System integration: API, RPA, or native connector — when to use each

When two systems do not talk, the reflex in many Mexican SMBs is «we need to integrate them.» But integrating can mean anything from a well-documented API to a robot clicking screens like a human. Choosing wrong yields expensive, fragile, or unauditable projects. This guide compares API, RPA, and native integration with criteria that matter in real operations: total cost, resilience to UI changes, traceability, and time to value.

At Devson Labs we see the same pattern: solid ERP APIs coexist with undocumented legacy portals and a spreadsheet «nobody can touch.» There is no silver bullet — only a conscious choice per system pair, aligned to data criticality and flow frequency.

API integration: the default when a clear contract exists

An API exposes structured operations: create order, query inventory, stamp invoice. Pros: typed data, formal auth, logs, versioning. Cons: needs documentation, test environments, sometimes license cost. Use it when the vendor offers stable REST/GraphQL, volume is high, or you need near-real-time bidirectional sync.

• Best for: CRM ↔ ERP, ecommerce ↔ WMS, billing ↔ PAC with SDK.
• Green flags: sandbox, OAuth or rotatable API keys, documented webhooks.
• Red flags: «beta» API with no SLA, opaque rate limits, unannounced breaking changes.

RPA: a temporary bridge — not permanent architecture

RPA replicates human actions on graphical UIs: open SAP, copy row, paste into portal. It works when there is no API and UI change is slow. It is fragile: a moved button breaks the flow. Maintenance cost grows with every source system update. Use it as a bridge toward migration or API, not as the backbone of critical revenue flows.

RPA saves the quarter; API saves the half-decade.

1. Identify no-API processes with low volume (<500 transactions/month).
2. Document each manual step with screenshots and exception rules.
3. Implement RPA with failure monitoring and alerts — not fire-and-forget.
4. Plan exit: target date for API, native connector, or system replacement.

Native connectors and iPaaS

HubSpot, Shopify, Salesforce, and many tools ship prebuilt connectors or marketplaces (Zapier, Make, n8n). They are fast for standard flows and reduce custom code. Limitation: complex logic, Mexican fiscal transforms, or very specific business rules may fall short. iPaaS shines at orchestrating SaaS; custom dev shines when generic connectors do not understand CFDI, regional price lists, or internal credit rules.

Practical decision matrix

• High criticality + high volume: API or official connector; avoid RPA.
• Legacy without API + low volume: bounded RPA with exit plan.
• Many standard SaaS tools: iPaaS first; custom only where it fails.
• SAT fiscal data: never RPA on SAT portal; use certified integration.
• Strict audit: API with immutable logs; RPA only with evidence recording.

Hidden costs nobody budgets

Initial quotes rarely include evolutionary maintenance: SAT catalog changes, new CRM version, OAuth credential rotation. A well-built API amortizes that cost. RPA multiplies it. Budget 15–25% annually of initial cost for critical integrations, plus monitoring and runbooks. If a vendor promises «integration in one week» without edge-case testing or idempotent retries, be skeptical.

How to start without analysis paralysis

List the five flows that burn the most hours between systems. Classify each pair with the matrix above. Prioritize one where API or connector exists and pain is measurable. Run a 4–6 week pilot with error and time metrics. At Devson Labs we combine n8n, APIs, and custom development as the business requires — we do not sell RPA where a $200/month API solves it, nor custom API where a native connector is enough.

Request an automation radiography if you have more than three disconnected systems and the team normalized chaos as «part of the job.» Integrating well is not picking the trendy tech — it is reducing operational risk while the business scales.

Real examples in Mexican SMBs

An industrial distributor connected Shopify Plus to ERP via API for inventory and pricing: sync every 15 minutes, alerts when floor stock differs from ecommerce. An accounting firm used RPA for six months to pull movements from a bank portal without API while negotiating SFTP access. A services company replaced Zapier with self-hosted n8n when ticket assignment rules exceeded 40 unmaintainable zaps. All three lessons matched: the right tech depends on system lifecycle, not vendor logo.

Avoid copying a US case study stack blindly: SAT catalogs, CFDI billing, and SPEI payments introduce rules generic connectors do not understand. Validate with accounting and operations before promising «integration ready» in a sales demo.

Governance: who owns the integration after go-live

Every integration needs an internal owner — not the contractor who built it, but someone in ops or IT mandated to review weekly logs, approve changes, and escalate incidents. Without ownership, the API rots when credentials rotate and nobody updates them. Document runbooks: what to do if stamping fails, Shopify webhooks return 500, or RPA stops at login.

• Integration inventory with criticality, owner, and last test date.
• Maintenance window agreed with business for major changes.
• Staging environment that mirrors fiscal catalogs, not just «prod copy».
• Quarterly review: is RPA still justified or is API available now?