Back to portfolio
AI Lead Scoring · Multi-Vertical · GDPR
LeadWise — AI Lead Qualification Engine
AI-powered lead qualification engine · Two verticals: Banking & Travel
Complete BA Specification Case Study · Business Analyst
FastAPI / Python
Claude API (Anthropic)
Salesforce CRM
GDPR Art. 6/17/21
Pydantic v2
Streamlit
Render
| Author | Zahra · Business Analyst |
| Verticals | 🏦 Banking · ✈️ Travel |
| Code | GitHub — zahraeh/LeadWise |
| Version | v1.0 · 2026 |
| Status | Portfolio · Sample/fictional data |
6BDD User Stories
10Business Rules
8REST API Endpoints
<45slead processing time
01
Business Context & Problem Statement
BankWise Retail is a fictional French retail bank positioned on the digital-first segment. As part of its seasonal campaign Livret A Boosté — Spring 2025, the bank generates several thousand marketing leads per week through its digital channels (Google Ads, email, comparison sites, push notifications).
This entire flow is currently processed manually by branch advisors, with no prioritization tool or structured GDPR compliance system.
Diagnosis of Pain Points
| Identified Pain Point | Operational Impact |
|---|---|
| 100% manual processing of inbound leads | Total dependency on human effort, frequent data-entry errors |
| Callback delay exceeding 72h | Loss of "Hot" leads to direct competitors |
| Conversion rate below 8% | Estimated lost revenue of +€30K/month |
| No scoring or prioritization | Overloaded advisors, poor resource allocation |
| GDPR compliance not under control | Risk of CNIL sanctions and regulatory disputes |
| No traceability of consents | Maximum exposure in the event of an external audit |
2025 Industry Benchmark
Average contact delay (leading banks): 4h
Conversion rate: 12–20%
Automated scoring: standard practice at 78% of digital banks
Average contact delay (leading banks): 4h
Conversion rate: 12–20%
Automated scoring: standard practice at 78% of digital banks
Project Objectives
| # | Objective | Target |
|---|---|---|
| OBJ-1 | Reduce processing time | < 24h |
| OBJ-2 | Improve conversion rate | 15% |
| OBJ-3 | Automate scoring | 100% |
| OBJ-4 | Ensure GDPR compliance | 0 CNIL non-conformities |
| OBJ-5 | Centralize tracking in Salesforce | 100% tracked |
| OBJ-6 | Contact 80% of leads within 48h | 80% |
02
Project Framing — Scope & Governance
Functional Scope
✅ In Scope
- Lead capture and qualification
- Consent management & GDPR traceability
- Automatic AI scoring (Claude API)
- Automatic assignment to advisors
- Full lead lifecycle tracking
- Real-time KPI dashboard
- Automatic deduplication
❌ Out of Scope
- Existing customer management
- Online subscription module
- Financial reporting / accounting
- Banking back-office integration
- Native mobile application
- Complaints management
- Customer loyalty module
Regulatory & Legal Constraints
| Reference | Requirement | Criticality |
|---|---|---|
| GDPR Art. 6 | Explicit consent required on every submission | CRITICAL |
| GDPR Art. 17 | DELETE /leads/{id} endpoint operational within 72h | CRITICAL |
| GDPR Art. 21 | Opt-out → immediate stop + data anonymization | CRITICAL |
| CNIL · Retention | Automatic deletion after 3 years if not converted | MAJOR |
| AML-CFT | KYC verification before any commercial offer | MAJOR |
| PSD2 | Encryption of banking data in transit and at rest | MAJOR |
RACI Matrix
| Stakeholder | Role | Framing | Specs | Tests | UAT |
|---|---|---|---|---|---|
| Marie Dupont | Product Owner | A | A | I | A |
| Jean-Marc Lefebvre | Marketing Lead | C | C | I | R |
| Sophie Arnaud | GDPR Counsel | C | R | C | A |
| Romain Petit | Tech Lead | I | C | R | I |
| Zahra ★ | Business Analyst | R | R | R | R |
R = Responsible · A = Accountable · C = Consulted · I = Informed
03
Functional Specifications
Business Rules
| ID | Description | Priority |
|---|---|---|
| RG-001 | Consent required — no lead is processed without explicit opt-in | CRITICAL |
| RG-002 | Score < 3 → Cold segment → automatic nurturing sequence | CRITICAL |
| RG-003 | Score ≥ 7 → advisor contacts within 24h (SLA) | CRITICAL |
| RG-004 | Prospect data deleted after 3 years if not converted | CRITICAL |
| RG-005 | Opt-out → immediate stop + anonymization | CRITICAL |
| RG-006 | Lead assigned to the advisor at the nearest branch (postal code) | MAJOR |
| RG-007 | Duplicate (email OR phone) → silent merge + log | MAJOR |
| RG-008 | Consent date/time archived in the database (audit trail) | CRITICAL |
| RG-009 | Score 4–6 → advisor assignment within 48h | MAJOR |
| RG-010 | Every AI call logged with input/output for audit | MAJOR |
User Stories — BDD Format
US-001 · Lead capture via web form
Must Have · 5 pts
"As a prospect interested in the Livret A Boosté offer, I want to fill out an online form so that I can be contacted by a BankWise advisor."
- Required fields: Last Name, First Name, Email, Phone, Postal Code, Source
- Explicit, mandatory GDPR checkbox — submission blocked if unchecked
- Automatic confirmation email within 5 minutes
- Lead created in Salesforce with status "Prospect"
- Consent timestamp archived (RG-008)
- Duplicate detection on email AND phone (RG-007)
US-002 · Automatic scoring by the AI agent
Must Have · 8 pts
"As a marketing manager, I want every new lead to be automatically scored by AI so that I can prioritize outreach."
- Score computed on a 1–10 scale in under 3 seconds
- Weighted criteria: source (25%), web behavior (20%), age (20%), postal code (20%), time of day (15%)
- Classification: Cold (1–3) / Warm (4–6) / Hot (7–10)
- Score and AI justification visible in Salesforce within 2 minutes
- Scoring log retained (input, output, timestamp) for auditability
US-003 · Automatic assignment to an advisor
Must Have · 8 pts
"As a branch manager, I want qualified leads to be automatically assigned to an available advisor."
- Assignment based on geolocation (postal code) AND workload
- Advisor notification: email + Salesforce alert upon assignment
- Priority queue if no advisor is available within the SLA
- Hot SLA: 24h | Warm SLA: 48h
US-004 · Right to object management (GDPR opt-out)
Must Have · 5 pts
"As a prospect, I want to be able to withdraw my consent at any time via an unsubscribe link."
- Unsubscribe link present in every marketing email
- Opt-out request processed and confirmed within 72h (GDPR Art. 21)
- Immediate stop of all marketing communications upon opt-out
- Data anonymized (retained only if legally required under AML-CFT)
US-005 · Real-time KPI dashboard
Should Have · 5 pts
"As a marketing manager, I want to view my campaign performance indicators in real time."
- Lead volume, conversion rate, opt-in rate, average contact delay
- Data refreshed every 60 minutes — filters: campaign, date, channel, segment
- CSV export available for external analysis
US-006 · Automatic lead deduplication
Should Have · 3 pts
"As an advisor, I want to avoid receiving the same prospect twice in my pipeline."
- Detection on email AND phone for every new submission
- Silent merge: enrichment of the existing lead record
- Notification to the original advisor with a diff of the information
04
Technical Architecture & AI Agent
Technology Stack
🖥️
Streamlit
Frontend — form & KPI dashboard⚡
FastAPI
Backend — REST API, flow orchestration🤖
Claude API
claude-sonnet — AI scoring & justification✅
Pydantic v2
Schema validation & business rules☁️
Salesforce
Simulated CRM (JSON) — leads & assignments📊
Locust
Perf. testing — 50/200/500 concurrent usersData Flow — End-to-End Architecture
1
Streamlit — Form
Prospect submits contact details + GDPR consent
2
FastAPI — POST /leads <0.5s
Pydantic validation, consent timestamp, deduplication
3
Claude API — AI Agent <3s
Profile analysis, 1–10 score calculation, text justification
4
FastAPI — Assignment Engine <0.5s
Advisor selection (geo + workload), SLA creation
5
Salesforce Mock — CRM <0.5s
Lead record created with score, segment, assigned advisor
6
Email + Notification <5 min
Prospect confirmation + advisor alert (email + CRM)
AI Scoring Matrix
| Criterion | Weight | Max Score | Min Score |
|---|---|---|---|
| Lead source | 25% | Direct email form | Unsolicited push notification |
| Web behavior | 20% | 3+ visits > 5 min | 0 product page visits |
| Age bracket | 20% | 25–45 years | <18 or >65 years |
| Geographic area | 20% | Paris, Lyon, Bordeaux | Isolated rural area |
| Submission time | 15% | Mon–Fri 9am–6pm | Night or weekend |
Segment Classification
Cold (1–3)No assignment — Email nurturing sequence: D+3, D+7, D+14
Warm (4–6)SLA 48h — Advisor assignment + standard notification
Hot (7–10) ★SLA 24h — Immediate assignment + urgent alert + SLA follow-up
API Endpoints — FastAPI
| Method | Endpoint | GDPR Constraint |
|---|---|---|
| POST | /leads | Consent verified |
| GET | /leads | Token required |
| GET | /leads/{id} | Token required |
| POST | /leads/{id}/score | Token required |
| PUT | /leads/{id}/statut | Token required |
| DELETE | /leads/{id} | Admin + mandatory logging |
| GET | /dashboard/kpis | Token required |
| POST | /leads/{id}/optout | No token required |
05
Test Strategy & Validation Results
Overview
| Level | Tool | Coverage | Status |
|---|---|---|---|
| Unit tests | pytest | 90% of code | DONE |
| Integration tests | pytest + httpx | All endpoints | DONE |
| Functional tests | Squash TM | 6 US / 10 BR | IN PROGRESS |
| GDPR tests | Squash TM | 6 GDPR articles | IN PROGRESS |
| UAT tests | Excel / Notion | 5 scenarios | TO SCHEDULE |
| Performance tests | Locust | 3 load scenarios | TO SCHEDULE |
Functional Test Cases
| ID | Title | Status |
|---|---|---|
| CT-001 | Valid form submission | PASSED |
| CT-002 | Blocked without GDPR consent | PASSED |
| CT-003 | Invalid email blocked (422) | PASSED |
| CT-004 | Hot lead scoring (score ≥ 7) | PASSED |
| CT-005 | Cold lead scoring (score ≤ 3) | PASSED |
| CT-006 | Automatic advisor assignment | TO RUN |
| CT-007 | GDPR opt-out via email link | TO RUN |
| CT-008 | Duplicate detection (same email) | TO RUN |
| CT-009 | Dashboard CSV export | TO RUN |
06
KPI Tracking & Delivery Summary
Target KPI Dashboard
| Indicator | Baseline | Target | Current Result |
|---|---|---|---|
| Lead → opportunity conversion rate | <8% | 15% | To be measured |
| Lead processing time | >72h | <24h | ~45s ✔ |
| Leads contacted within 48h | — | 80% | To be measured |
| Opt-in consent rate | — | 60% | To be measured |
| AI scoring response time | — | <3s | ~1.8s ✔ |
| System availability | — | >99% | To be measured in prod |
BA Skills Demonstrated
Requirements gathering & analysisAdvanced
User Story & spec writingAdvanced
GDPR regulatory complianceAdvanced
Test strategyAdvanced
Applied AI for business use cases (Claude API)Intermediate
KPI-driven managementAdvanced
6BDD User Stories
18%target conversion rate
5destinations covered
<24hadvisor callback time
01
Business Context & Problem Statement
LinguaWise International is a fictional language-travel agency, inspired by the EF Education First model. Its flagship campaign, Intensive English Course — Summer 2025, generates several thousand prospect inquiries per week via Google Ads, travel comparison sites, web forms and social media.
These leads are currently processed manually by travel advisors, with no prioritization or structured pipeline. The conversion rate stalls below 10%, well behind the sector's leading players.
Diagnosis of Pain Points
| Identified Pain Point | Operational Impact |
|---|---|
| 100% manual processing of inbound inquiries | Assignment errors, frequent oversights, uneven workload |
| Callback delay exceeding 72h | Loss of hot leads to competitors (EF, Kaplan) |
| Conversion rate below 10% | Estimated lost revenue of +€25K/month |
| No scoring or prioritization by destination/budget | Generalist advisors instead of destination specialists |
| GDPR compliance not structured | CNIL risk on sensitive data (minors, passports) |
| No prospect lifecycle tracking | Follow-ups based on gut feeling, no nurturing sequence |
Destinations Covered — Summer 2025
🇬🇧 London
🇮🇪 Dublin
🇺🇸 New York
🇲🇹 Malta
🇨🇦 Toronto
Target: 18–35 year-olds + professionals looking to improve their English
Sector benchmark: conversion rate 15–25%
Sector benchmark: conversion rate 15–25%
Project Objectives
| # | Objective | Target |
|---|---|---|
| OBJ-1 | Reduce callback delay | < 24h |
| OBJ-2 | Improve conversion rate | 18% |
| OBJ-3 | Automate AI scoring | 100% |
| OBJ-4 | Ensure GDPR compliance | 0 CNIL non-conformities |
| OBJ-5 | GDPR opt-in rate | 65% |
| OBJ-6 | Contact 85% of leads within 48h | 85% |
02
Project Framing — Scope & Governance
Functional Scope
✅ In Scope
- Prospect capture and qualification
- Consent & GDPR traceability
- AI scoring by destination and profile
- Assignment to the specialist advisor
- Full lead lifecycle tracking
- Real-time KPI dashboard
- Automatic deduplication
❌ Out of Scope
- Online booking and payment
- Visa and passport management
- Accommodation / housing module
- Native mobile application
- Existing customer management
- Loyalty module
- Accounting back-office
Regulatory & Legal Constraints
| Reference | Requirement | Criticality |
|---|---|---|
| GDPR Art. 6 | Explicit consent required — minor prospects require parental consent | CRITICAL |
| GDPR Art. 17 | DELETE /leads/{id} endpoint operational within 72h | CRITICAL |
| GDPR Art. 21 | Opt-out → immediate stop + data anonymization | CRITICAL |
| CNIL · Retention | Automatic deletion after 3 years if not converted | MAJOR |
| Sensitive data | Passport/ID data collected only at booking, not at lead stage | MAJOR |
RACI Matrix
| Stakeholder | Role | Framing | Specs | Tests | UAT |
|---|---|---|---|---|---|
| Sarah Morel | Product Owner | A | A | I | A |
| Antoine Lebrun | Marketing Lead | C | C | I | R |
| Claire Vidal | GDPR Counsel | C | R | C | A |
| Karim Benali | Tech Lead | I | C | R | I |
| Emma Rousseau | Sr Travel Advisor | C | C | R | R |
| Zahra ★ | Business Analyst | R | R | R | R |
R = Responsible · A = Accountable · C = Consulted · I = Informed
03
Functional Specifications
Business Rules
| ID | Description | Priority |
|---|---|---|
| RG-T01 | Consent required — parental consent required if prospect < 18 years old | CRITICAL |
| RG-T02 | Score < 4 → Cold segment → automatic email nurturing sequence | CRITICAL |
| RG-T03 | Score ≥ 7 → destination specialist advisor contacts within 24h (SLA) | CRITICAL |
| RG-T04 | Prospect data deleted after 3 years if not converted | CRITICAL |
| RG-T05 | Opt-out → immediate stop of all communications + anonymization | CRITICAL |
| RG-T06 | Lead assigned to the specialist advisor for the chosen destination | MAJOR |
| RG-T07 | Duplicate (email OR phone) → silent merge + log + advisor notification | MAJOR |
| RG-T08 | Consent timestamp archived (GDPR audit trail) | CRITICAL |
| RG-T09 | Score 4–6 → advisor assignment within 48h | MAJOR |
| RG-T10 | Average AI score < 6 for a destination → alert to marketing manager | MAJOR |
User Stories — BDD Format
US-T01 · Submitting a language-travel inquiry
Must Have · 5 pts
"As a prospect interested in a language-immersion trip, I want to fill out an online form so that I can be contacted by a LinguaWise advisor specializing in my destination."
- Required fields: Last Name, First Name, Email, Phone, Destination, Duration, Budget, Desired Date
- Explicit GDPR checkbox — parental consent mandatory if age < 18
- Automatic confirmation email within 5 minutes
- Lead created in Salesforce with destination and budget segment
- Consent timestamp archived (RG-T08)
US-T02 · Automatic scoring by the AI agent
Must Have · 8 pts
"As a marketing manager, I want every prospect to be automatically scored based on their profile and destination so that I can prioritize advisors."
- Score computed on a 1–10 scale in under 3 seconds
- Criteria: source (25%), departure urgency (20%), budget (20%), traveler profile (20%), time of day (15%)
- Classification: Cold (1–3) / Warm (4–6) / Hot (7–10)
- AI justification visible in Salesforce within 2 minutes (e.g., "Budget > €3,000, departing in 6 weeks")
US-T03 · Assignment to the destination specialist advisor
Must Have · 8 pts
"As a sales director, I want every lead to be automatically assigned to the advisor who is an expert in the requested destination."
- Assignment based on chosen destination AND advisor availability
- If specialist unavailable → escalation to the senior travel advisor (Emma Rousseau)
- Advisor notification: email + Salesforce alert upon assignment
- Hot SLA: 24h | Warm SLA: 48h (RG-T03, RG-T09)
US-T04 · Right to object management (GDPR opt-out)
Must Have · 5 pts
"As a prospect, I want to be able to withdraw my consent at any time via a link in every email I receive."
- Unsubscribe link present in every communication
- Opt-out processed and confirmed within 72h (GDPR Art. 21)
- Immediate stop of all marketing communications
- Data anonymized in Salesforce CRM
US-T05 · KPI dashboard by destination
Should Have · 5 pts
"As a marketing manager, I want to view performance indicators for each destination in real time."
- Lead volume, conversion rate, average score per destination
- Filters: destination, date, channel, segment · Refreshed every 60 minutes
- CSV export for monthly reporting
US-T06 · Automatic deduplication
Should Have · 3 pts
"As an advisor, I want to avoid receiving the same inquiry twice for the same destination."
- Detection on email AND phone for every new submission
- Silent merge with destination/budget enrichment if changed
- Notification to the original advisor with the diff
04
Technical Architecture & AI Agent
Technology Stack — same as the Banking vertical
🖥️
Streamlit
Travel form + KPI dashboard⚡
FastAPI
REST API — lead flow orchestration🤖
Claude API
AI scoring — destination + profile✅
Pydantic v2
Travel business schema validation☁️
Salesforce
Simulated CRM — leads & assignments🚀
Render
Cloud deployment — app available onlineData Flow — End-to-End Architecture
1
Streamlit — Travel form
Prospect enters destination, budget, desired dates + GDPR consent
2
FastAPI — POST /leads <0.5s
Pydantic validation (valid destination, budget > 0, consistent dates), deduplication
3
Claude API — AI Agent <3s
Traveler profile analysis, departure urgency, budget — 1–10 score + text justification
4
FastAPI — Assignment Engine <0.5s
Selection of the destination specialist advisor + availability check
5
Salesforce Mock — CRM <0.5s
Lead record created with destination, score, segment, assigned advisor
6
Email + Notification <5 min
Confirmation email to the prospect + destination advisor alert
AI Scoring Matrix — Travel Vertical
| Criterion | Weight | Max Score | Min Score |
|---|---|---|---|
| Lead source | 25% | Direct form / referral | Untargeted social ad |
| Departure urgency | 20% | Departure < 6 weeks | Departure > 12 months |
| Stated budget | 20% | > €3,000 | < €1,000 |
| Traveler profile | 20% | Professional, 25–40 years old | Minor without parental consent |
| Submission time | 15% | Mon–Fri 9am–6pm | Night or weekend |
Segment Classification
Cold (1–3)No assignment — Email nurturing: D+3 (destination brochure), D+7, D+14
Warm (4–6)SLA 48h — Destination advisor + personalized quote
Hot (7–10) ★SLA 24h — Senior advisor + priority offer + SLA follow-up
05
Test Strategy & Validation Results
Overview
| Level | Tool | Coverage | Status |
|---|---|---|---|
| Unit tests | pytest | 90% of code | DONE |
| Integration tests | pytest + httpx | All endpoints | DONE |
| Functional tests | Squash TM | 6 US / 10 BR | IN PROGRESS |
| GDPR tests | Squash TM | 5 GDPR articles | IN PROGRESS |
| UAT tests | Excel / Notion | 5 scenarios | TO SCHEDULE |
| Performance tests | Locust | 3 load scenarios | TO SCHEDULE |
Functional Test Cases
| ID | Title | Status |
|---|---|---|
| CT-T01 | Valid form (destination London, budget €2,500) | PASSED |
| CT-T02 | Blocked without GDPR consent | PASSED |
| CT-T03 | Invalid email blocked (422) | PASSED |
| CT-T04 | Hot lead scoring (budget >€3K, departure <6 weeks) | PASSED |
| CT-T05 | Cold lead scoring (budget <€1K, departure +12 months) | PASSED |
| CT-T06 | Assignment to the New York specialist advisor | TO RUN |
| CT-T07 | GDPR opt-out via email link | TO RUN |
| CT-T08 | Duplicate (same email, different destination) → enrichment | TO RUN |
| CT-T09 | Parental consent required if age < 18 | TO RUN |
06
KPI Tracking & Delivery Summary
Target KPI Dashboard
| Indicator | Baseline | Target | Current Result |
|---|---|---|---|
| Lead → booking conversion rate | <10% | 18% | To be measured |
| Advisor callback delay | >72h | <24h | ~45s ✔ |
| Leads contacted within 48h | — | 85% | To be measured |
| Opt-in consent rate | — | 65% | To be measured |
| Average qualified score (≥ 6/10) | — | ≥ 6 | To be measured |
| AI scoring response time | — | <3s | ~1.8s ✔ |
BA Skills Demonstrated
Requirements gathering & analysisAdvanced
User Story & spec writingAdvanced
GDPR regulatory complianceAdvanced
Applied AI scoring (Claude API)Intermediate
Multi-vertical test strategyAdvanced
White-label adaptation (config.py)Advanced
07
Appendices — Business & CRM Glossary
- Lead CRM
- A prospect who has expressed commercial interest and provided explicit GDPR consent
- Opt-in GDPR
- A prospect's explicit, informed consent to have their data processed for marketing purposes
- Opt-out GDPR
- The act by which a prospect withdraws consent — triggers the immediate stop of all processing
- Scoring AI / CRM
- Assigning a 1–10 score that assesses a lead's maturity and conversion potential
- Hot Segment CRM
- Lead with score ≥ 7 — contact SLA: 24h — priority advisor assignment
- Warm Segment CRM
- Lead with score 4–6 — contact SLA: 48h — standard processing
- Cold Segment CRM
- Lead with score 1–3 — no assignment — automatic nurturing sequence
- Nurturing Marketing
- A sequence of automated communications designed to mature a lead toward purchase readiness
- White-label Product
- Architecture that lets the same LeadWise engine be deployed into a new industry via config.py alone
- SLA Contract
- Service Level Agreement — a committed level of service (e.g., contact within 24h)
- RACI Governance
- A responsibility-assignment matrix: Responsible, Accountable, Consulted, Informed
- Audit trail Compliance
- A traceability log that allows the full history of every system action to be reconstructed
Author's note: This document covers both verticals of the LeadWise project — Banking (BankWise Retail) and Travel (LinguaWise International). All names, data and scenarios are fictional and created for educational purposes. Source code is available on GitHub — zahraeh/LeadWise.
Want to know more?
Check out the source code, the Notion documentation, or my other projects.