Case Study

Azonova Sales - From unstructured contact files to an actionable pipeline

Azonova Sales is a focused CRM that turns PDFs, spreadsheets, documents, photographed tables, device contacts, and manual notes into reviewed, structured sales records. It then helps users prioritize leads, move opportunities through a pipeline, and act on follow-ups.

AI extraction Review before save Lead pipeline WhatsApp workflows Owner-isolated data
Open Azonova Sales Read the Azonova Sites case study
Azonova logo

Import, verify, prioritize, and follow up

The product combines document intelligence with practical CRM controls. AI accelerates data entry, while users remain in control of what is saved and how each lead is handled.

Contents

1) Product problem and design principle

Valuable sales contacts often arrive in formats that traditional CRMs handle poorly: event lists, scanned tables, PDFs, spreadsheets, screenshots, address books, and informal notes. Manual entry is slow, while direct AI import can create silent errors.

  • Reduce repetitive entry: extract many contact fields from mixed source material.
  • Keep human control: show a sample and editable records before committing them.
  • Make records actionable: connect imported data directly to stages, priority, scoring, and follow-up.

The core principle is assisted accuracy: AI prepares the work, and the user approves the CRM state.

2) End-to-end sales workflow

Source material to sales action
Product flow
Import
Files, images, device contacts, or manual entry
Extract
AI schema output and source-aware processing
Review
Correct fields and confirm records before save
Act
Prioritize, contact, follow up, export, and close

3) Multi-source AI import pipeline

Import is designed as a pipeline rather than a single upload box. Each source type has different quality risks, so preprocessing, extraction, and status tracking are kept explicit.

Import and extraction pipeline
Document intelligence
Capture
PDF / sheet / image / contacts
  • Accept structured files, documents, photographed lists, screenshots, and supported device contacts.
  • Record source type, filename, status, timestamps, and import metadata.
Prepare
quality recovery
  • Auto-orient and enhance difficult images using grayscale, normalization, sharpening, and resizing.
  • Create full-page, table-band, and field-focused image assets to improve extraction from dense layouts.
Extract
strict structured output
  • Request schema-constrained contact records rather than free-form model prose.
  • Capture identity, company, communication, location, qualification, source, and follow-up fields.
Review
user-controlled commit
  • Present extracted contacts for correction and selection before database insertion.
  • Keep failed, partial, and completed import states visible in history.

4) Review, normalization, and deduplication

AI output is treated as input, not truth. Records pass through bounded field normalization and can be reviewed before they become part of the user's working pipeline.

  • Consistent fields: names, contact details, locations, tags, stages, scores, and monetary attributes are normalized.
  • Bounded values: stages, entity types, priorities, temperatures, channels, ages, and scores use controlled ranges.
  • Duplicate checks: matching prefers email, then phone, then name and company combinations.
  • Source preservation: records retain import source and source-file context for later filtering and audit.

5) CRM model and pipeline operations

The data model supports people, companies, leads, and opportunities without forcing every imported row into the same shape. Users can move work through a clear sales lifecycle.

  • Pipeline: new, qualified, contacted, proposal, won, and lost stages.
  • Prioritization: 0-100 score, low-to-urgent priority, and cold/warm/hot lead temperature.
  • Segmentation: industry, segment, tags, company relationships, source files, and location fields.
  • Work queue: search, filter, sort, bulk operations, and next-follow-up ordering.

6) Communication, follow-ups, and exports

The CRM keeps common next actions close to the record instead of making users move data into separate spreadsheets or messaging tools.

  • Communication actions: phone, WhatsApp, email, website, LinkedIn, and in-person preferences.
  • WhatsApp templates: reusable messages with contact and company placeholders.
  • Follow-up planning: per-contact dates, notes, stages, and pipeline status.
  • Portable output: selected CRM fields can be exported to Excel, CSV, JSON, PDF, or PNG.

7) Security and operational traceability

Sales records are private working data. Authentication is enforced at API boundaries, ownership is checked on reads and writes, and database row-level policies isolate each user's contacts and imports.

  • Owner isolation: contact and import rows are keyed to the authenticated user.
  • Row-level security: database policies restrict select, insert, update, and delete operations.
  • Import history: source, status, extracted count, errors, and timestamps provide operational context.
  • Abuse controls: sensitive AI import flows support verification and bounded processing limits.

8) Engineering value

Azonova Sales demonstrates an end-to-end AI feature where model output is only one part of the system. The useful product includes source handling, image recovery, schemas, validation, review UX, secure persistence, operational history, CRM controls, and export.

Explore sales.azonova.com, or contact Azonova to build document intelligence, workflow automation, CRM features, or practical AI capabilities into an existing product.