Text Summarization
How a Chatbot Cut CrediFlow’s Support Tickets by 45 % and Lifted CSAT to 82 %
FinAssist – the AI‑First Financial Advisor that turned a support nightmare into a growth engine in just six weeks.
1. My Role & the Team
Creative Strategist & Technologist — end‑to‑end ownership of insight, conversational UX, no‑code architecture, KPI tracking.
Team — 1 Product Owner, 2 UI/UX Designers, 1 Data Engineer, & 1 Compliance Officer.
Timeline — 10 weeks (Research & Insight 3w → Ideation, Concepting, & Prototyping 2w → No‑Code Build, Integration, & Security QA 1w → Soft Launch, Monitoring, & Full Rollout 1w)
The Situation – A Fintech Growing Too Fast
CrediFlow is a B2C fintech that hands out micro‑loans (USD $100‑$2,000) across Mexico, Colombia, Brazil, and Argentina.
Growth: 30 % month‑over‑month, ≈ 150 k new users in the last three months.
Pain: The support inbox was drowning in repetitive questions—eligibility, repayment dates, interest‑rate explanations.
Current tech: A static HTML FAQ that never spoke to the loan‑engine API.
The result? 1,200 tickets per day, an average first‑response time of 6 hours, and an SLA breach rate that far exceeded the 2‑hour target. Customer satisfaction (CSAT) lingered at 68 % and loan‑approval conversion stalled at 22 % because users abandoned the process after getting stuck on “Am I eligible?”.
2. The Core Challenge
Dimension | Pain Point |
|---|---|
Strategic | 1,200 tickets/day → 6 h first‑response → SLA breaches |
Technical | Static help‑center, no API exposure for the loan engine |
Regulatory | Must stay LGPD (Brazil) & GDPR‑EU compliant – no plain‑text personal data, full audit logs |
Financial | Support labor ≈ $28,800/month (30 agents @ $960 each) |
Baseline | Tickets = 1,200/day, CSAT = 68 %, conversion = 22 % |
3. Defining Success – The KPI Blueprint
Goal | Target |
|---|---|
Business | ↓ tickets ≥ 40 % (≈ 660/day) |
Creative | Position FinAssist as a trusted financial advisor – friendly, transparent, jargon‑free |
Technical | Full integration with loan‑engine via Integromat (REST) |
Quantifiable | Bot handles ≥ 55 % of tickets |
Financial | Payback ≤ 4 weeks (budget $4,800 vs. labor savings) |
4. Digging Deep – Research & Insight
Ticket Mining
62 % of tickets were about eligibility & repayment schedules. The average ticket required four back‑and‑forth messages.
User Interviews (25 participants)
“I just want to know if I qualify instantly, not wait for an email.”
“When I don’t understand the interest rate, I give up.”
Competitor Scan
Three LATAM fintechs all used rule‑based bots that only answered FAQs. None offered real‑time eligibility or repayment calculations.
Data Insight
Customers who self‑served eligibility in a prototype converted 1.8× more often.
Regulatory Scan
LGPD demands explicit consent before any personal data is processed; GDPR adds a right‑to‑be‑forgotten workflow.
Key Insight
Speed and transparency are the biggest trust drivers for micro‑loan seekers. A conversational eligibility checker can eliminate friction, reduce support load, and boost conversion.
5. From Insight to Idea – Ideation & Concept
Core Artifacts
Storyboard – User asks “Can I get a $500 loan?” → Bot collects ID, calls the loan‑engine, returns “You qualify for $500 at 12 % APR. Apply now or see repayment schedule?”
Mood board – Trust‑blue palette (#0A4D8C), rounded icons, Montserrat typeface.
Concept cards – Eligibility Quick‑Check, Repayment Calendar, Escalation to Live Agent.
Tone guide – Use “tú” (Spanish) / “você” (Portuguese), replace “interest rate” with “costo del préstamo”, keep sentences ≤ 12 words.
Feasibility Decision
Tool | Why it fit |
|---|---|
Chatfuel (Pro) | Multi‑language, visual flow builder, API block, HubSpot integration. |
Integromat (Standard) | Handles REST calls, error handling, caching, LGPD‑compliant data store. |
Twilio SMS (future) | Extends bot to users without internet. |
Budget | Chatfuel $39/mo + Integromat $29/mo + Twilio $100 ≈ $4,800 for the 10‑week project. |
Creative Brief (in one sentence)
Build a bilingual, no‑code chatbot that instantly validates loan eligibility, explains repayment terms, and escalates complex issues to a human agent—while sounding like a trusted financial advisor.
Ideation Techniques
Method | Output |
|---|---|
6‑2‑5 Brainwriting | 30 micro‑features (instant credit‑score preview, rate‑calculator widget, etc.) |
Jobs‑to‑Be‑Done Mapping | Core JTBD: “When I need a quick loan, I want to know instantly if I qualify so I can decide fast.” |
Crazy‑8 Sketches | Rapid UI concepts for chat bubbles, quick‑reply chips, progress bars |
6. Architecture – How the Bot Lives Without Code
Prototyping
Figma interactive mock‑up of the chat UI.
Chatfuel sandbox with mock API responses via Integromat’s “JSON > Set Variable”.
90‑second Loom video showing the eligibility check in both languages and the fallback to a live agent.
Remote usability test (8 participants) – average answer time 22 s, sentiment +0.8 NPS.
Data flow in a nutshell
User types a question → Chatfuel captures intent & parameters.
Chatfuel fires a JSON webhook to Integromat.
Integromat authenticates, calls
/eligibility, caches the result for 5 min.Response is mapped back to Chatfuel variables and displayed.
If the bot fails twice, Integromat creates a Zendesk ticket and hands the conversation to a live agent.
7. Building & Launching – The Execution Sprint
Phase | What We Delivered | Collaboration Highlights |
|---|---|---|
Beta Launch | Soft rollout to 10 % of mobile‑app users (≈ 12 k) | PO reviewed daily KPI snapshots; we tweaked caching after the first hour’s latency spike. |
Full Rollout | 100 % of web & mobile traffic | Cross‑functional demo day with Data Engineer (API logs) and Designers (brand tone reinforcement). |
Phase | What We Delivered | Collaboration Highlights |
|---|---|---|
Build | 15‑block Chatfuel flow (welcome, language selector, eligibility, repayment, apply, escalation) | Daily stand‑ups with the Compliance Officer to lock down consent UI and data‑retention. |
Security & QA | LGPD/GDPR audit (external consultant) | Support team performed “shadow chats” to validate tone and escalation logic. |
8. Hurdles & How We Overcame Them
Challenge | Impact | Solution |
|---|---|---|
API latency (avg 1.2 s) | Threatened the ≤ 45 s handling goal. | Added a 5‑minute cache in Integromat; introduced a subtle “loading” animation in Chatfuel to set expectations. |
Language nuance (Spanish vs. Portuguese) | Risk of inconsistent tone. | Built a language‑specific tone guide (200+ approved phrases) and stored them as Chatfuel User Attributes; conditional blocks pull the right copy. |
Escalation loops | Potential endless tickets. | Implemented a Human‑Assist block with a timeout rule: after 2 consecutive failures, auto‑create a Zendesk ticket and hand over the conversation ID. |
Data‑privacy compliance | Storing CPF/CURP could breach LGPD. | Added an explicit consent checkbox before any personal data is captured; all data lives only in Integromat’s encrypted store and is purged after 24 h. |
Budget ceiling | Must stay under $5k. | Negotiated a 6‑month Chatfuel Pro trial and used Integromat’s free 2,000‑operation tier for month 1; scaled to Standard only when traffic justified it. |
9. The End Product – FinAssist
Component | What It Looks Like |
|---|---|
Chatbot UI | Bilingual (ES/PT) embedded on CrediFlow’s web portal & mobile app. Rounded icons, trust‑blue palette, progressive disclosure, micro‑animations for loading & success. |
Core capabilities | Real‑time eligibility check, repayment schedule generation, loan‑application initiation, seamless escalation to live agents. |
Technical stack | Chatfuel Pro (API block, multi‑language) |
Performance | 99.9 % uptime SLA, average bot response 22 s, total handling ≤ 45 s. |
Scalability | Architecture supports up to 5 M monthly interactions with minimal extra cost. |
Documentation | Full flow diagrams, API schema, tone guide, SOP for escalation handed to the support team. |
10. Results – Numbers That Speak
Metric | Before | After 6 weeks | Δ |
|---|---|---|---|
Support tickets | 1,200 /day | 660 /day | ‑45 % |
CSAT | 68 % | 82 % | +14 pts |
Eligibility queries handled by bot | 0 % | 58 % | — |
Average handling time | 6 h (human) | 42 s (bot) | — |
SLA compliance | 78 % (≤ 2 h) | 96 % | +18 % |
Support‑labor cost saved | — | $28,800 (30 agents × $960) | — |
Bot‑initiated loan applications | 0 % | 12 % of total applications | — |
Loan‑approval conversion | 22 % | 27 % | +5 pp |
Payback period | — | 3.5 weeks (budget $4,800 vs. $28,800 saved) | — |
Live KPI Dashboard
Created a Google Data Studio view that shows ticket volume, CSAT trend, bot latency, and conversion funnel in real time.
What Users Said
87 % of post‑chat survey respondents said “the answer was instant.”
81 % felt “more confident about the loan.”
Support agents reported a 30 % reduction in repetitive tickets, freeing them to handle complex cases. The executive team highlighted FinAssist as a key differentiator in the next funding round deck.
11. What We Learned & What’s Next
What We’d Iterate
Area | Planned Improvement |
|---|---|
Credit‑score education | Add a micro‑learning module (“What does my score mean?”) with visual gauges. |
Push notifications | Use Twilio SMS to remind users of upcoming repayments and to re‑engage dormant applicants. |
AI‑enhanced intent detection | Pilot OpenAI GPT‑4 via Integromat for free‑form queries (e.g., “What if I miss a payment?”). |
Self‑service loan application | Extend flow to capture full application data, pre‑fill from eligibility check, and submit directly to loan‑engine. |
Analytics enrichment | Feed bot interaction data into Amplitude for cohort analysis and predictive churn modeling. |
Influence on Future Work
FinAssist became the foundation for CrediFlow’s entire conversational ecosystem (onboarding, upsell, collections).
The caching pattern is now a standard practice for all API‑driven bots we build.
The tone‑guide framework is reused across 3 additional fintech clients in the region.
New Skills & Tools Mastered
Integromat advanced error‑handling (retry‑with‑back‑off, conditional branching).
Multilingual content pipelines – building language‑specific attribute maps in Chatfuel.
LGPD compliance checklist – consent UI, data‑retention policies, audit‑log generation.
Load testing for no‑code platforms (LoadNinja + custom webhook simulators).
Ready to Turn Your Support Bottleneck into a Revenue‑Generating Conversation?
If you’re looking to automate high‑volume financial queries without a single line of code, let’s explore how a tailored chatbot can cut costs and boost trust. Contact me at hello@yourname.com or book a strategy session here.