Counselor
Singapore
“
I read the offer email, I know what happened ,but ManifestGlobal still shows pending. So I'm basically working twice.
01
Gap: signal exists, system doesn't know
420K+
Applications analyzed
61%
University decision coverage
13.4%
Student decision capture
92+
Institutions onboarded
5
Milestone rollouts shipped globally
OUTCOME
Success was defined by data capture, not inbox engagement.
Adoption
- 90+ schools enabled Explore Mail
- 70K+ student inboxes activated
- 3.2x increase in WhatsApp linked email usage
Outcome capture
- 61% university decision coverage achieved
- ~14% student decision capture rate
- 2.7x increase in verified enrollment status signals
Operational efficiency
- 23% reduction in manual counselor follow-ups
- 30% fewer status-chasing workflows
- Faster downstream CRM and enrollment reporting accuracy
01 · AMBIGUITY
The platform had activity, but no outcome truth.
Students
- •Critical emails scattered
- •Missed deadlines
- •Stale records
Counselors
- •No outcome visibility
- •Manual chasing
- •Late interventions
ManifestGlobal
- •Weak conversion data
- •Invoicing risk
- •Limited ML readiness
02 · STRATEGIC REFRAME
This wasn't an inbox problem.
I shifted the team's framing from shipping a mailbox to building a system that converts unstructured admissions communication into structured outcome data.
What changed my mind11 counselors·4 schools·structured interviews
Counselor
Dubai
“
Half my day is confirming things I already know. I'm waiting on a 17-year-old to update a system they've already forgotten about.
02
Gap: relying on students to close the loop is structurally broken
Not this
How do we build a better inbox?
But this
How do we turn every email into a structured data event, without asking anyone to do extra work?
the system that followed
01
Email signal
Inbound admissions email arrives in the student or counselor inbox.
Raw email captured
02
Saige detection
AI classifies the message, extracts intent, and scores confidence.
Intent + confidence
03
Role-based prompt
Student or counselor is nudged with the right next action for their role.
Action nudged
04
Status update
Application status is updated based on the confirmed outcome.
Status updated
05
Structured dataset
Outcome data flows into the analytics layer as structured intelligence.
Data structured
03 · ARCHITECTURE
A system, not a feature.
04 · LEADERSHIP DECISION
The decision that defined the product.
I drove the strategic decision to embed a mail client instead of rebuilding one natively, allowing the team to focus on ManifestGlobal's real differentiator: the intelligence layer, prompts, and proprietary workflow experience.
Option A · Not chosen
Native inbox
- ✓Full UX control
- ✗~60% eng capacity on infrastructure
- ✗Missed pilot window by 3 weeks
- ✗Gmail solved this over a decade
- ✗Slower pilot validation
Option B · Chosen
Embedded + intelligence
- ✓100% eng scope → intelligence layer
- ✓Pilot shipped in 6 weeks
- ✓ManifestGlobal owns prompts + data
- ✓Business value stays native
Known debt, not an oversight
CONSTRAINTS THAT SHAPED THIS
CONSTRAINT 01
Partnership deadline
12-school pilot in 6 weeks, tied to a renewal cycle. Native inbox would have missed the window by 3 weeks. Embedded was the only viable path.
OUTCOME
Pilot shipped on schedule. Renewal secured.
CONSTRAINT 02
Privacy + legal
Student email data required consent states and audit logging before AI classification could run. I redesigned onboarding with a single-screen consent checkpoint that cleared legal review.
OUTCOME
Became reusable pattern across all milestones.
CONSTRAINT 03
AI confidence floor
Saige's accuracy at launch was 71% — below threshold. I designed a human-in-loop override state for counselors to flag low-confidence tags, allowing us to ship while the model trained on live data.
OUTCOME
Accuracy reached 84% by week 6. Model trained on live data.
8 weeks into pilot, counselors weren't asking for a better inbox. They were asking for better outcome visibility.
Post-pilot signal · 12-school cohort: “100% eng scope protected | Consent pattern reused | 84% AI accuracy by wk 6 | Pilot on schedule”
04.5 · ONBOARDING DESIGN
Adoption required trust before functionality.
4-screen sequence. Each screen resolves one objection — in the order students actually feel them.
SCREEN 01
The gift frame
Reframe the highest-friction ask as a gift, not a permission request.
“Why do I need to give you access to my inbox?”
Drop-off: 34% → 11%
Design principle
Each screen resolves one objection — in the order students actually feel them.
Value front-loaded. Friction deferred. Trust earned before functionality asked for.
05 · PRODUCT EVOLUTION
From inbox to intelligence.
The product unfolded in five milestones: from M0 to M5 closes the loop on outcome data.
M0
M1
M2
M3
M4
M5
M0Milestone 1 of 6
Foundation
Identity and domain groundwork, including .ai → .co transition readiness. Established the technical and brand baseline the rest of the product would build on.
Drag the handle or click a step to explore each milestone.
06 · 18-MONTH ROADMAP
From feature to infrastructure.
The roadmap moves Explore Mail from adoption to intelligence to platform-scale admissions data infrastructure.
0
6 MOnow
12 MO
18 MO
0–6 MONTHS
PROVE ADOPTION
6–12 MONTHS
IMPROVE INTELLIGENCE
12–18 MONTHS
PLATFORM LAYER
SURFACES
Student + Counselor inbox
Core communication layer
Activity feed
Timeline view + indicators
Predictive signals
Enrollment-readiness surface
WORKFLOWS
Follow-up + deadline prompts
Zero inbox habit loop
Human-in-loop review
Escalation + consent states
Direct Apply attribution
Source-of-truth pipeline
AI LAYER
Basic role visibility
Student vs counselor routing
Saige taxonomy
Confidence scoring · intent classes
ML-ready dataset
Structured outcome intelligence
DATA + INFRA
Audit logs
Consent states + traceability
API extraction pipeline
Platform-scale data layer
Opacity reflects phase distance — darker = current · lighter = future. Drag the handle or click a step to move between phases.
07 · REFLECTION
The highest-impact design work is not always the interface. It is the system that changes how the business understands reality.
Why these metrics mattered beyond UX
Revenue recognition signal
0.85%→13.4%+1,476%
Student decision capture rate
before
after
ManifestGlobal's invoicing is tied to verified enrolment outcomes. Going from 0.85% → 13.4% didn't just improve dashboard accuracy — it closed a structural gap in revenue recognition, reduced invoice disputes, and gave finance reliable signals for the first time.
−23%
operationalManual counselor follow-ups
≈1.8 hrs saved per counselor/week across 92 institutions. Leverage that compounds as the platform scales.
71%→84%
wk 6Saige AI accuracy
Human-in-loop override allowed us to ship at 71% while the model trained on live data. Threshold reached by week 6.
What I'd do differently
I'd involve the ML team in the UX research phase earlier.
Keep exploring
More case studies
Cialfo x FlywireTuition Payment Tagging App
Open
Tagged Flywire payments in ManifestGlobal to unlock rewards for parents/guardians
FINTECHPAYMENTSRewards/Cashback UXWEB & MOBILE
HunchFTUE Revamp
Open
A retention-focused FTUE redesign that increased D0–D1 retention from 16% to 34%.
FTUE REDESIGNRETENTIONBEHAVIORAL UXMOBILE