Onboarding & Template Recommendations
Expertise: UX/UI Role: Product Designer Company: Venngage When: Nov 2017
TL;DR
Background:
Venngage is a simple design platform for busy professionals. The tool offers thousands of pre-built templates that can be customized in a user-friendly drag & drop editor. As a B2C and B2B SaaS product, Venngage empowers individuals and teams to create professional designs, manage all their content, and maintain brand consistency with ease, even if they lack design experience.
As the number of templates increased, users encountered difficulties in finding relevant templates for their design needs. This was especially problematic for new users discovering the tool for the first time.
Goal:
Increase creation rate for templates page
Deliverables:
Release a new onboarding flow incorporating recommendations
Release an MVP recommendation algorithm leveraging existing search engine, user data, and templates.
Impact:
Creations rates increased by 27% — from 66% to 82% during the time of the initial release
More users were finishing onboarding (up 20%) — Once the desktop version was released, the UX was released to all acquisition and mobile flows.
This experience paved it’s way for future iterations such as incorporating a user’s brand, recommendation algorithm improvements, and updated curations.
Back in 2017, I was one of two designers brought on to support Venngage's brand new onboarding flow that incorporated template recommendations based on people's goals.
Discovery & Research
Defining the user problems
Our library housed thousands of easy to use pre-built templates and because of this, first time users often experienced content overload.
We spoke to users, analyzed existing user flows, sorted through support documentation and user chats, and compiled the information into a user journey. This gave the team a holistic view of the customer experience by uncovering moments of both frustration and delight.
While we envisioned an onboarding experience that felt tailored, it became clear we needed to address these issues first:
Users are asked generic screener questions with no personalization. After onboarding, templates shown are the same for every user.
Most users don’t want to create from scratch, yet, that's the first thing they see after onboarding. The metrics showed that users were more likely to complete having started from a template vs from blank.
Busy interface (overuse of icons, decorators with no function) & unnecessary copy with no real value given to users
The hypothesis & goal measures
Defining the goals, metrics, and baselines that will be used to measure the success of the hypothesis
This was the main hypothesis we wanted to test:
We believe that recommending templates to users based on their goal and design preferences will increase template creation rate by 20% (from ~66% to 80%) because users will have curated options without having to spend time browsing.
The goal measures:
Registration to Creation %
Registration to Finish Onboarding %
Registration to Upgrade (All plan types)
Understanding the limitations and dependencies
Building a “good enough” recommendation algorithm to validate our ideas
We needed to figure out how to build a recommendation algorithm without actually building it from scratch.
We leveraged Algolia (our search engine platform) and used existing facets.
Manually curated templates based on templates users were already creating. I knew this iteration would not scale well but it was fast to build and the idea needed to be vetted.
Grouped and tagged templates with style attributes. We needed to build similarity graph for subcategories, styles, content type to populate recommendations.
We had basic template data like category, subcategory and publication date but we needed stylistic tags in order to recommend templates based on aesthetics. I did a design attribute mapping activity grouping similar designs in categories. This was an experiment to figure out whether style was an important factor when someone chooses a template. In the end, style wasn’t as important as content and category but it was a learning experience.
Once we broke down the template data, we formulated a game plan to test the recommendation quality.
Setting boundaries and time-boxing tasks
We were aware that there were highly advanced recommendation systems out there — some of our competitive research took a look at Pinterest's recommendation system. We also had to consider what a new onboarding meant for all our acquisition channels. Ultimately, I decided to scope the project down keeping in mind our resources. Without a time limit, work expands and I wanted to validate our idea, see whether it impacted our creation numbers, and then figure out whether we had an appetite to scale it.
I scoped the project down and A/B tested users coming in from the homepage, blog, and static pages — basically any pages that tested the full onboarding flow. We didn’t want to increase complexity by introducing alternative flows. This gave us enough time to iterate on the UI, fix any UX errors and test out all major and minor flows in time for release.
Balancing revenue goals
We knew we had to maintain upgrades from the templates page since it was part of the top 5 conversion flows in the product. I saw this as an opportunity to improve conversion rates by experimenting with the distribution of paid templates vs free, this is a rule we added to the algorithm later on.
Defining the Vision
Working backwards from the vision
We wanted to approach recommendations as a feature that would support multiple phases as the product scales. This led to the idea of defining the feature vision early on. This practice helped us put risky ideas down on paper — pushing the team to build toward it.
We wanted the team to work toward a recommendation vision that would:
Automatically suggest relevant templates based on the user’s role and job to be done
Be able to generate coherent sets of designs using their brand logo and colors
Design Phase
Sketches & UI Explorations
Building interactive prototypes for quick feedback
In order to keep development costs low, I spent more time building an interactive prototype for user testing. When flows didn’t make sense or new ideas were presented, I was able to go back to the drawing board and iterate.
We ran 2 methods of user testing — live interviews with existing users and task based testing on the Usertesting platform. Here were some learnings after our feedback sessions:
People did not scroll much. We needed to ensure onboarding categories were visible above the fold and pare it down to the most essential categories.
People wanted to select more than 1 template
The recommendations had no logical order and people couldn’t find their initial selections
Implementation
Releasing the feature in reasonable chunks
1st release included:
Full onboarding experience for users coming from homepage, blog, and static pages
Exposure to 30% of new users and monitored for 2 weeks
2nd release included:
Tweaking the recommendation algorithm by adding category, subcategory and style tags so the recommendations rely less on manual curations
Lightweight onboarding for template specific sign ups
Mobile version for all flows
Changing the paid template threshold so we could funnel more users through a paid flow
Exposure to 50% of new users and monitored for 2 weeks
Final release:
Changing the paid template distribution back to 50% free and 50% paid. Although there was a higher volume of users at the top of the funnel, the conversions were the same whether we had more or less paid templates.
Released to 100% of users
Learnings after release
Impact & Reflections
Creations rates increased by 27% — from 66% to 82% during the time of the initial release
More users were finishing onboarding (up 20%) — Once the desktop version was released, the UX was released to all acquisition and mobile flows.
A version of this recommendation onboarding flow is still being used at Venngage with curations updated as the template quality improved.
While the creations increased for first time users, recurring creations from this category dropped significantly. As time progressed, we found tags like style to be less important and sub-categories and recently added templates to be more relevant. I’d like to go back and experiment with algorithm rules (like adding collaborative filtering) and refresh recommendations periodically so users find new templates waiting for them on a regular basis.
Future Opportunities
We know that templates are easy to use and an essential starting point to speed up the design workflow. In order to differentiate the recommendation experience from other design tools, there’s opportunity to double down on business and enterprise use cases by generating pre-built templates that not only incorporates their brand but industry-specific topics as well. Instead of plugging in content and customizing a design, people will have the added convenience of making minor tweaks and sharing their designs right away.
