★★★★★Bespoke software for B2B·One-off build fee, low monthly retainer·You own the software

Integration

System integration: getting your business software to talk to each other

System integration is the plumbing that moves data between the separate systems your business already runs - CRM, finance, stock, a portal - so nobody re-types it. Done badly it's a tangle of one-off links; done well it's a small number of predictable connections.

Haystak · 11 August 2026 · Updated 11 August 2026 · 8 min read

Diagram showing system integration through a central hub connecting several business applications

System integration is the work of connecting two or more pieces of software so that data created in one place appears, correctly and on time, in another. That might be a new order in your e-commerce site appearing in your finance system, or a customer's address update in your CRM reaching your delivery software without anyone copying and pasting.

Most businesses arrive at this problem gradually. They buy a CRM, then an accounting package, then a stock system, then a portal - each one good on its own, none of them aware the others exist. System integration is how you make the set behave like one system rather than four.

Point-to-point tangle vs a hub

The first integration is easy: connect system A to system B directly. The second is still fine. By the fifth system, you have direct links between most pairs of them, and each new system means several new connections rather than one.

That point-to-point pattern is why integration projects that started cheap end up expensive to touch. Nobody wants to change the CRM's customer record because four other links quietly depend on its exact shape.

ApproachHow it worksWhere it goes wrong
Point-to-pointEach system talks directly to each other system it needsConnections multiply; one field change breaks several links at once
Integration hub / middlewareEach system talks to one central layer, which routes and translatesCosts more to set up; needs someone to own it
Shared databaseSystems read and write the same data store directlyFragile, and most modern SaaS products won't allow it anyway

A hub - sometimes a proper integration platform, sometimes a small custom service we build for exactly this purpose - means each system only has to know how to talk to the hub. Adding a sixth system is one new connection, not five.

Every field needs a source of truth

The question that derails most integration projects is deceptively simple: when the same piece of information exists in two systems, which one is right? Customer address, product price, stock level - pick a system as the single owner for each, and make every other system a reader of that value, not an independent editor.

  • One writer per field. If both the CRM and the finance system can edit a customer's payment terms, they will eventually disagree.
  • Everyone else reads. Other systems display the value or use it, but changes flow back through the owning system.
  • Write it down. A one-page table of field, owning system, and who is allowed to change it settles most arguments before they start.

Sync vs event-driven integration

There are two broad ways to move data between systems, and the choice affects how quickly problems show up and how expensive they are to fix.

  1. 01Scheduled sync. Every hour, or overnight, a job copies changes across. Simple to build and reason about, but there's a window where the two systems disagree, which matters for stock or pricing.
  2. 02Event-driven. The source system announces a change the moment it happens (an order placed, a status updated) and the other system reacts immediately, typically via a webhook. Near real-time, but needs more care around retries and ordering.

Most businesses don't need everything in real time. Stock counts and order confirmations usually do; a marketing mailing list update usually doesn't. Matching the method to how much the delay actually costs you keeps the project proportionate.

Error handling and reconciliation

The part of system integration that gets skipped under time pressure is what happens when a transfer fails. A network blip, a system down for maintenance, a record that doesn't validate on the other end - all of these happen regularly, and a good integration expects them rather than being surprised by them.

  • Retries with backoff, so a temporary failure doesn't need a human to notice and re-run it.
  • A dead-letter queue or error log that someone actually checks, not one that silently fills up.
  • Reconciliation reports - a scheduled check that counts and compares totals between systems, catching drift before a customer does.
  • Idempotency, so replaying a failed message twice doesn't create a duplicate order or invoice.

What breaks in practice

In our experience, the same handful of things cause most of the pain after go-live, and none of them are exotic.

What breaksUsual cause
Duplicate customers or ordersNo agreed matching key between systems, or retries without idempotency
Prices or stock out of stepSync interval too slow for how fast the value actually changes
Integration silently stops workingNo monitoring - a vendor changed their API and nobody was watching
Fields quietly go blankA field mapping missed during a vendor's own product update

Most of these are cheap to prevent and expensive to unpick after the fact, which is why we build monitoring and reconciliation into integration work as standard rather than as an optional extra.

Buying vs building the integration layer

Off-the-shelf integration platforms (iPaaS tools) are a genuinely good answer when you're connecting mainstream SaaS products with well-documented, well-used APIs - they save you building plumbing that already exists. Custom integration earns its keep when a system has no usable API, when the volume or reliability requirements are unusual, or when the logic connecting the systems is itself a piece of business rules rather than a straight data copy.

A short discovery - mapping every system, every field, and who owns it - tells you which situation you're in before anyone commits to a build. If you're weighing that up, our pricing page has a sense of what different scopes of integration project typically involve, or you can talk it through directly.

✦ Where this fits

More on this from us: our system integration services.

Questions we get asked

Common questions

What is system integration in simple terms?

It's connecting separate software systems so that information created in one automatically reaches the others that need it, instead of someone re-typing it by hand.

How long does a typical integration project take?

A single well-documented connection between two mainstream systems can take days. Integrating several older systems with no APIs, or building an event-driven hub, is a matter of weeks. Mapping the systems first tells you which you're facing.

Do we need an integration platform (iPaaS), or custom code?

If every system involved has a modern, documented API and your rules are simple field mapping, an off-the-shelf platform is usually the right, cheaper answer. Custom code earns its place when a system lacks an API or the logic in between is genuinely complex.

What happens if an integration fails halfway through?

A properly built integration retries automatically, logs the failure somewhere a person will see it, and is designed so re-running it doesn't create duplicates. Reconciliation reports then catch anything that still slips through.

Can old, unsupported software be integrated at all?

Often yes, though not always through a clean API - sometimes via its database, an export file, or screen automation as a last resort. It's worth checking early, since it changes cost and reliability considerably.

Tell us which systems don't talk to each other We'll tell you what we'd build.

We'll map the connections you actually need, tell you honestly where an off-the-shelf tool will do the job, and quote the rest properly. Get in touch via our contact page.

No pitch deck · We map your process · You own the software