Solving low-quality profiles with AI automation
TLDR
Problem
Incomplete professional profiles get 65% fewer customer interactions, creating a quality spiral that hurts both sides of our marketplace.
Solution
An AI powered profile builder that uses professionals' website content to turn low-quality profiles into high-quality ones seamlessly.
Outcome
Launched - pending metrics. Increase in profile quality. Increase in consumer interaction with professional profiles. Proved we can use AI to enhance profiles.
Team / Role
Product Manager, Engineering team, Insights analyst, Specialist AI agency. Research, testing, product design.
Context
Bark connects millions of consumers with professionals across 1000+ categories globally.
Incomplete professional profiles get 65% fewer customer interactions, creating a quality spiral that hurts both sides of the marketplace.
Complete profile
Incomplete profile
Customer problem
Laura, a yoga instructor, spends hours copying information from her website into our platform. She abandons when asked to manually describe her services - information that already exists on her website.
+ Prioritised
Building a profile takes a lot of time and effort
The existing profile builder had significant usability challenges
- Deprioritised
Users didn't understand the importance of a good profile
Solution
Build an AI powered profile builder to seamlessly turn low quality profiles into high quality ones.
We knew that 70% of professionals had a websites on Bark which contained the majority of the information needed for a perfect profile. We were challenged to build a functioning MVP in 2 months which could be launched to existing users with low profile quality.
Design considerations
User attributes
Existing user, Active, Has a website, Low to medium profile completeness (means we need to design for new + existing profile info)
User goal
Users should be able to seamlessly view/accept/edit suggested info and add it to their profile
User pain points
Not sure what information is currently on their profile, Don't change my profile without my permission, They don't trust AI generated suggestions, The generated content isn't good enough, Cognitive overload - there is a lot of information on profiles
User success criteria
Time taken to use the feature (mindful of loading times), Completion rate of reviewing AI suggestions, Acceptance rate of suggestions
Designing for AI principles
Source transparency
Users need to understand what the AI is doing and what specific data sources it used to create personalised suggestions, making them feel higher quality.
Co-creation
Users need to feel they are working with the AI. They can override, edit, or reject AI outputs - they remain the decision-maker, AI is the assistant.
Graceful Failure
When AI gets it wrong, make it easy to correct the mistakes.
Key design decisions
New Flow or add to existing Profile Builder
Choice
Retrofit existing builder or start fresh?
Considerations
Existing UX debt would negatively impact completion rates and time taken
Decision
Built dedicated AI flow separate from legacy profile builder
Impact
Faster to build, better UX, avoided known usability issues
Existing profile builder
Section-by-Section vs Big Bang Validation
Choice
Review all AI suggestions at once or break into smaller sections?
Considerations
Existing research showed us the more we could make the feature look like a customer facing profile the easier it would be for customer to understand
Decision
Section-by-section validation for progressive disclosure
Impact
Reduced cognitive load, higher completion rates
Concept 1
Concept 2: Step 1
Concept 2: Step 2
Background Website Extraction
The challenge
Eliminating the load time associated with web scrapping and AI generation. Load times could be up to 5minutes.
Decision
AI scrapes websites in background, presents curated suggestions. Users see an in product message prompting them to view their AI suggestions.
Impact
This eliminated load time as a pain point and removed API rate limit constraints
High level user flow
Eligible professional is on Bark
See in product modal to view their AI suggestions
Open AI suggestion flow
Accept/skip suggestions
Informed that changes have been made, ability to view their public profile
Step 2 deepdive:
Messaging rationale
Time efficiency ("under 2 minutes")
Directly addresses the known pain point of time-consuming profile creation
Clear value proposition ("win more work")
Focuses on the end benefit professionals care about most
Gentle approach ("suggested improvements")
Non-threatening language - users maintain control
Result
9.6% click rate
(60% above Chameleon's strong engagement benchmark)
Solution design
Rationale
Separate flow to avoid UX debt and increase completion rate of the flow
Loading suggestions in the background to avoid the customer losing interest due to long wait times
Section by section flow to increase completion rate and reduce cognitive load
Editing on hover, reduces noise on page, providing immediate visual feedback
Customer testing
From 6 moderated interviews we grew confidence in our design decisions.
The solution was successfully validated across 4 areas
Source attribution
Users clearly understood content came from their website
Information hierarchy
Distinguished existing vs suggested content
Interaction model
Knew how to edit and add suggestions
User control
Understood they could skip suggestions entirely
Changes from customer testing
Initial modal messaging about 'incomplete profiles' created negative priming. Reframing to focus on 'suggestions' and 'improvements' significantly improved user receptivity.
"This would have saved me so much time copying and pasting from my website"
"I can't believe how fast it was."
Figma prototype used for testing
Outcomes
Just launched, awaiting full results. Preliminary results include:
+ Achieved 80% conversion rate - significantly outperforming typical profile creation benchmarks of 20-40%
+ A professional can take their profile from blank to medium quality in about 1 minute, saving them time and improving the marketplace experience for buyers.
+ De-risked AI investment
+ Scalable pattern for AI profile creator to build on