key attributes include Length, Genre, Rating, Director, Actor, and Subtitles.

User Behavior Events

Personalization engines need a 360-degree view of the customers, incorporating information from several sources, most importantly, visitor behavior tracking

In e-commerce, trackable user actions include

  • product page view; 
  • product added to or removed from cart; 
  • product saved, shared, or bought, etc.

In video streaming, some examples of relevant user actions are

  • view – when the user has viewed the video detail page;
  • watch – applicable when a certain percentage of a video has been played

These user actions are measured in both industries: 

  • search
  • rate
  • add or remove from favorites
  • click on recommendation.

At the end of this phase, an Assessment Document is produced by the Yusp team, as an inventory of the client’s data and IT infrastructure. This will serve as the technical basis for the next step, Data Integration Design. 

Data Integration: Design and Testing

This is where the actual work begins. Having assessed the existing technical environment, we propose what needs to be built in order to run a personalization system that meets the objectives set together.

Based on the learnings of the Preparation and Assessment phase, the Yusp project team delivers the Integration Design Document (IDD) – a detailed guide to the integration process, a sort of reference manual tailor-made for the client team. It includes all essential information about the applied endpoints and solutions, complete with examples. It describes the method of implementation – for instance, what data types will be used, how Yusp will receive the data, and how the recommendations are sent back. 

Once this document has been approved by the client, the testing phase begins. It consists of

  • delivering the actual implementation tasks on both Yusp’s and the client’s end, such as setting up the data connections, starting the data synchronization, and configuring the personalization system; 
  • verifying that all has been done according to the business requirements specified earlier; 
  • and testing the functionalities of the personalization solution before going live. 

Data integration and testing can take a few weeks, as agreed at the end of the Assessment phase. Its length depends on how much of the existing technical assets we can use as is, and what elements of the personalization solution need to be built from scratch.  

Proof of Concept Period

As the saying goes, the proof is in the pudding. For this reason, first, we test a limited scope of the personalization system on a sample of end-users and measure its added value in terms of the KPIs set in the Business Requirement Definition. 

This can take the form of an A/B test – for instance, if a client wishes to test their existing personalization provider against one or more competitors, or if they prefer to compare two or more personalization solutions before going ahead with either of them, this is the time to do it. If they’re new to personalization, A/B testing the default setup against a personalized one can help them measure the added value of personalization. 

However, A/B testing is not an indispensable part of Proof of Concept. Sometimes, PoC merely means checking if all the nuts and bolts of the personalization engine are working properly to provide an enhanced user experience.  

Based on the results, we finetune the configuration in order to maximize return on investment or any other aspect of personalization our client can benefit from. At this point, we’re ready to roll out the full scope of personalization to the audience at large.

Ready to Engage?

From proposal through preparations to integration and testing, these steps are crucial for laying the foundations of a lasting, mutually beneficial business relationship. As you can see, they’re clear-cut and well-documented, yet flexibly adaptable to your circumstances and needs. 

So, how long, complicated, and expensive does setting up a recommendation system get? It depends on the complexity of the requirements, and the scope of personalization, but all in all, it can be achieved in a matter of weeks. An experienced and competent vendor team can help make the process seamless and cost-efficient.  Hopefully, this overview has helped to answer at least some of your burning questions. But if you’ve got more, check out our answers here. Or, if you’re ready to take the next step, get in touch with us to request your own personalization roadmap.

What to read next

Join our newsletter

Get to know the ins and outs of personalization