Choose a native connector when your process matches what it was built for, an integration platform when you need to connect several systems and have someone to maintain the flows, and a custom API integration when your rules are specific to your business or the cost of a failure is high. Most businesses use more than one. The mistake is picking an approach for the whole company instead of for each data flow.
This guide explains the three options in practical terms, what each costs you beyond the subscription, and a framework for choosing per integration.
The three options in plain terms
Native connectors
A connector built by one of the vendors, or a partner, that links two specific products. You install it, fill in settings, and it runs. Examples include a CRM's built-in email integration or an ERP's Shopify connector.
Middleware and integration platforms
A platform that sits between systems and runs flows you configure: Zapier, Make, and n8n at the lighter end, and enterprise iPaaS products such as Boomi or Celigo at the heavier end. They provide prebuilt connections to many apps plus a visual or low-code way to map data between them.
Custom API integration
Code written for your business that calls each system's API directly. It can run as a small service, inside an existing application, or as scheduled jobs. You own the logic, the error handling, and the maintenance.
What each option costs beyond the price tag
| Native connector | Integration platform | Custom integration | |
|---|---|---|---|
| Time to first result | Fastest | Fast for simple flows | Slowest to start |
| Fit to unusual rules | Limited to its settings | Good, until logic gets complex | Whatever you need |
| Visibility into failures | Varies widely by vendor | Run history and alerts, if configured | As good as you build it |
| Ongoing cost driver | Subscription and workarounds | Usage pricing and flow maintenance | Engineering time |
| Main risk | Vendor changes or drops features | Flows nobody understands a year later | No owner after launch |
The hidden cost in every column is manual work. A connector that handles 95% of orders still leaves the other 5% for a person to fix by hand. Over a year, that can cost more than the difference between options.
A framework: decide per data flow
List each flow you need, for example "Shopify orders to ERP" or "new CRM deal to project tool," and score it on these questions.
- How standard is the process? If a thousand other companies do it the same way, a connector probably fits.
- What happens if a record is lost or duplicated? A missed marketing tag is minor. A missed invoice or a duplicated shipment is not.
- How complex are the rules? Count the conditions and transformations. A flow with many branches is hard to maintain in a visual builder.
- How much volume? Usage-priced platforms become expensive at high volume. Custom code has a fixed cost that does not rise per record.
- Who will maintain it? A connector needs an admin. A platform needs someone who understands the flows. Custom code needs an engineer.
- How often will it change? Processes that change monthly favor tools your team can edit. Stable, critical flows favor code with tests.
A simple rule of thumb: standard, low-risk, low-volume flows go to connectors or platforms. Unusual, high-risk, or high-volume flows go to custom code. Everything in between depends on who will own it.
If you are trying to decide this for a set of systems right now, write down each flow and what breaks if it fails. We can go through that list with you on a Systems Discovery Call.
Reliability requirements that apply to every option
Whatever you choose, ask how it handles these. A vendor or developer who cannot answer clearly is a warning sign.
- Duplicates. Many systems deliver events at least once, which means sometimes twice. Shopify's webhook guidance recommends using a delivery ID to ignore duplicates. The integration must recognize a repeat and skip it.
- Retries. When an API call fails, it should retry with a delay, and retries must not create a second record. Payment providers such as Stripe support idempotency keys for exactly this reason.
- Reconciliation. A scheduled check compares both systems and reports differences, because events can be missed.
- Alerts. Failures reach a named person, not only a log.
- Replay. After a fix, you can reprocess failed records without re-entering them by hand.
We cover these failure modes in more detail in Why automations fail silently and create duplicate records.
A common path that works
Many businesses start with connectors and a platform, then move the flows that hurt most into custom code. That is a reasonable path, as long as the move is deliberate. Signs a flow is ready to move:
- It has more branches and workarounds than steps.
- Someone checks its output by hand every day.
- Usage pricing for that flow has become a noticeable line item.
- A failure in it has caused a customer-facing problem.
For Maia, a client of ours, the answer was middleware that connects five ecommerce storefronts into a single source of truth for orders, inventory, and reporting. Five stores with their own connectors would have meant five places for data to disagree. You can see this and other work on our case studies page.
Questions to ask before you commit
- Where can I see every failed record, and why it failed?
- What happens if the other system is down for an hour?
- How are duplicates prevented?
- How do we change a mapping, and who can do it?
- If we leave this tool or vendor, what do we take with us?
Common questions
Is custom integration always more expensive?
Upfront, usually yes. Over time it depends on volume, platform pricing, and how much manual correction the other options require. Compare total cost over two or three years, including staff time.
Can we mix approaches?
Yes, and most businesses should. The important part is documenting which tool owns which flow, so nobody builds a second integration for data that already moves.
Next step
If you have a list of systems that do not talk to each other, or integrations nobody trusts, book a Systems Discovery Call. It is 45 minutes on your tools, workflows, and bottlenecks. Bring a list of the systems involved and the flows that cause the most manual work. You leave with a clear picture of what to build, whether you hire us or not.
Farooq Ch
Oct 2026