Domain retention insights for .nz ccTLD
InternetNZ is the guardian of the .nz namespace. We strive to make .nz a safe and reliable space for everyone. Understanding domain retention allows us to understand the current market and predict future demand. As the .nz namespace has reached a plateau of growth, improving domain retention has become more important.
This article provides insights into domain retention of the .nz ccTLD. It will begin with an overview of the current .nz register’s domain age distribution and year-on-year growth rate. It will then examine the relationship between domain age and retention rate, including a view of domains created during the COVID-19 lockdown. Next, it will explore the key factors influencing domain retention and examine their relationship with domain retention over time. Finally, it will analyse the relationship between the registrars’ business types and their performance in domain retention.
Directly jump to the conclusion.
So that you can better understand our analysis, below is a refresher of some of the concepts associated with the operation of the domain name register:
- Domain name: a domain name is an identification string that allows customers to reach the services of a domain holder, such as email, web, eCommerce, etc.
- Registrar: a domain name registrar is a company that operates as an intermediate between end-users and the registry.
- Registrant: a registrant is a person who registers domain names.
- Register: register is the whole .nz domain space.
- 1 year matured window: most of the domains are with a yearly subscription. Therefore, it is most accurate to check a domain’s retention when the time between its registered date and its analysis date is over one year. If a domain’s registered date to the analysis date is over 1 year, we say the domain satisfies the 1 year matured window.
Registration age distribution
We found from the register that many domains did not renew. Most of the domains use a yearly subscription cycle. Therefore, as Figure 1 shows, 31% of the domains only last for 1 year, as the registrants did not renew the subscription.
(* Only churn domains were included in the below two charts).
The cumulative percentage chart at Figure 2 shows the register coverage on a specific domain age. For example, 45% of the domains only make it to 1.75 years old or less.
New registration growth rate
The new registration Year-On-Year (YOY) growth rate has reached a plateau, as we can see from Figure 3. However, the lockdown in New Zealand due to COVID-19 in April 2020 boosted the demand for the Internet and new domains. We can see on the chart that there was a 46% YOY growth in April 2020.
A domain’s current age is the most important feature for predicting domain retention. Cohort analysis is a type of behavioural analytics in which you group your users based on their shared traits to better track and understand their actions. Here we used the domain’s registration date to form cohort groups to explore domain retention patterns in moving monthly windows from January 2016 to June 2021.
As most of the domains are with a yearly subscription, the retention pattern has a yearly cycle. The churn rate decreased over time as the domains matured as Figure 4 shows. 12 months after registration, at the first renewal time, 30% of the registrations churned, while 70% retained. Churn rate decreased from 30%, 15%, 10%, 5% respectively at each renewal.
The retention pattern has been very consistent over the past few years. Each cohort has a similar retention rate. The April 2020 cohort group consists of domains registered during COVID-19 lockdown in New Zealand, and the chart shows that it has the same retention pattern as the other cohorts.
Key features related to domain retention
Feature reduction, or dimensional reduction, is a technique that uses fewer features to represent the meaningful properties of the original data. Reducing features brings many benefits, such as avoiding sparse data which makes machine learning models hard to reach statistical significance.
A decision tree is a commonly used method for variable selection (dimensional reduction), assessing the relative importance of variables, making predictions etc (). The algorithm is non-parametric, which is perfect for our skewed data.
In this section, we used a decision tree to select important features by reviewing features’ coefficients. We then measured the features with Odds Ratio, which will be introduced later. The below features were explored in a decision tree, with the most important ones highlighted in bold.
- Parent Group (registered under co.nz, net.nz, org.nz, nz, etc.),
- whether the registrant is using the privacy option,
- domain name length,
- email name length,
- registrant type (organisation, person),
- do they use free/paid email（free emails: .gmail.com, .hotmail.com, .outlook.com, .yahoo.com, .xtra.co.nz, .icloud.com),
- do they have a secured website (has an HTTPS site),
- DNS activity (number of DNS queries received).
Coefficients from the decision tree are convenient for data scientists in selecting important features, but cannot easily be used as a business matrix for decision making. We wanted to translate the ‘feature importance’ to a more business-ready statistic so that it can be used directly. Therefore, we used Odds Ratio to evaluate the key features.
An Odds Ratio is a statistical matrix commonly used to explore the relationship between diseases and specific exposures (). For example, people who smoke cigarettes are X times more likely to have lung cancer than people who do not smoke. In this domain retention analysis, it will be “domains with X feature are Y times more likely to retain over 1 year than domains who do not have X feature.”
- Odds Ratio > 1 means greater odds of association with the exposure and outcome.
- Odds Ratio = 1 means there is no association between exposure and outcome.
- Odds Ratio < 1 means there is a negative association between the exposure and outcome.
Figure 5 and Figure 6 show each key feature’s Odds Ratio. We can see from the “query groups” in Figure 5 shows domains receiving a high amount of DNS queries are 5.6 times more likely to have retention > 1 year than those that received a low amount of queries.
In figure 6:
- The “registrant_type” shows that domains registered with organisations are 1.9 times more likely to have retention > 1 year than those registered with individuals.
- The “with_paid_email” feature shows that domains with registrants using paid email services are 2 times more likely to stay in the register over 1 year than those that use free email services.
- The “with_email_service” feature shows that domains using domain hosts’ email services are 1.1 times more likely to stay in the register over 1 year than those who do not.
Odds Ratio and Churn Rate relationship
By averaging the different feature’s Odds Ratio, we are able to produce a score for each registrar, making comparison across registrars easier.
Figure 7 shows the churn rate and average Odds Ratio from the domains registered with a single registrar (R10) over the last few years. We can see that as the average Odds Ratio increased, the churn rate decreased.
Registrar (R8) in Figure 8 also shows the same pattern. The reason behind the pattern is because Odds Ratio quantified “How important are these features for domain retention?”, as Figures 5 and 6 showed. For a given domain, its Odds Ratio can be seen as its “quality”. At the registrar level, it can be seen as a portfolio’s quality.
Figure 9 shows several big registrars’ average Odds Ratio over time. Registrar (R14) has the best and most consistent performance. Several registrars had some dips and jumps in their average Odds Ratio due to business acquisitions.
With a closer look into registrar R14, we found that its business type (brand protector) meant it naturally met several conditions important for domain retention. Brand protectors purchase domains for companies in order to protect the companies’ brand names. For example, they might purchase chanel.nz, chanel.io, chanel.ai., etc. for the Chanel (a French luxury fashion brand) company. These types of domains are registered with organisations, using paid email services (see blue highlighted rows in Figure 10). These domains are rarely visited and have long lifespans.
The lesson learnt here is that when using average Odds Ratio to compare registrars, we must keep in mind that the nature of registrars’ businesses is not the same. The nature of the business will influence their average Odds Ratio.
Another example of this is “domain investing” registrars. Their domain names are short. These domains also have a short lifetime because they will be deregistered if they are not profitable (see pink highlighted rows in Figure 10).
This analysis explored several aspects of domain retention. We learnt that:
- A domain’s current age is the most important feature for retention. In general, the older the domain, the more likely it’s going to stay.
- Domains with these features are more likely to stay in the register over 1 year:
- registered using big registrars,
- registered with an organisation,
- have high levels of DNS activity,
- uses domain host’s email services,
- uses paid email services.
- Domain retention patterns:
- Registrations have the highest churn rate in the first year (30% churn rate).
- Registrations reach maturity in the third year (5% churn rate).
- Registrations registered during the COVID-19 lockdown period in April 2020 have the same retention pattern as other cohorts.
- Market growth:
- The register growth has reached maturity stage. Less and less new domains are being registered every year.
- COVID-19 was a surprising boost to new domain registration. However, the boost did not last long. Previous analysis for .nz namespace changes before and during COVID-19 lockdown can be found here.
- Odds Ratio is an effective indicator. When the average OR increased, the retention churn rate decreased.
- Registrars from different business types operate differently. The characteristics of these business types help us understand why some registrars have a higher Odds Ratio than others.
 Song Y, Lu Y., 2015, ‘Decision tree methods: applications for classification and prediction’, Shanghai Arch Psychiatry, v27(2): 130–135. [NCBI]
 Szumilas M., 2010, ‘Explaining Odds Ratios’, J Can Acad Child Adolesc Psychiatry, v.19(3): 227–229. [NCBI]