Multiplier • Separation Agent • 2025
Making separation cases compliant and self serve for HR admins through AI.
Multiplier • Separation Agent • 2025
Making separation cases compliant and self serve for HR admins through AI.
my role
End to end product design, information architecture, AI conversation design
team
Product manager
Lead engineering
HR and Legal advisor
Product designer (me)
timeline
12 weeks
Q4’ 2025
my role
End to end product design, information architecture, AI conversation design
team
Product manager
Lead engineering
HR and Legal advisor
Product designer (me)
timeline
12 weeks
Q4’ 2025
core problem
"We thought hiring a global workforce was our biggest challenge—until we had to start separating them."
core problem
"We thought hiring a global workforce was our biggest challenge—until we had to start separating them."
separations previously
Complex, multi-regional employee terminations were handled through scattered Zendesk tickets and heavy manual Ops dependencies, creating bottlenecks and severe legal compliance risks.
separations previously
Complex, multi-regional employee terminations were handled through scattered Zendesk tickets and heavy manual Ops dependencies, creating bottlenecks and severe legal compliance risks.
Multiplier Ops team
Often initiating a multi-day, back-and-forth email thread hindered by timezone delays
HR clients
Requests a termination through a Zendesk ticket
Multiplier Ops team
Often initiating a multi-day, back-and-forth email thread hindered by timezone delays
HR clients
Requests a termination through a Zendesk ticket
solution
Designed a 0-to-1 AI Separation Agent that automates contract data extraction and enforces regional labor laws and guides HR Admins through complex, compliant workflows with a self-serve, step-by-step experience.
solution
Designed a 0-to-1 AI Separation Agent that automates contract data extraction and enforces regional labor laws and guides HR Admins through complex, compliant workflows with a self-serve, step-by-step experience.
addressing the unmet needs
addressing the unmet needs
Lack of centralisation
scattered tools → hard to keep trail
“A dedicated system that embeds the entire process would be easier... our current system is an Excel sheet”
Operational In-efficiency
manual process → delay
“This delay is costing me money I will have to pay the contractor. We are +5 days from this email. Please advise."
Lack of centralisation
scattered tools → hard to keep trail
“A dedicated system that embeds the entire process would be easier... our current system is an Excel sheet”
Operational In-efficiency
manual process → delay
“This delay is costing me money I will have to pay the contractor. We are +5 days from this email. Please advise."
Knowledge gap
lacks legal expertise → dependent and error prone
“The easy and the fast route is termination by mutual agreement. Anything else will cost a complex termination fee.”
Trust deficit in AI
wants speed → but with control
“If I’m going to act with something non human, I need to have the confidence of control.”
Knowledge gap
lacks legal expertise → dependent and error prone
“The easy and the fast route is termination by mutual agreement. Anything else will cost a complex termination fee.”
Trust deficit in AI
wants speed → but with control
“If I’m going to act with something non human, I need to have the confidence of control.”
contextual adaptation & system visibility
Dynamic Procedure list
A dynamic procedure list that actively updates itself as the context of the termination shifts.
contextual adaptation & system visibility
Dynamic Procedure list
A dynamic procedure list that actively updates itself as the context of the termination shifts.
what I did
Paired a vertical compliance checklist with a conversational stage.
the trade-off
We traded the simplicity of a standard wizard for adaptive flexibility.
what I did
Paired a vertical compliance checklist with a conversational stage.
the trade-off
We traded the simplicity of a standard wizard for adaptive flexibility.
The Admin mentions the employee has absconded. The side panel dynamically re routes the checklist from PIP steps to AWOL steps in real-time.
The Admin mentions the employee has absconded. The side panel dynamically re routes the checklist from PIP steps to AWOL steps in real-time.
human in the loop
AI Parsing & Human validation
Extracts key dates and clauses from messy PDF contracts → intentionally pauses the workflow to force the HR admin to explicitly verify the data before any legal action is generated.
human in the loop
AI Parsing & Human validation
Extracts key dates and clauses from messy PDF contracts → intentionally pauses the workflow to force the HR admin to explicitly verify the data before any legal action is generated.
the ideal
AI instantly ingests any contract, maps local laws, and executes the termination.
the reality
Contracts are often messy, scanned PDFs. Blindly relying on AI extraction posed a massive legal risk.
the ideal
AI instantly ingests any contract, maps local laws, and executes the termination.
the reality
Contracts are often messy, scanned PDFs. Blindly relying on AI extraction posed a massive legal risk.
the adaptation
I designed mandatory "Data Validation" screens. The AI pre-fills data (Start Dates, Probation Periods), but the HR Admin must review and edit before any legal notices are generated.
the adaptation
I designed mandatory "Data Validation" screens. The AI pre-fills data, but the HR Admin must review and edit before any legal notices are generated.
Showcases the user dropping a PDF contract → AI parsing state → "Quick Details" view. The zoom-in on the Edit Button emphasizes the human in the loop requirement.
Showcases the user dropping a PDF contract → AI parsing state → "Quick Details" view. The zoom-in on the Edit Button emphasizes the human in the loop requirement.
transparent trade-offs
Strategic Risk Mitigation
the interface proactively compares termination options → explicitly highlighting the "Legal Risk" to naturally steer admins toward safer, less litigious routes like Mutual Separation Agreements.
transparent trade-offs
Strategic Risk Mitigation
the interface proactively compares termination options → explicitly highlighting the "Legal Risk" to naturally steer admins toward safer, less litigious routes like Mutual Separation Agreements.
what I did
The AI is architected to pitch a Mutual Separation Agreement (MSA) as the primary path, even if the user asks for formal termination.
the trade-off
In many jurisdictions, formal terminations are high-risk. MSAs mitigate litigation risk. We traded immediate severance cost for Long-term Legal Protection.
what I did
The AI is architected to pitch a Mutual Separation Agreement (MSA) as the primary path, even if the user asks for formal termination.
the trade-off
In many jurisdictions, formal terminations are high-risk. MSAs mitigate litigation risk. We traded immediate severance cost for Long-term Legal Protection.
The Agent captures the intent, identifies legal risk, and the canvas zooms in on the "Mutual Separation" recommendation card.
The Agent captures the intent, identifies legal risk, and the canvas zooms in on the "Mutual Separation" recommendation card.
preventive guardrails
Compliance Hard Stops
the system actively intervenes and blocks the user from bypassing mandatory due process steps → preventing actions that would result in an illegal dismissal.
preventive guardrails
Compliance Hard Stops
the system actively intervenes and blocks the user from bypassing mandatory due process steps → preventing actions that would result in an illegal dismissal.
the ideal
A touchless, fully automated separation process.
the reality
High-risk markets require strict procedural due process. AI cannot assess emotional volatility or nuanced retaliation risks.
the ideal
A touchless, fully automated separation process.
the reality
High-risk markets require strict procedural due process. AI cannot assess emotional volatility or nuanced retaliation risks.
The HR Admin types "I’ve had enough, just terminate." The Chatbot triggers a "Hard Stop" warning, preventing a legally risky rush.
The HR Admin types "I’ve had enough, just terminate." The Chatbot triggers a "Hard Stop" warning, preventing a legally risky rush.
intent translation
Automated Drafting & Execution
the AI drafts formal legal notices → presenting them in an integrated, editable email preview for a final human sign-off before dispatching to the employee.
intent translation
Automated Drafting & Execution
the AI drafts formal legal notices → presenting them in an integrated, editable email preview for a final human sign-off before dispatching to the employee.
The AI drafts the formal notice based on the case history. The user reviews, edits, and sends it directly through the chat interface.
The AI drafts the formal notice based on the case history. The user reviews, edits, and sends it directly through the chat interface.
impact
By transitioning from scattered Zendesk tickets and manual email threads to a unified AI Copilot, we transformed the HR Ops workflow:
impact
By transitioning from scattered Zendesk tickets and manual email threads to a unified AI Copilot, we transformed the HR Ops workflow:
75%
Separation Zendesk Tickets
MoM comparison for Q1 for Philippines and Indonesia
100%
Procedural Compliance
zero procedure related tickets
75%
Separation Zendesk Tickets
MoM comparison for Q1 for Philippines and Indonesia
100%
Procedural Compliance
zero procedure related tickets
~20min
Separation case flow average time
down from 4-6 hrs
Risk Mitigation
AI pushes for Mutual Separation
for cases with high legal risks
~20min
Separation case flow average time
down from 4-6 hrs
Risk Mitigation
AI pushes for Mutual Separation
for cases with high legal risks
