In 2018, machine learning and its game-changing capacity for automated predictions of visitor behavior and purchasing trends currently dominate digital personalization. Machine learning’s flame is burning bright, but whether it can deliver the competitive edge depends on overcoming a key challenge: the need for structured, accurate, and consistent data. This is the gold dust of machine learning because the quality of every model depends on the data that feeds it. You can train and retrain a model ad infinitum, but it is doomed to failure if your data is poorly structured, inaccurate, and inconsistent.

Enter the data layer — an object of infrequent discussion, understood by few, but so fundamental to machine learning and the data-driven decisions that power high-impact and scalable personalization.

If personalization is a journey that takes us down the road to understanding customers, their purchasing habits, wider browsing behavior, and likely future trends, then the starting point on this journey is the data layer. If you set out on the wrong foot without a correctly implemented data layer that’s aligned with business requirements and goals, it will be painful and costly. You will achieve nothing more than inconsistent data, overly intensive manual data processes, and undelivered promises.

This is why it is important to begin the journey with a well-implemented data layer — not one plucked from thin air without a sound strategic basis, but one based on a well-thought-out plan, communicated across your business, with clearly defined goals. This is the key to personalization, not machine learning or AI. Those are merely tools. They can offer nothing without the business-critical information that drives them.

What is a data layer?

Stated simply, a data layer is a dynamic and flexible container for all the useful information you want to collect from each of your website’s pages. When we talk about “useful information,” we are referring to data points.

On a more technical level, a data layer is a collection of key data points exposed via a suitable data structure on the frontend of the website with JavaScript. Consisting of key/value pairs, the data layer is typically updated on key events, such as a page rendering, a login, a signup, a product view, or a transaction. The complexity can be a single variable, such as a user id, or multiple nested variables, often seen with product or transaction data.

Decide which data to gather

How do you know what data points are important on a page? The answer to that is straightforward: It’s whatever is important to your business. Answering the following questions is a great place to start.

  • What information can I gather that will help me and my business understand the interaction between my visitors, my site, and my products?
  • How can I drive visitors to my site?
  • How can I offer a truly engaging experience?
  • How can I encourage visitors to spend more on the products that I want them to spend their money on?
  • How can I make sure that visitors come back often?

The starting point for implementing a data layer is, therefore, a wider business discussion that incorporates a clearly defined strategy, encompassing business requirements and goals. Defining the “why” informs the “what,” not the other way around. In our experience, there is little or no value in implementing a data layer without a clear and well-communicated strategy that defines what we hope to achieve and what will we do to achieve it.

As an example, if the business goal (why do we want to gather the information) is to increase customer loyalty and the revenue from those customers, then a strategy of understanding who the most loyal customers are, what products they purchase, and how much they spend will inform the “what” (what information to gather from a transaction confirmation page). In this case, the following data points would be useful:

  • Transaction information that defines what consumers purchased.
  • User information that defines who made the purchase and on what device, along with their journey, product views, and transaction history.
  • Contextual information that defines when and where consumers make their purchases.

The important point to understand is that once you have decided what information is important for each of your pages to gather, a data layer provides the means to collect that information and store it in a structured and readable way.

The benefits of separating the data layer

By organizing a website’s pages into common types and emitting the same events on each of those page types, the data layer can generate structured and readable data that you can then use across your digital personalization platform, CMS tool, or any other software that your business has integrations with, whether in browser or backend systems. In fact, one of the key selling points of a data layer is that it is platform agnostic and reusable. This makes the task of convincing other stakeholders in your personalization strategy, especially those involved in implementation, of the advantages in opting for a data layer much easier.

It is worth noting that as your business requirements and goals inevitably shift over time, you can question, re-define, divide, and conjoin the data in your layer. And the separation of the data layer from the underlying page structure means that you can make changes to the data layer in isolation from the markup and will not affect it with errors generated elsewhere on the page. Engineers can, therefore, make changes to a page, safe in the knowledge that those changes will not impact the integrity of the data layer and the consistency of the generated data.

Final thoughts

Clearly, any engineer familiar with a data layer can implement one, but you can only obtain the maximum value if all the stakeholders across your business are clear about the objectives. Why do you want to collect transaction information? What product information is important to your marketing and sales teams? What is the purpose of tracking a user through an ID? If you can answer these and similar questions, you and your business will be perfectly placed to begin the task of implementing the data layer.

Some more words of wisdom: Start small and aim big. Iterate, build in complexity as you understand more about what information you need to collect to fulfill your objectives. One of the most common implementation mistakes is to shoot for the most complex and comprehensive data layer possible. Wrong — start smart and small. Gather information, understand how to use it, and what it means. Aim for deep data before big data.

Graham Cooke is the founder and chief executive officer at Qubit, the pioneer in data-first customer experiences.