Warning
🚧 Work in Progress: This page is currently under construction. Content may be incomplete or subject to change. To contribute, see the contribution guide.
ADR-NNN: [Decision Title]
| Field | Value |
|---|---|
| Status | Proposed |
| Date | 2026-05-07 |
| Decision makers | Leandro Crespo |
| Supersedes | N/A |
Use Case Catalog — Integration Platform Selection
Companion to [PAX-EA] 4. Integration Architecture — §5.0 Integration Platform Selection Guide.
Maps concrete Patria scenarios to the recommended platform (or combination of platforms) based on the current decision matrix.
0. Preliminary Notes
Integration platform vs. compute runtime. Sensedia API Management exposes and governs APIs, but the backend implementing the business logic runs on a runtime (Cloud Run, Cloud Functions, FastAPI on Cloud Run). These runtimes are not integration platforms and must not be evaluated against Sensedia/N8N/Airflow/etc.
Mandatory governance. Every API consumed by third parties (clients, counterparties, regulators) or published to the internet must go through Sensedia API Management — this is not optional; it is a mandatory security control (auth, rate limiting, observability, auditing).
1. Quantitative Criteria for Tiebreaking
When a use case fits more than one platform, apply the following thresholds:
| Dimension | Threshold | Indicated Platform |
|---|---|---|
| Volume per execution | > 100k records | Airflow (Cloud Composer) |
| Volume per execution | < 10k records and latency matters | API Management + microservice |
| Required p95 latency | < 200 ms | API Management + Cloud Run (min instances ≥ 1) |
| Event frequency | > 100 events/sec | Sensedia Event Hub |
| Criticality | Regulated production / external client / contractual SLA | NOT N8N, NOT UiPath as permanent production |
| Transformation volume | GB–TB | Airflow + BigQuery |
| Transformation volume | KB–MB inline | Sensedia iPaaS |
2. Data Exposure and Consumption via API
| Concrete Use Case | Platform(s) | Rationale |
|---|---|---|
| Simple DB query/extraction (e.g., fund positions from Geneva) | Sensedia API Management + microservice in FastAPI on Cloud Run | Encapsulate data access in a versioned service layer |
| Exposing logic currently residing in a stored procedure | Microservice (Cloud Run + FastAPI) that invokes the SP + Sensedia API Management | Avoid coupling the API contract to the DB schema; the SP remains an internal detail |
| Low-latency API (< 200 ms p95) for point queries with no business logic | Sensedia API Management + Cloud Run with min instances ≥ 1 | Cloud Functions discarded due to cold start |
| Aggregation of multiple backend systems into a single endpoint (BFF) | Sensedia API Management + microservice on Cloud Run | Cloud Run to maintain backend connection state |
Allowed exception: expose an SP directly via iPaaS (Camel) only if (a) the SP is already stable, (b) the cost of a microservice is not justified, and (c) the contract is documented as internal/transitional.
3. System-to-System Integration
| Concrete Use Case | Platform(s) | Rationale |
|---|---|---|
| Sending information between systems via synchronous API (e.g., price review) | Sensedia iPaaS + Sensedia API Management | There is transformation or routing between calls |
| Simple synchronous call without transformation | Sensedia API Management directly to the backend | iPaaS would be over-engineering |
| Event-driven notification when a fund valuation is updated | Sensedia Event Hub (publish) + consumer on Cloud Run | Standard pub/sub pattern |
| Near-real-time synchronization between Geneva and CRM | Preferred: Sensedia Event Hub (CDC from Geneva) + consumer. Alternative: iPaaS if complex orchestration/transformation across multiple systems is required | Events for decoupling; iPaaS only if orchestration logic is needed |
| Incoming webhook from external SaaS (e.g., signed DocuSign) → trigger internal production action | Sensedia API Management (reception) + Sensedia iPaaS | Production-critical → iPaaS, not N8N |
| Incoming webhook for non-critical internal automation / prototype | Sensedia API Management (reception) + N8N | Only if flow failure does not affect SLA toward clients/regulators |
4. Data Pipelines (High Volume / Batch)
| Concrete Use Case | Platform(s) | Rationale |
|---|---|---|
| Daily trade ingestion from OMS to data lake (Bronze → Silver → Gold) | Airflow (Cloud Composer) | Standard Medallion pattern |
| Nightly reconciliation Geneva ↔ data warehouse | Airflow (Cloud Composer) | High volume + task dependencies |
| XBRL ingestion from CMF to BigQuery | Airflow (Cloud Composer) — download via HttpOperator/PythonOperator + processing | Do not split across N8N+Airflow if Airflow can handle it alone |
| Scheduled market data extraction (Bloomberg, Refinitiv) | Airflow (Cloud Composer) + specific connector | Schedule + retry + lineage on a single platform |
| File download from portal with login/captcha/UI → load to GCS for Airflow to process | UiPath or N8N (download only) → GCS → Airflow | Split only when the download requires UI automation |
5. UI Automation / Legacy Systems Without API (RPA)
| Concrete Use Case | Platform(s) | Rationale |
|---|---|---|
| Automated login to banking portal to download statement | UiPath | UI with session/captcha/heavy JavaScript |
| Mass data entry in legacy desktop application without API | UiPath (unattended bot) | No alternative available |
| Regulatory report generation in system with proprietary UI | UiPath | No alternative available |
| Legacy data extraction → API exposure (transition) | UiPath (extracts) → Sensedia API Management (exposes) | Strangler pattern for modernization |
UiPath should be treated as a transitional solution. Whenever possible, expose the system via API and replace the bot with a Sensedia integration.
6. User Task Automation / AI-Augmented
| Concrete Use Case | Platform(s) | Rationale |
|---|---|---|
| Email download and formatting (e.g., broker confirmations) | N8N | Non-critical task, no external SLA |
| Document classification with LLM (KYC, contracts) | N8N + MCP / LLM | Internal productivity; if promoted to regulated production, migrate to Cloud Run |
| Smart notification in Slack/Teams on business event | N8N | Low volume, non-critical |
| Web scraping of simple HTTP/HTML source without login + load to GCS | N8N | No complex authentication or heavy JS |
| Web scraping with login/session/heavy JS | UiPath | UiPath’s native capability |
| Quick prototype connecting SaaS tools (Notion, Asana, Slack) | N8N | Iteration speed |
N8N must not execute processes where a failure affects SLA toward clients, counterparties, or regulators. For those cases, migrate to iPaaS / Cloud Run.
7. Approval Workflows / BPM
| Concrete Use Case | Platform(s) | Rationale |
|---|---|---|
| Purchase order approval with multiple approval levels | Fluig | Formal approval with audit trail |
| Employee onboarding (forms + signatures + routing) | Fluig | Human workflow with compliance requirements |
| Contract approval → upon approval, provision access in target system (no transformation) | Fluig (approval) + Sensedia API Management → target system | Direct API call post-approval |
| Contract approval → upon approval, execute integration with transformation/orchestration | Fluig (approval) + Sensedia iPaaS | There is transformation between Fluig and the target system |
| IT change request / incident | ServiceNow (not Fluig) | ITSM, not business BPM |
Fluig vs. ServiceNow: Fluig manages business workflows (procurement, contracts, HR). ServiceNow manages ITSM (incidents, change requests, ticketing). They are complementary.
8. Typical Hybrid Scenarios at Patria
| Concrete Use Case | Combination |
|---|---|
| Bot downloads file from legacy portal → pipeline processes it | UiPath → GCS → Airflow (Cloud Composer) |
| Expense approval triggers direct call to ERP | Fluig → Sensedia API Management → ERP |
| Expense approval requires transformation before reaching ERP | Fluig → Sensedia iPaaS → ERP |
| LLM interprets PDF → publishes event → consumer updates CRM (production) | Cloud Run (LLM) → Sensedia Event Hub → consumer |
| Same flow in prototype / non-critical phase | N8N (LLM) → Sensedia Event Hub → consumer |
| API exposes data computed by batch pipeline | Airflow (computes and persists in BigQuery) → Sensedia API Management on Cloud Run |
9. Anti-Patterns to Avoid
| Anti-Pattern | Correct Platform |
|---|---|
| N8N as production API gateway | Sensedia API Management |
| N8N as high-volume ETL (> 1M records/day) | Airflow (Cloud Composer) |
| N8N in regulated production process / with external SLA | Sensedia iPaaS or Cloud Run |
| UiPath over a system that already exposes an API | Sensedia |
| Airflow for webhooks or synchronous flows | Sensedia (API Management / iPaaS) |
| Airflow for processes requiring human decision / approval | Fluig |
| Fluig for IT tickets | ServiceNow |
| Fluig for technical integrations without a human component | Sensedia |
| Sensedia for heavy transformations (GB–TB) | Airflow + BigQuery |
| Sensedia API Management for heavy batch processing | Airflow |
| Stored procedure exposed directly as a REST API | Microservice (Cloud Run + FastAPI) + API Management |
| Cloud Functions for APIs with latency < 200 ms p95 | Cloud Run with min instances ≥ 1 |
| Chaining 3+ platforms for a simple use case (fragmentation) | Consolidate in the platform with the best primary fit |
| External / internet-facing API without going through Sensedia API Management | Always Sensedia API Management (security mandate) |
10. Decision Tree
Apply in order; take the first match.
- Batch pipeline / high volume (> 100k records) or heavy transformations (GB–TB)? → Airflow (Cloud Composer)
- Legacy system without API requiring UI automation (login, captcha, desktop)? → UiPath (transitional)
- Formal approval with audit trail, signature, compliance, or human routing? → Fluig (or ServiceNow if ITSM)
- Non-critical internal automation, AI-augmented, or prototype, without external SLA? → N8N
- Asynchronous / event-driven communication (pub/sub, CDC, near-real-time decoupled)? → Sensedia Event Hub
- Synchronous request-response communication with transformation / orchestration / routing? → Sensedia iPaaS (+ API Management if exposed externally)
- Synchronous request-response communication without transformation? → Sensedia API Management directly to the backend (Cloud Run / Cloud Functions / FastAPI) If there is still ambiguity between two platforms, apply the quantitative criteria from §1.
