ABMS
One Morning 'Untangling' 5,000+ Leads: How AI Helped NexaFlow Clean Up Their Leads Module
ABOUT NexaFlow
Leads pour in from 5 different sources: Website, Zalo OA, Facebook Ads, SMS Broadcast, Telesales. Each source brings its own set of fields. But all are 'dumped' into one block — making the UI a place that stores both business data and technical data. Important information drowns among import fields. Sales have to 'dig' for data instead of processing leads.
RESULTS ACHIEVED
Block structure: from 1 chaotic block to 10 purposeful blocks. Duplicate fields eliminated: 10 pairs removed. Import fields separated: 22 fields in dedicated block (can hide). Header when opening record: Name · Company · Status. QuickCreate: 6 fields in natural sequence. Data safety: 0 records lost. Time: One morning instead of 'no one dares to touch'.
KEY RESULTS
10
Hidden duplicates
22
Import fields separated
10
New structured blocks
6
QuickCreate fields
0
Records lost
1
Morning
BUSINESS CHALLENGES
One Chaotic Block
All data from 5 sources dumped into one block. Sales staff don't know where to look first. Important info drowns in technical fields.
10 Hidden Duplicates
10 duplicate fields existed but no one noticed. Same data under different names causing filter and report inconsistencies.
22 Import Fields Mixed In UI
Technical tracking fields (uid, tracking_code, sms_batch_id...) mixed with business fields, making the interface cluttered and hard to use.
QuickCreate Sequence = 0
All fields had quickCreateSequence = 0, making the quick create form unusable. No one had time to fix the 'basic stuff'.
Read Data Structure from PostgreSQL
AI read entire structure: blocks, fields, column names, data types
Cross-checked with Prisma schema to detect inconsistencies
SQL query: SELECT fieldname, fieldlabel, block, typeofdata...
Detect Duplicates — Field Right in Front but No One Handled
AI compared labels/fieldnames, checked data patterns
Example: salutation vs salutationType — both store titles but different names
Prevented errors in filters and reports
Separate 22 Import Fields from Main UI
Metadata from SMS broadcast: uid, tracking_code, timestamp, batch_id
For power users, this is 'junk drawer'
Moved all to Block 10 (hidden) — kept for traceability, can hide/collapse
Self-Check & Detect Missed Field (cf_1764)
After restructuring, AI re-ran checklist
Found cf_1764 (Position/Title) still visible in old block, not moved yet
Didn't just fix, also verified after fixing
QuickCreate: First Time Truly 'Quick'
6 fields in actual input order
Name → Company → Phone → Email → Source → Owner
Before: sequence = 0. After: purposeful sequence.
CLIENT TESTIMONIALS
"Leads pour in from Facebook, Zalo, website, SMS... each channel different. Staff open records but don't know where to look first. AI read data structure, analyzed, restructured. Almost no briefing needed. It understood operational logic very quickly."