User onboarding

Context for this project

This on boarding journey for the Family Tree product is the site’s most successful new user acquisition tool. Users typically took 5-8 minutes to get through the existing journey, which was essentially a series of forms.

Our goal:

We wanted to improve this experience for the user. Using our basic AI we wanted to add delight to this journey, and help users create a more accurate picture of their family’s history, ultimately increasing user satisfaction and subscription rates.

My role

I lead on design in this project, working closely with a Product Manager and 5 Engineers.  Our UX Researcher helped arrange user testing sessions; however I created the testing plans, facilitated the sessions, and analysed the findings.  Protoyping and testing sessions were done on paper, Sketch and Invision, using Principle to animate certain aspects when needed. Handover to developers was done using Zeplin.

Mapping the existing journey, user testing and UX audit

I mapped out the journey, which consisted of 8 screens in total, and planned a user testing session to observe users complete this and identify issues. Each page in the journey consisted of an identical form (no fields were mandatory), and required users to know information about their family. Completing this journey produced a basic family tree. Depending on the information provided by the user, there would be 3-7 members on the tree, some with hints for them to review.  Further exploration of the product at this stage (searching for other ancestors, reviewing any records etc) would require a subscription.

User testing the existing journey

I ran a user interview and testing session (using the existing journey) with 5 new users to Findmypast. The goal was to watch them complete the existing flow and understand what they struggled with.

Positives:

• Users found the overall process easy and quick to complete
• Typeahead feature recently introduced across Findmypast for all UK locations was positively received and sped up the process

Learnings:

• Each step saw a reduced amount of fields completed, due to lack knowledge about older family members
• Most users don’t know their Grandparent’s ‘year of birth’ and ‘maiden names’ (where applicable), this impacted our AI’s ability to offer hints on those ancestors
• Dropdown menu to select birth years. Though no problems were identified, this slowed users down
• Form fields titles later in the flow confused users. eg. ‘Your father’s fathers first name’
• We asked users about their deceased ancestors and when they died. User’s seemed more able to recall their year of death, based on where they were in their own life. As opposed to trying to recall an ancestor’s year of birth.
• Ask better questions that fit with what users know about their family

Suggestion to improve the journey

• The design of the form pages needed to be refactored to better utilise the screen size for mobile
• Removing redundant copy and UI elements
• Create a form that adapts to what users tell us
• Language changes for male/female ancestors
• Ask if ancestors are deceased/alive - better chance to capture relevant information about these people
• Have fields populate depending on the above decision by the user
• Personalising the journey to give context to the user
  eg. If the user’s father’s is ‘Ben Smith’, we’d ask users for ‘Ben’s Father’s name’ opposed to ‘your father’s fathers first name’
• Ask for information users are able to answer. eg. If a relative is alive, ask for their ‘age’ not ‘year of birth’
• Pre-populate the form fields where possible, to reduce the unnecessary user input.  Eg. populate surnames in the paternal lines.

A before and after state are illustrated here:

These changes were implemented over the course of a couple of engineering sprints (legacy code slowed progress). We A/B tested this and our change in field form questions got a 5% increase in Year of Birth data for each person in the user’s family. Though for Grandparents ‘Year of Birth’ data was still low (40% of users could provide the information).

Improving user delight during the journey

We wanted to expose the AIalgorithms of Findmypast (known as ‘Hints’ in the product) to these users. This could help them complete the journeywith more accurate information, and delightthem through proactively showing them information about their family. This required engineering work to bring this feature in front of the paywall.

Our hypothesis: by providing ‘hints’ to users and showing them exciting finds about their family early on; they would build a more accurate family tree, making for a better experience in the product. They would stay longer and be more likely to subscribe.

Workshopping the new journey

Through our user testing and research, we knew that making our product more helpful and smart would improve the user experience.

With the engineers and product manager, we decided the best way to test this hypothesis was early in the journey.  So after a user gave us their details and registers on the first 2 pages, we’d have enough information to show them their birth certificate.
‍

Showing the user their birth record was useful because:

• It validates their data
• The record has their mother’s maiden name, which we could present to the user and populate on the form field page
• We do this early in the journey, without asking for it from the user, they get the benefit of less effort filing in form fields and the business gets the correct data first time round

If they accepted the hint, then we’d pre-populate relevant information on the next form for their mother (and we’d also then have their grandfather’s surname in most cases). Testing this on the site, we would see if it impacted the on boarding completion rate.The below screen flow illustrates how we designed the journey early in the journey.

I held a quick user testing session on this journey, results were very positive. When we had information to show users, they felt the product was helping early in their experienced, “it made the site more credible”.

We A/B tested this on our live site. In the experiment users who accepted a hint from us were two times more likely to subscribe when they completed the onboarding journey.

Using the AI to show the user their father’s information

Building on this success, we implemented this throughout the on boarding journey. With the user’s surname and their mother’s maiden name, we would have our Hints algorithm search for a marriage record between the user’s mother and father. This was chosen as it as still near the front of the journey and could build momentum for the user if our AI got it right.

Presenting this information to users, giving them their father’s name and their parent’s wedding date, without ever asking them who their father was, tested very well. When accepting this hint, the information for the father would populate the next form field, reducing the work the user needed to do.

Catering for the sad journey

In situations where:

• users had common surnames, we could have too many hints.
• Incomplete data from the user (didn’t know their father/had a different surname or maiden to their father), we had no hint.

In the first example, if a user was to decline the first hint, we would show them (at most) two more suggestions.  These were based on the algorithm’s relevancy factor. We’d only show a high relevancy scored hint. After that, we’d progress the user to the next ancestor page. In the case of no hints being available, the user just progressed to the next ancestor page.

Census Hints

Census records have the most information about a person and most times even their immediate family, where they lived and what jobs they did. The UK government releases this data (in almost all cases) 100 years after the census date. For a user, find a grandparent you know in a census and you’ll likely learn a lot about additional family members you probably never knew about.

If we could find a user’s grandparent’s census record from 1911 or 1939 (the 2 most recent available data sets), and show this to them in their on boarding experience, we’d be able to show a user who their great-grandparents were. This would be the ‘wow moment’ may casual users are after, and to get it at the end of this journey was something we wanted to test.

• We would get a tree built on more accurate information, enhancing our ability to help the user further
• The user got a ‘wow moment’ and additional family they most likely didn’t know about to add to their tree

Testing this concept

Testing a paper prototype with users validated the delight and surprise when we would show them their ancestors information.  Running these sessions using actual data from a user’s family can lead to very emotive responses from users.  2 of the 5 users we tested with asked to keep the page showing their ancestors details, as they thought it was an amazing find for them.All five users loved finding more information on family they didn’t know or learnt something new about.  For example an occupation or an address that surprises them or even the names of their great aunts and uncles.

Not overwhelming the user with information

It wasn’t unusual to have large families in 1911. To make these finds more relevant to users, our testing showed that we’d need to reduce long lists of family, to the top ancestors a user would want to see. Users were generally most interested in their grandparents and great grandparents. Bothers and sisters of grandparents were generally of less interest to our casual users.So my design prioritised the user’s grandparent (who typically was a child at the time) and their parents (user’s great grandparents).

Based on the completion data we had from this journey, the fact users don’t know enough about their grandparents details to give our AI enough to work with,  we estimated that about 5% of users would be presented with this census hint.

We decided the engineering effort required to make this a reality in the on boarding journey wasn’t worth it. We’d need a way to get accurate birth information (date and place of birth) for the grandparents, we already knew a way that we could help users do this, that needed validation and appears later in this case study.

Asking better questions so our AI can help the user discover more

Our interviews had indicated that users are better able to recall when their grandparents passed away. Compared with questions about their grandparents birth.  

So when a user used our updated form to say a grandparent was deceased. Our form design added a question about that person’s year of death.

Our AI could use this information to see if records for that name and date existed and we’d present that to the user.This had two purposes:

• We would get a tree built on more accurate inforFrom this data, we’d learn that person’s exact Date of Birth (if the record was after 1969); or, just their Year of Birth (in records before to 1969). Once confirmed by the user, we could use this information to provide the user with more relevant hints in the product.mation, enhancing our ability to help the user further
• It’d continue to demonstrate the product was helping them, and they’d learn more information about their family.

Below is how the information was presented to a user before they progressed onto the rest of the journey.

Improving registration rates

Our data indicated high exit rates (40%) at our registration page, originally the second screen in this journey.

Our hypothesis: Move registration page to the end of this journey, let users access the ‘hints’ we’ve added to the journey. Let them experience value and develop an understanding of what the site could offer them, by the time they reach the registration screen at the end, they’d be invested enough to register.

In addition to letting users experience the product, we wanted users to be able to briefly see the tree they’d built before we ask for registration details. I designed the page registration that would display as a ‘toast’ over the family tree, the user would not be able to interact with the product until the registration was complete.

Use of a progress bar and copy indicating to the user they’re only two final steps away from being able to explore their family tree.

Our Reg page had 4 checkboxes. (and they wonder why there is a big drop off at this page)

• 2 for terms and conditions (1 for Findmypast, the other for a partnership with a company in the US)
• 2 for marketing preferences (1 for Findmypast, the other for partner sites)

Though I checked in with our legal team, we needed to keep these 4 checkboxes.

My solution was to breakdown the account creation process and opt-ins.

• I moved the privacy opt-ins (essential to agree to to progress) to the user form (first page).
• After the user completed the journey, we’d ask for their email and password creation.
• Marketing preferences were now the final screen in this journey.

Transitions added to the card interactions to make the experience look and feel more modern, animation work done in Principle app.

Results of this test

Moving the registration to the end of this journey actually resulted in a reduction of total registrations.  Whilst this was quite worrying, we saw that the experiment cohort actually had a higher total amount of subscriptions purchased than those in the control, they also had higher repeat visits to the site and accepted more with hints.  The business had concerns about the decrease in registrations but with the higher engagement metrics the decisions was made to push the experiment out to all users.