My Role
Led product strategy and design for a scalable scam intervention framework on Instagram, defining how and when to warn people about potential scam interactions. Partnered with Product, Engineering, Data Science, Legal, and Content Design.
What shipped
A multi-level intervention framework based on risk precision, alongside a system of contextual warnings across key interaction points (messaging, following, commenting). Enabled teams to apply the right level of friction based on confidence in scam signals.
The impact
Reduced harmful interactions between victims and scammers while maintaining user trust. Launched warnings across key surfaces that led to a 28% reduction in estimated scam-related reports, with users reporting increased confidence in Instagram’s safety protections.
Problems
Most scam interactions begin with user-initiated engagement, not inbound messages
91% of first interaction with scammers come from content posted by suspected scam accounts
47% of scam conversations are initiated by victims after viewing content
Applying high friction universally erodes trust and harms legitimate users
Users who experience scams reduce engagement or leave the platform entirely
Goals
Create a framework to map intervention type to risk precision
Increase heed rate of scam warnings
Reduce victim-initiated engagement with suspected scam accounts
Protect users while maintaining trust in the platform
Scope and constraints
It is extremely difficult to determine with absolute certainty that an account is a scam account and is violating community standards. Interventions aim to give potential victims enough information to determine if an account is a scammer, without over-enforcing on benign accounts
We worked closely with Legal to determine what language we could and could not use when revealing information about why we suspected a certain account may be a scammer
Solutions needed to work across multiple entry points (content, profiles, messaging, comments)
Process
Developing a friction framework
Before designing individual warnings to target specific interactions, we needed to design a scalable system for how we would intervene and assign precision levels to each type of intervention.
Audited existing intervention patterns and mapped them by friction level
Partnered with Product and Data Science to align intervention types with risk confidence, balancing people problems with business impact
Established a system of Block, Reduce, and Warn to guide enforcement and design decisions:
Warn—used for the lowest confidence and focused on victim-facing interventions. Within this level, we broke down warnings into low versus high friction on a smaller precision scale
Reduce—used for medium precision and included downranking content posted by suspected scam accounts
Block—used for our highest confidence level and included blocking features for scammers
Designing a scalable warning system
Using our new intervention framework, I designed a system of warnings for use within the Warn level of precision. After we determined the most common pathways that lead to financial scams, I explored how we might warn victims about to engage in specific interactions, creating a low friction, medium friction, and high friction variant for each interaction.
Design principles:
Match friction to confidence—apply stronger interventions as certainty increases
Intervene at the moment of risk—focus on high-intent actions, not passive browsing
Preserve trust while preventing harm—avoid over-enforcement on legitimate users
Solutions
Messaging warning
Warned users attempting to message suspected scam accounts after viewing their content, targeting one of the most common scam entry points.
Follow warning
Expanded follow warnings to trigger from content, not just profiles. Introduced additional context (account location, username history) based on research with scam victims, who reported using these signals to assess trustworthiness.
Comment warning
Warned users engaging with scam content via comments, a common pathway for scammers to identify and target victims.
Impact
Reduced scam-related harm, with launched warnings driving a 28% reduction in estimated violating reports (a proxy metric for measuring scams)
Established a scalable framework for scam interventions
Enabled consistent, data-informed approaches to applying friction
Improved user trust by signaling proactive safety protections
Drove adoption through reusable patterns across surfaces
Extended to Threads, informing the first message warning system for the platform