Challenges building a SaaS SEO from the trenches: SEO and UX at Scale. Allies or Rivals?

Enric, Josu & Marcos
DWX Team
May 17, 2021
15 Min read or listen

When working on a project like DWX, our primary goal is to discover how machines can help us tackle problems and accomplish big tasks that escape our supervision. We must also be aware that there is a risk of distancing ourselves from the individual, that we can forget what defines us and what differentiates each one of us.

Optimizing the conversion of your organic traffic means aligning the users' moment and their search intention with what we propose on our site and the way we do it.

This is easy to write and understand, but it poses risks and implications that are not always obvious when done on a large scale. And it is our responsibility to know this fact to give an answer that is efficient and fair with all the people involved. Therefore, if we delegate our work to machines, we don’t want them to make the same mistakes on a larger scale.

Josu Tellaeche invited Marcos Herrera and Enric Viladoms, Head of Product and Co-Founder  and UX Lead at DWX, to reflect upon this topic to an open debate that we want to share with you.

In this session, we want to address the relationship between SEO and UX when the scale of our site does not allow us to have everything under control manually.

I would say that both teams have been working in a less isolated way for a while, although there is still room to improve mutual understanding and collaboration—even more so when the size of the site shoots up in terms of URLs volume and DB items.

In that case, both teams have to live together—and make amends (if necessary)—to face the challenges of a greater enemy jointly.

It is time to tackle together the challenge of managing a live site with millions of URLs that responds to search intentions—both in external engines and in our search engine—and see how we align this purpose with our business objectives.

It is useless for me to manage some pages as landings if later I will not be able to convert because something does not respond well to the search intention, we haven't associated the main keywords with what the user expected to find, there are not enough DB products, we have not defined well the proposed internal navigation... or if you later want to continue the search within the site and our search engine does not respond correctly—it is not able to anticipate or facilitate the queries.

I think that all these challenges at scale make SEO and UX work more coordinated than it is currently internalized in many companies.

Usually, UX and SEO professionals are confronted on these issues because we occupy the same space. Navigation through internal linking has a brutal impact on the user's own experience, on how the user will discover the content of each page. And that is the space we share with UX.

In the end, when we are talking about internal linking, we have to understand that we are presenting some content; that we are suggesting. As if we were some kind of sales agent 2.0.

Let's take the example of a store! When you go into a shop, you have a person who attends you; who advises you on what you will buy.

In a hardware store—where to be honest, I never know what I am doing, and I need help—a person comes, sees what needs you have, gives you advice, shows you a catalogue, takes you where you want. And I believe that the links and the content fulfil that function as well.

If you have an e-commerce homepage and present many products at first, you can overwhelm the user—who has just entered the store and does not need the content you are showing at that moment.

As a user, if I have just entered your store, what I want is for you to guide me, identify my needs and suggest the links that will take me to the site where I can answer my questions. And I need you to do it little by little, like a good sales agent.

I think the sales agent's example is a good one; I like it. It is just that the dilemma comes from the fact that we talk to different people. UX professionals speak and define with the user in mind, and SEO professionals come down on the side of the robot—of a search engine with a code-behind—and pursue everything that encourages more people to enter the site. 

SEO seeks more volume at scale, responding to search needs to display the results when the time is right. There is no discussion there.

The recommendations and improvements usually made are always related to data; you can rely on data in SEO.

When defining the user in UX, you must directly trust users, asking them if everything is going as expected. For this reason, the needs of each department are different, and it can seem that we are sometimes competing.

The sales agent's speech looks pretty appropriate to me, but in many cases, you say: we are going to this category page, we need to distribute its authority because it is very well positioned; we are going to create links to other categories in leading positions; we will put the most viewed products and the breadcrumbs so that the user knows where everything is, and then we will start showing products.

In other words, you have just landed looking for something, and you can go anywhere.

In some cases, links are related, but in other cases, they are not. If I come for a hammer and you show me saws because they are having a lot of success. That competes with getting a good user experience. 

In a secondary positioning, once the products in that category have been viewed, displaying modules—such as related or similar products—makes much more sense.

Although I think that sometimes we place too many links for reasons that may seem somewhat forced.

The sales agent’s example continues to be the content hierarchy that we usually do from UX.

For example, in a network of schools that we have as clients in our agency—in their homepage—we did what we always do in UX; we understood the needs according to the landing used to invite someone to the site.

When we analyzed the sector, we realized that most of the schools and universities homepages were designed and directed at recruitment—when the part of recruitment and enrollment is done at a particular time of the year. For the rest of the year, the visits are recurrent from users who want to consult a specific type of content. But instead of providing that well-ranked content during the year with the enrollment campaigns working on their landings, the homepages were not designed for the daily user.

Of course, when you do this type of exercise, it is very well received by users. If I go to the web to see the school cafeteria menu and the events calendar... I do not want to be overwhelmed with content about how wonderful you are, your facilities and more. I already know them.

In any case, this is still a controlled approach if we talk about the page volume level, and you can undertake it in the form of research.

In both teams, you investigate the user. But at the UX level, you focus more on the qualitative aspect, identifying risks and stoppers that can be found and anticipating errors based on what the homepages need us to show to users. And at the SEO level, you focus on searches, on how those users are using search engines. 

They are still two parts of the same investigation.

I really like that example.

It is true that, in SEO, the criterion of "the more links, the better" is usually used, but it does not have to have the negative impact that you mention.

If we go back to the sales agent's example... Why not remove some links?

If I enter the 'About Us' section of a site, for example, do I need to enter again?

Depending on where the user has entered the site and what they do on it, this 'About Us' may or may not have to do with it. This example, which is easy to understand, is clearly extrapolated.

For example, if I'm in the 'Dog Food' category, I probably don't have to look at cat food, but all other dog products.

You may not have to show me all the links to other animals either, but you should show me a single link to see them if I want to.

This is arguable at the UX level, but it defends itself quite well at the SEO level because it is a silo-linked architecture. I link from top to bottom and eliminate links to other categories to funnel the user. This involves removing non-relevant links and adapting links to user behaviour.

You have said something that I think is the key. We take things for granted because we believe it makes sense for everyone, but it is not necessarily true.

If you have dogs, you are not interested in cats. OK. But many times, the categorization is not so obvious and, therefore, we use UX exercises and techniques—such as card sorting, which understands how the logic of product group categorization that we can have from users works.

Categorization is not always obvious. And, if you do not establish rules that follow a clear logic if the site scales, that can be considerable chaos because all do not always share a single criterion.

You have to find the structure that works best for the most significant number of users, assuming that there will always be someone who can interpret it differently. 

Card sorting resembles a model that is being worked on in the SEO community—the Next Best Action. We are talking about data-driven links.

Based on the data that I have of users, I will decide which links I present in a page type to guide them to that specific conversion.

You use predefined behaviour models to understand the next best action for a user in a given cohort. You can only do that on sites with a lot of traffic.

This technique will become increasingly important in the years to come when we no longer use cookies and rely on cohorts.

Next Best Action is a common area for UX-SEO. SEO data is entered—using two algorithms—to understand the relationship between the contents of different pages. These algorithms are Tf-idf (Term frequency - Inverse document frequency) and LDA (Latent Dirichlet Allocation).

These two topic extraction models allow us to identify the distance between the semantics of two specific URLs.

This overlaps with what Enric said. If there is a lot of distance, you can break the link. That solves the automation that you noted, Josu, shortening the distances based on these two models.

You have to find the structure that works best for the most significant number of users, assuming that there will always be someone who can interpret it differently.

Just no one can implement these techniques. It is difficult to get out of the standard of 'I have the main navigation menu with these categories and—it does not matter who enters or at what moment—it will remain the same.'

In the case of a pet food store. What are you going to do? Are you going to hide the rest of the categories from users because they entered the web for a specific animal?


That is also arguable.

Look at the case of trade marketing: how the shelves work in supermarkets and how the circuits make you explore because they know that a lot of the purchase is impulsive, not programmed.

Imagine that they applied the same techniques. "I know this user has four things on his shopping list, so that I will remove everything else." Do you know how much that would affect an average shopping basket?

I know there are many differences from that example. But by trying to make the user do what you want, you are paying the price.

We have to find a balance between making you dizzy with too many options and oversimplifying. You have to leave the door open to other types of shopping... I may have entered for dogs, but I see that you have fish. And look, I want to buy a fish.

One of the risks of customizations is that they can come back to haunt you if you do not do them very well. If something does not work, the user perceives it, and that is detrimental to UX.

There is no clear path to customization. The costs and the risk of your users penalizing you are high.


You have to test everything. And we are not talking about removing links, but about hiding them, presenting them in another way. You have to test the correct form.

There is a big difference with trade marketing because you are captive in a physical store, and they manage you as they want. In an online store, I cannot make users browse all the categories because they will leave.

Everything is more immediate. What e-commerce wants is that you buy in 30 seconds, that you do not have time to think that perhaps Amazon is cheaper. It does not want you to think about how long it would take for the product to arrive from China.

Yes and no. It is clear that when you have customers in a physical space, you have them in a captive point. And yet, you also have to take care of them throughout their experience. But I do see similarities between trade marketing and UX hierarchical organization of the content.

We have to adapt as much as possible to users. But we also have to leave other options for them to discover things beyond their initial intention.

The first thing is to cover that initial intention. Then we have to broaden the focus of the users. They need to be able to find out where they are and what other options they have. We must not mislead them but give them a little breadth.

Many times people are not clear about what they want. Users have a need, an intention or a desire to give themselves a prize. But they still need to define everything a bit more.

That is when you have to guide them. But you are not always clear on how you have to personalize that process.

I think it is challenging to find the balance in eliminating noise—in not doing it excessively. 

All of this gets more and more complicated by doing it at scale.

We have to adapt as much as possible to users. But we also have to leave other options for them to discover things beyond their initial intention.

It is impossible to define precisely where to stop. What you can do is understand your type of website. If you have a horizontal website—where many different topics are dealt with—you will surely end up in one way, and if you have a vertical one, you will do it in another way.

If it is a unique product—such as training courses—you have to directly remove all the links and funnel. The type of web marks a lot the strategy to follow.

I share this vision. It is not very easy to package and define at the level of a single user. Specifying general types of users loses more and more force because each user has a different behaviour afterwards.

Netflix, for example, seems to handle thousands of user types, and all the recommendations are generating sub-groups, even changing the covers.

It is increasingly difficult for me to treat users by type because we are not sure how to do it. There is a lot of data to assess, and it requires many subdivisions. 

It is increasingly difficult for me to understand the types of users that may be on a site and how I will treat them differently.

That happens unless you have historical data and the necessary volume of the same "you cannot's." And even so, you have to keep testing.

We—as users—see everything that is being tested and what remains as the standard. In Amazon or ebay, for example, you see related content and internal linking for similar purchases, semantic affinity or affinity of tastes with users with similar behaviour.

You also see more products from the same manufacturer, which allows you to level up—if you feel an affinity with a brand.

If you discover a brand that you like because you are related to its values, designs, materials and want to go back, you are like a salmon against the tide that goes up the funnel to discover what else that brand has.

And there is an opportunity to guide the user. Although it is far from the pure functionality of the product that the user was looking for, its price or its reviews... 

Because… What about intangibles when we talk about scale? You can also work on the brand there and seek to generate a connection with it. 

The same happens with the site itself, working on the credibility of the portal. We can seduce users and make them want to see what else we have because they like our philosophy, values, logistics process, etc. As a user, I will do the exercise of searching the portal because it is capturing me at the branding level.

There is the option to go backwards in the funnel, and we have to make that easier with navigation.

I like to compare the user to a salmon going against the tide. But in that case, you know that it is not only about that user—about Josu Tellaeche. It is about another 80 million people with the exact shopping habit. And I have to see how I work for those users who want to go up the funnel.


That is very good because it makes life easier for me. I may be at a point where I want that to happen, but I have also created my criteria. I learned the hard way, searching and searching—many times outside of what corresponds to my tastes.

If you personalize everything for me, I will never have a chance to shape my criteria. Everything is shaped by learning from opportunities. And if you remove the friction, my criteria will be flat instead of polyhedral.

That is one of the risks out there. The risk that we end up without thinking because everything we do is so chewed up and related to the above that in the end, it leads us to radicalize our positions, our way of thinking.

It is interesting because we are beginning to think the opposite that we stated at the beginning.

That sales agent 2.0, which will eliminate your links and the risks that we are commenting on, is telling you to test—even to accept that you will leave out some users.

Your conversion may increase if you assume that you can lose users who previously converted. You will have to balance and decide if you can come up with another strategy for those users or if you will simply assume that you lose them. Hyperfunnelization has some costs.

When I am removing links at the SEO level, I am doing it from pages that were previously receiving traffic—or not—and authority and that now will no longer do so. Eliminating links can affect the user and can affect the performance of that page, whether or not it had results.

That is a power and a risk, especially when you enter AI, where you are already seeing concrete examples of how biases are being transferred.

In the end, the opinion of the programmer is being transferred. When it comes to establishing the criteria and learning areas of model training, biases can become norms if, later on, these intelligences decide what other users have to see, understand or learn.

And, of course, the principal opinion is still the one that comes from the same people: heterosexual white men with a wealthy life.

If we look at it within the world of machines that make decisions at scale, evil will continue to be transmitted in that sense. And if that is going to make us close doors so that people can see another reality... I know I'm getting philosophical here, but that's the reality. 

In any automation and artificial intelligence, which ends up making decisions that you program, your biases will be reflected. And that has a substantial impact on who will receive the content or will be guided.

Following what Marcos said, we can isolate users who are not relevant enough for us on our site. But if you leave them isolated, it may have an impact on your business. 

Perhaps it implies that an area of your business ends up discontinued without an apparent reason. Imagine that some people—who work in that area—are affected because it was decided that their work was not going to be visible at a specific moment.

The product stops working on the market. And those people will lose their jobs.

Your conversion may increase if you assume that you can lose users who previously converted.

In any case, we have to make decisions at the business level. If some categories have to be eliminated for the greater good, this must be done. You do not have to be afraid to make those decisions.

I am not talking about being afraid; I am talking about being fair. When you have responsibilities, they come with a heavy burden. You have to try to do what is right, not to make no decisions.

I am talking about not applying your own biases to be fair when weighing, knowing that someone will lose.

Users who don't convert are a cost; they consume bandwidth and are not welcome. If they do not contribute to the conversion in the funnel, they are not welcome on the web. Traffic that is not converting is not welcome.

Apart from the fact that they are consuming resources, there is a probability that they will speak badly about you, your catalogue, your products, what you offer... And that has an impact on the brand. So you have to make an effort to try not to receive these types of users or to be able to discard them at an opportune moment. It is hard, but for users who do not convert, we can try to convert. If it does not work, they go away.

We have to find the right level of affinity. Returning a little to the beginning topics, we have to understand the search intention well to only connect with people with a clear intention that we can respond to.

There needs to be an alignment between what you are looking for, when you are looking for it and what we show you so that you end up converting within our site.

We should not look for poor quality traffic that comes from any source. If we do this, we will have to test and fine-tune. 

In the end, users are free and can do whatever they want, but we have to look for that affinity.

In the real world, it is the same. You do not want people to come to your business for the wrong reasons—to use free Wi-Fi, for example—or to create discomfort in others.

These people will complicate the owner's existence. They will be occupying a place that others should occupy; they will generate noise, etc. There has to be an affinity. I offer something, and if you do not like it, go.

But how do you do that user filter?

Identifying the taxonomies that do not have a good performance. If there is one that is bringing you traffic that is not relevant, you attack it at the product level, at the structure level, at the SEO level, at the UX level, and ultimately you make a decision based on the core of the product. This happens a lot in SEO. SEO brings you traffic, and UX tries to get the user to convert. But SEO can bring you shitty traffic.

Become a beta tester
by joining the DWX BETA Program

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.