Technology12 October 2026 · 5 min read

When two of your systems should be talking to each other

Most businesses do not have a data problem. They have a translation problem: the till knows what sold, the website knows what was ordered, the accounts package knows what was invoiced, and a person is the only thing joining them together.

That person is usually doing it on a Monday morning, from memory and a spreadsheet, and they are the single point of failure nobody has written down.

How to spot the ones worth joining

Somebody types the same thing twice. The clearest signal there is. Every double entry is a weekly cost and an opportunity for the two systems to disagree.
Two reports should match and do not. If reconciling takes a morning, the reconciliation is the integration, done by hand.
A decision waits for a person to be in. If nobody can say whether something is in stock, paid for or booked without asking a specific colleague, the information is in a system nothing else can see.

What an integration actually is

Two systems agreeing on a shared fact, continuously, without anybody retyping it. In practice that is usually an API on one side and a scheduled job or a webhook on the other, with a record of what moved and what failed.

The unglamorous part is where the work is: what happens when one side is down, what happens when the same record arrives twice, and what happens when the two disagree. A connection that cannot answer those three questions will quietly corrupt both systems rather than one.

When not to integrate

When the job takes ten minutes a month. Automating it will cost more than it saves, forever.
When the process is about to change. Wiring together a workflow you are replacing in the spring is paying twice.
When one side has no proper way in. If the only route is screen-scraping somebody’s portal, that is a connection that breaks every time they redesign a page. Sometimes it is still the right call; it should be a decision, not a surprise.

The question to ask a supplier

Not "does it integrate". Everything claims to. Ask what happens to your data if you leave, whether there is an API you can use without a partner agreement, and whether the connection costs per site per month forever. The answers tell you whether you are buying a system or renting access to your own numbers.

The detail of how we join them is on the integrations and data page, and the workflow side of it, the chasing and the reminders nobody should be doing by hand, is on process automation.

Questions

Before you ask.

How do I know which systems should talk to each other?

Follow the double entry. Wherever somebody types the same thing into two places, or where two reports should match and do not, you already have an integration, done by hand and at risk of being wrong.

What makes an integration reliable?

Handling the unglamorous cases: one side being down, the same record arriving twice, and the two sides disagreeing. A connection that cannot answer those three questions corrupts both systems rather than one.

When is integrating not worth it?

When the manual job takes ten minutes a month, when the process is about to change anyway, or when the only route in is scraping somebody’s web portal, which breaks whenever they redesign it.

What should I ask a software supplier about integration?

Not whether it integrates, because everything claims to. Ask whether there is an API you can use without a partner agreement, what happens to your data if you leave, and whether the connection is charged per site per month forever.

Can systems from different suppliers work together?

Usually, yes, if each exposes a proper API. Where one does not, the question becomes whether a scheduled export is enough, and that should be a decision made up front rather than discovered later.