A few years ago at Mediahuis, we wondered how we could stay relevant to our readers in a complex, digital world where readers have different behaviours and habits. To find the answer, we began personalising our news across different brands and improving our readers’ experience. 

While we’re still convinced that personalisation is the right path — and we had some big successes — news personalisation takes effort, the right strategies, and the right technology to make it work. It’s a process of ups and downs. 

Here’s what we learned about news personalisation and what our next steps will be. 

1. Take a holistic approach to personalisation.

By greeting the reader with the relevant article and providing a seamless transition to the next one, you keep readers on your platform longer. People who stay longer are more likely to subscribe and less likely to churn. We take a holistic personalisation approach using a full-suite personalisation engine powered by AI. 

Here’s how it works: Froomle technology (a full-suite personalisation engine provider) is integrated into our Web sites, e-mails, and video content. It runs our raw data through machine learning algorithms to recommend relevant articles to readers. The algorithms are fully tunable to a specific use case or business rule. To optimise the result of the specific use case, we add filtering capabilities, diversification, impression discounting, and take user intent into consideration. Then, we A/B test everything.

2. Forget about the homepage, focus on the detail page.

You might be tempted to start with the highest traffic point for maximum performance. The homepage, in and of itself, is all about performance. 

However, the same reasons that make the homepage a compelling place to start also make that a poor choice. If you want to grow a personalisation roadmap in the long run, you need to start small.

Instead of the homepage, begin with a high-traffic article page, get results, A/B test everything, and communicate your results internally. That makes it easier to get your team on board with experimentation and testing new technologies. 

3. Don’t be mistaken: The work doesn’t stop.

If you’ve successfully rolled out an algorithm for one newsroom, that doesn't mean it will be equally successful for another. Just as your reader audience is not the same across every newsroom, the algorithm is not the same for all our news brands. We learned that when we created the exact same fully personalised newsletter for three of our brands — one popular and two regional papers. 

When we saw the first results with our popular brand, we were excited. First, we generated 110% more traffic from these personalised e-mails compared to the control group. Secondly (and this is important when talking about personalisation), not only did we succeed in building more traffic, but also in offering a more diverse range of content to our readers.

We thought this was going to be easy. We’d just roll out to all our brands and be done. 

When we started rolling out to our two regional news brands, we saw different — even negative — results in some cases. That’s when we learned that we cannot assume that all news brands are the same.

4. Stop showing the same article over and over again.

As publishers, we produce so much content — every day, again and again. We have a lot of quality journalism, and publication teams struggle daily to push those articles into all our channels.

We saw an opportunity to make our channels smarter through personalisation. Our lists on article details were always based upon the same general settings: most-read, more of a section, etc. Even more frustrating, these lists show the same articles — even if you have read them — over and over again. As a consumer-facing company, this meant we weren’t relevant enough to our reader and we weren’t creating enough value. 

Personalising those lists was only part of the answer to this problem, because even with personalisation, some articles kept getting recommended. What truly put us on the road to better results was impression discounting.

We discovered that, after the third time, you show the same recommendation, the probability that somebody clicks is maximal. It means people are aware of the article you recommended. After showing the article eight times, however, the probability of clicking that article is less than 50%. We then stop showing that article and replace it with another. 

This is what we call impression discounting. This gave us better results in terms of time on-site, and we managed to show more relevant, quality content to our readers.

5. Just keep testing.

Our data taught us it’s always better to show a personalised list vs. a non-personalised list, as we saw CTR improvements up to 50% on our personalised boxes. We learned this by continuously A/B testing groups.

Here's an example of article personalisation for one of our brands:

Approach 1 recommends randomly selected most-recent articles. Due to inferior performance, Approach 1 was replaced by better experiences early in the process. Approach 2 suggests most-read articles. 

Approach 3 is a 1:1 personalisation. It considers user behaviour, various filtering capabilities, diversification, and some business rules tailored specifically per news brand. Approach 4 is a full-suite personalisation, factoring in impression discounting. Furthermore, within Approach 4, we test several algorithms with different weights to learn which combination works best for a specific brand.

When upgrading to a 1:1 personalisation from “most popular” articles, we saw an uplift of about 48%. However, a holistic full-suite personalisation can increase the CTRs by about 60%.

For an optimal, relevant, and personal experience with your brand, you’ll need to adjust the technology according to your specific audience and business rules, then track performance by testing. 

6. Adoption is a journey.

Test small, get CTR uplifts, but also think about the next step. It’s common to get caught up in tracking all the pages’ success, but your role as a product manager is bigger than that. You need to lead your newsroom into personalisation by educating your team and following the adoption curve of all news brands. 

Educate your team with personalisation sessions; show them the possibilities and how they can implement this and use it to their advantage. At Mediahuis, we often give personalisation sessions, together with our strategic partner, Froomle. We noticed some newsrooms have a much steeper adoption curve than others. In addition to these sessions, we also track how many pages are personalised per brand. We learned that adoption really depends on enthusiasm and understanding in the newsrooms.

7. Integrate the right people in your roadmap.

Everybody on your team has a crucial role: journalists, the technical team, and the digital marketing lead of the newsroom, but only a few have the decisive power on personalisation. Involve the right people throughout the whole process. Talk directly with the final decision-makers and editors in chief, listen to their feedback and concerns, and look for answers together.

Define some clear use cases and just start doing. It’s better to start with small improvements, test, and show some of the first results to your stakeholders.

What does the future hold?

Our focus in 2020 is to increase functionality and roll out to all our channels and brands. 

In 2020 we’ll also work on personalised messaging. We believe we can make our newsletters and push messages better through personalisation. To effectively bring users back to our platform, these (brief) attention-grabbing messages must observe and meet the person's preferences, behavior, and context. 

We’re excited to work further on this and we’re looking forward to sharing some of our results.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .