Business

Webhooks, Polling or Queues: Choosing an Integration Pattern Between Business Systems

November 11, 2026 · 4 min read · Etzal Group Team

When two business systems need to share data, somebody has to decide how one learns that something changed in the other. There are three common answers. The choice of integration pattern looks technical, and its consequences are operational: whether an order can go missing, how late stock figures can be, and what happens at two in the morning when one system is down.

Pattern 1: polling

One system asks the other at intervals: "Anything new since I last asked?"

How it works. A scheduled job calls an API every few minutes, fetches records changed since a stored timestamp or sequence number, processes them and updates the marker.

Strengths.

  • Simple to build and to reason about.
  • The receiver controls the pace and cannot be overwhelmed.
  • If the receiver is down, it catches up when it returns, because it asks from where it stopped.
  • Works with systems that offer no event notifications.

Weaknesses.

  • Data is late by up to one interval.
  • Most calls return nothing, which wastes capacity and may hit rate limits.
  • Careless use of timestamps can skip records that were saved during the query.

Fits. Nightly or hourly synchronisation of master data such as products, price lists and customer records. Any case where a delay of minutes is acceptable.

Pattern 2: webhooks

The source system calls an address you provide the moment something happens.

How it works. You register a URL. When an order is paid, the source sends an HTTP request describing the event. Your system acknowledges quickly and processes it.

Strengths.

  • Close to real time.
  • No wasted calls.
  • Simple on the sending side for the event's owner.

Weaknesses.

  • If your receiver is down or slow, the event may be lost unless the sender retries, and retry policies vary.
  • Events can arrive twice, or out of order.
  • The endpoint is public, so it must verify that the call really comes from the source.

Fits. Payment confirmations, order status changes, incoming messages. Any event where minutes matter.

Make it safe. Verify the signature on each request. Store the raw event first and acknowledge immediately, then process from your own store. Record event identifiers to ignore duplicates. Run a periodic reconciliation by polling, to catch anything that never arrived.

Pattern 3: message queues

A broker sits between the systems. Senders publish events to it and receivers take them when ready.

How it works. The source writes a message to a queue or topic. The broker keeps it until a consumer confirms it has been processed. Several consumers can read the same stream for different purposes.

Strengths.

  • Sender and receiver are decoupled. A slow or offline receiver does not block the sender.
  • Messages are retained and retried.
  • Handles bursts by letting work wait in line.
  • One event can feed accounting, the warehouse and analytics separately.

Weaknesses.

  • Another piece of infrastructure to run and monitor.
  • Requires discipline about message formats and versions.
  • Still delivers duplicates in failure cases, so consumers must be safe to repeat.

Fits. High volume, several consuming systems, or processes where a lost event is unacceptable.

Four questions that decide it

  1. How fresh must the data be? Seconds points to webhooks or a queue. Minutes or hours allows polling.
  2. What happens if an event is lost? If the answer is "a customer is charged and nothing ships", you need durable storage and reconciliation, whichever pattern you use.
  3. How many systems need the same event? One receiver suits a webhook. Several suit a queue.
  4. What does the other system support? Many products offer only an API. Then polling is the pattern, and that is fine.

Rules that apply to every pattern

  • Make operations repeatable. Use unique identifiers so that processing the same event twice has no extra effect.
  • Expect out-of-order arrival. Compare versions or timestamps before overwriting newer data with older.
  • Keep a failed-items list. Events that cannot be processed after retries go to a place where a person can see and resolve them.
  • Reconcile. A daily comparison of totals between systems finds problems that no alert reported.
  • Log by event identifier. When finance asks where an order went, you can answer in a minute.
  • Agree on ownership. For each field, one system is the source of truth.

A common combination

Many reliable integrations combine the patterns: a webhook for speed, a small internal queue so that processing never blocks the acknowledgement, and a nightly poll that reconciles everything. Each covers the others' weak points.

Designing and operating integrations between ERP, point of sale, warehouse and accounting systems is a core part of our Enterprise Software work. If records are going missing between two of your systems, tell us which ones and we will suggest where to look.

Summary

Poll when delay is acceptable or events are not offered. Use webhooks for speed, with verification, storage and reconciliation. Use a queue for volume, several consumers or guaranteed delivery. In all three, make processing repeatable and compare totals every day.

Frequently asked questions

What is the difference between a webhook and an API?

An API is something you call when you want data. A webhook is a call the other system makes to you when something happens. With an API you ask. With a webhook you are told. Most integrations use both: a webhook announces an event and an API call fetches the details.

How do we avoid processing the same event twice?

Give every event a unique identifier and record the identifiers you have handled. Before acting, check whether the identifier is already recorded. This makes the operation safe to repeat, which matters because retries and duplicate deliveries are normal in every pattern.

Do we need a message queue for a small integration?

Often not. If two systems exchange a modest number of events and a short delay is acceptable, polling or webhooks with a retry table are enough. A queue becomes worthwhile when volume is high, several systems consume the same events, or a slow receiver must not hold up the sender.

Author

Etzal Group Team

Etzal Group Team, helping businesses grow with technology. Contact us.

Related service

Free consultation

Not sure where to start? We will map the right system for your business, free.