# First Projects

We begin with two parallel verticals: Customer Support (Atendimento) and Financial Operations (FinOps). Both have engaged stakeholders, bounded scope, and pain points that are felt daily.

These aren't pilot projects that get discarded. They're the first production implementations of the governed data platform.

---

# Customer Support (Atendimento)

# The Current State

Atendimento relies heavily on Metabase for dashboards and as an informal CRM. The team views customer data, tracks interactions, and monitors operational metrics. Data flows in from multiple sources:

• Total IP - The telephony platform (dialer, inbound calls, 0800)
• Infobip - Communication channels (WhatsApp, chatbot, push notifications)
• Google Analytics - Web interaction data

Léo (Data Engineering) has built views that consolidate communication data, and the team consumes these through Metabase dashboards.

# The Pain

• Unrestricted access - Everyone can query any table, including production databases and raw logs. There's no permission model.
• Expensive queries - Log tables stored as single JSON columns require full scans. One misconfigured dashboard cost tens of thousands of reais.
• No historical tracking - To understand how an indicator looked last month, analysts manually reconstruct it. There's no time-travel or snapshot capability.
• Duplicated dashboards - Teams copy existing dashboards and tweak filters. The same KPI exists in dozens of versions with no clear "official" source.
• Blocked external sharing - Security concerns (LGPD, lack of granular permissions) prevent sharing dashboards with external partners like retailers.

# MVP Scope [to be validated]

The MVP focuses on establishing governed data flows for Atendimento's core needs:

1. Consolidate data sources - Bring Total IP, Infobip, and GA data into governed tables with proper schemas, partitioning, and documentation.
2. Create certified views - Replace ad-hoc queries with maintained, tested views or physical datamarts that serve as the single source of truth.
3. Catalog everything - Register data assets in the data catalog with ownership, lineage, and quality indicators.
4. Migrate key dashboards - Move priority dashboards to the governed data sources. Flag migrated vs. legacy in the catalog.
5. Establish access controls - Implement proper permissions so teams access only what they need.

# Expected Benefits

• Reduced query costs - Optimized tables and views eliminate expensive full-scan queries.
• Single source of truth - One version of each KPI, documented and owned.
• Historical visibility - Time-travel capability lets teams see how indicators evolved.
• Path to external sharing - With proper access controls, sharing with partners should becomes viable.
• Template for other teams - Cobrança (Collections) has similar patterns and can follow the same playbook.

---

# Financial Operations (FinOps)

# The Current State

FinOps handles two critical processes: credit assignment (transforming loans into CCBs for fund purchase) and reconciliation (matching payments to contracts and reporting to fund managers).

The reconciliation process runs daily:

1. A Spark pipeline queries internal tables and builds a CSV file
2. The file is uploaded to the fund manager (Milênio) via API
3. Milênio processes and returns events via webhook (approved, rejected, errors)
4. The team handles exceptions and resubmits as needed - all before 4pm

Key systems involved:

• IUGU - Payment provider, source of truth for boleto payments
• Milênio API - Fund manager integration for reconciliation
• Internal BigQuery tables - Contracts, installments, invoices, proposals

# The Pain

• Incorrect data sent daily - The pipeline pulls payment values from the wrong table. The correct information exists in IUGU/invoices, but the current process uses installments, which has discrepancies.
• Overly complex pipeline - The Spark job has grown into a maze of transformations with business logic scattered across multiple places. No one fully trusts what it does.
• No reliable history - When something goes wrong, there's no trustworthy audit trail to understand what was sent and when.
• Time pressure - The 4pm deadline means errors discovered late can't be properly investigated. The team often resubmits knowing issues remain.
• Renegotiation complexity - The way renegotiated contracts are modeled creates additional reconciliation challenges that compound the data quality issues.

# MVP Scope [to be validated]

The MVP focuses on creating a reliable, auditable reconciliation process:

1. Establish source of truth for payments - Create a governed table that pulls from IUGU (the correct source), properly transformed and validated.
2. Rethink retrieving data from BigQuery - think of an intermediate relational source of data that gets replicated data out of transactional services.
3. Simplify the pipeline - Replace the complex Spark job with a clearer transformation (likely dbt) that's easier to understand, test, and maintain.
4. Enable time-travel - Structure data so the team can query "what did we know on date X" for any reconciliation period.
5. Capture webhook responses - Persist Milênio's responses in queryable tables, not just logs, enabling analysis of rejection patterns.
6. Build operational visibility - A dashboard showing reconciliation status, error rates, and trends - before the 4pm deadline.

# Expected Benefits

• Correct data to fund managers - Payments reconciled using the actual source of truth.
• Faster problem resolution - When Milênio rejects records, the team can quickly identify why and resubmit correctly.
• Audit trail - Full history of what was sent, when, and what the response was.
• Reduced operational stress - Confidence in the data means less firefighting and fewer last-minute scrambles.
• Foundation for credit assignment - The same patterns (governed sources, clear transformations, audit trail) apply to the credit assignment process next.

---

# Shared Patterns

Both MVPs establish patterns that scale. The Tooling page details the specific tools that enable these patterns.

The tools and practices proven in these verticals become the blueprint for expanding to other areas: Cobrança, Growth, Credit, and beyond.

---

# Key Stakeholders

These are the people who feel the pain daily and will validate whether the solutions actually work.