How to choose an iPaaS in 2026: ten questions to ask

An integration platform is a long-term choice. Once hundreds of flows run on it, changing is expensive. At the same time, the market is moving faster than it has for years, with AI assistants, agents and MCP arriving in every product. Here are ten questions that help separate what matters from what is marketing.

The foundation

1. Does it connect to the systems you actually have? Count connectors for your own systems, not the total. Check how the connectors for your most important systems (ERP, CRM, HR, finance) are maintained, and whether they cover the operations you need.

2. Where does it run, and where is the data processed? Cloud only, or also runtimes in your own data centre or network? Which regions? For Norwegian organisations, data residency within the EU/EEA and the ability to reach on-premises systems safely are often decisive.

3. How are errors handled and monitored? Ask to see what happens when a flow fails at 03:00. Retry, dead-letter handling, alerting and tracing across systems (for example OpenTelemetry) matter more in daily operations than most features.

4. What about API management and events? Most organisations need APIs and event streaming as well as integration flows. Is it one platform or several products that must be bought and run separately?

The AI layer

5. What can the AI assistant do, and what needs approval? All vendors now have an assistant. The difference is whether it only suggests, or also changes live integrations. Look for staged changes and an audit trail.

6. How does it handle AI agents? Can agents be built, registered and governed on the platform? Can it govern agents built elsewhere? Ask what is generally available today.

7. Does it support MCP? Can your integrations and APIs be exposed as MCP tools, and are MCP servers governed like APIs, with access control, versioning and logging? Gartner points to MCP as a way to avoid lock-in in AI integration.

8. How is AI cost controlled? Language models cost money per call. Can you set limits, route to cheaper models, and see what each agent costs?

Commercials and people

9. How is it priced as you grow? Per connection, per flow, per message, per task or per user? Model three years of growth and ask for a price at that volume. AI features are often priced separately.

10. Who will build and run it? The best platform is the one your team, or your partner, can actually operate. Look at the local partner ecosystem, training, and how easy it is to recruit people with the skills.

How to use the questions

Start with a short list of three or four platforms, based on the systems you have and your cloud strategy. Analyst reports such as Gartner’s Magic Quadrant for iPaaS can help build the list, but should not decide it. Then run a proof of concept of one to two weeks with a real case: one flow that touches your most important system, including error handling. Ask every vendor to answer the same ten questions in writing. The differences will be clearer than any demo.