Start with workflow differentiation, not feature count
The first question is whether the workflow represents a distinctive business capability or a common administrative function. Attendance, expense claims, basic approvals, and standard CRM activities are usually well served by mature SaaS products. Building them from scratch also means taking responsibility for security patches, mobile support, audit trails, accessibility, and continuing user support.
Custom development becomes more defensible when the workflow directly affects pricing, delivery speed, service quality, or how operational data becomes a decision. A service process that dynamically routes work according to customer terms, product conditions, and live equipment data is not merely a ticket form. Its value lies in the organization’s rules, exceptions, and feedback loops.
- Favor SaaS: Standard industry behavior is sufficient and a different workflow provides little competitive advantage.
- Favor custom software: The rules encode operational knowledge, customer experience, or a core service capability.
- Consider a hybrid: Use SaaS for commodity functions while keeping differentiated logic in an owned service, integration layer, or workflow engine.
Change frequency matters more than the current requirement list
A long requirement list does not automatically justify custom software, and a short list does not guarantee that SaaS will fit. Examine how often rules change, why they change, and whether those changes are predictable. When updates follow regulations or common accounting practices, a specialist vendor can often maintain them more effectively. When changes come from market experiments, customer contracts, operational learning, or evolving AI models, dependence on a vendor roadmap may slow the business.
Engineers should also separate configuration changes from code changes. If administrators can safely modify fields, forms, states, and notifications, a configurable SaaS product may be flexible enough. If each change affects transaction ordering, decision logic, exception recovery, or new data sources, an advertised API is not sufficient evidence. The required fields may be unavailable, rate limits may constrain the process, or event delivery may lack the ordering and idempotency guarantees the workflow needs.
- Infrequent and standardized: Prefer SaaS and avoid maintaining a commodity capability.
- Frequent but configurable: Evaluate workflow-enabled SaaS or a low-code platform with version control and governance.
- Frequent and domain-specific: Keep control of the domain model, business rules, tests, and release schedule.
- Change pattern still unclear: Run a prototype or limited pilot before committing to a long contract or a complete rebuild.
Include integration, governance, and exit costs
Subscription pricing is visible, but total ownership also includes implementation, data cleanup, interfaces, identity management, training, extensions, and eventual migration. Custom software has its own frequently omitted costs: monitoring, backups, security updates, cloud operations, automated testing, documentation, and knowledge transfer. SaaS transfers part of the technical burden to a vendor; custom development exchanges greater internal responsibility for greater control.
Integration is often the dividing line. If a workflow spans LINE, ERP, CRM, IoT devices, and a data warehouse, define system boundaries and data ownership before connecting tools directly to one another. Decide which component owns identity, master data, retries, audit records, and reconciliation. For enterprise assistants and RAG systems, document upload alone is not knowledge governance; source permissions, document refresh, answer traceability, and sensitive-data isolation must also be designed.
- Data portability: Can the organization export records, attachments, relationships, history, and permission metadata in usable formats?
- Integration reliability: Are stable APIs, webhooks, sandboxes, version policies, and failure diagnostics available?
- Governance: Do data location, access control, audit, and retention behavior meet operational requirements?
- Vendor dependency: If pricing, functionality, or service availability changes, can the workflow be moved within an acceptable period?
Test operating scenarios instead of watching demonstrations
Select a small set of common, complex, and high-risk scenarios, then require the SaaS trial or custom prototype to complete them end to end. Include invalid data, duplicate events, manual review, insufficient permissions, an unavailable external system, recovery, and later investigation. A polished demonstration proves that the happy path works; exception handling reveals the likely operating cost.
The decision record should cover differentiation, change frequency, configuration limits, integration complexity, governance, exit difficulty, and internal maintenance capability. Avoid hiding every trade-off inside a single weighted score. Identify non-negotiable constraints first, then compare the remaining choices. Data control may be a hard requirement, while a preferred screen layout may be safely postponed.
In practice, the strongest design is often layered rather than entirely purchased or entirely built. SaaS can handle mature commodity capabilities, an integration layer can normalize identity, data, and events, and owned services can contain the rules that change frequently or distinguish the business. Whichever model is chosen, define data ownership, interface contracts, monitoring, and an exit plan early so that fast implementation today does not become an immovable constraint later.