Integration counts mean nothing

How a number gets big

Count anything that sends a webhook, and a product gains twenty integrations in an afternoon.

Count every app reachable through a general automation platform, and the number reaches the hundreds without a line of code written for any of them.

Count read-only imports, one-way exports and things a partner built and abandoned, and there is no upper bound at all. None of this is fraud. It is just a number that measures nothing you care about.

Our five, and what they each do

Here is the whole list, with the part that matters: what each one needs, and what it produces.

• Slack: posts notifications, reports and scheduled announcements into your workspace. Needs a bot token.
• OpenAI: powers HashBot, the policy assistant, and the risk signals on meetings.
• Fireflies: imports meeting recordings and transcripts, filed against the right project, with action items and notes.
• ZKTeco attendance device: syncs punches from an attendance terminal, configured by device address on the local network.
• Turnstile: bot protection on sign-in, managed by the platform, with nothing for you to set up.

The question that actually matters

Not how many. Which one, and how deep.

Ask whether the integration is one-way or two-way. Ask what happens when it fails, and whether anyone is told. Ask what field-level detail comes across, because a summary sync and a record sync are different products.

Ask what the integration needs before it works. Ours need a bot token, an API key, a device on your network. Each is a small setup job somebody has to own.

Why fewer tools need fewer connectors

Much of the integration count in this category exists to join one tool to another. An HR system gets joined to a project tool, and a project tool to a capacity planner.

PeopleMuster keeps HR and project delivery in one database. Leave reaches the capacity plan without a sync. Timesheets reach the utilisation report without an export. The connector that would join them does not need to exist.

What an integration costs after it is switched on

The count also hides the ongoing part. Every live connection is a small piece of maintenance somebody inherits.

A key expires and a feature stops. A meeting arrives filed against the wrong project and stays there until someone moves it. A device drops off the network and the sync quietly falls a day behind.

None of those announce themselves. They show up as a report that looks slightly wrong, a fortnight later, and the cause takes longer to find than the fix.

So the right question about a connection is not only whether it works. It is who notices when it stops, and how.

Fewer connections, better understood

Five is not a badge, and a longer list would not be a scandal. But a short list has one property worth keeping: each one can be explained in a sentence and checked in a minute.

Each of ours needs a specific thing before it does anything. A token, a key, a device address. That setup is visible, which means a failure is traceable.

Keys are stored per tenant and masked after saving, which the product says on its own screen. Small detail, and the sort of thing a logo wall never tells you.

A page that names five and describes them honestly gives a buyer more than a page that names a hundred and describes none.

The export question, which usually matters more

For most buyers the real requirement is not a live connection at all. It is getting data out in a shape something else can read.

Attendance exports. Leave exports. Device data exports. A monthly file that lands somewhere predictable solves more problems than a connector that has to be watched.

Exports are also easier to reason about. A file has a date on it, a person who produced it, and a copy you still hold next year. A sync has none of those unless somebody built them.

So before asking whether a product connects to a system, ask what it will hand you. That answer is usually simpler and more durable.

Counting honestly, if you must count

There is a version of the count that would be worth publishing, and nobody publishes it.

Connections the vendor wrote and maintains themselves. Connections that write as well as read. Connections a customer can set up in under an hour without help.

Those three numbers would be small for almost every product in this category, and they would tell a buyer something real.

The hundred-logo version survives because it is easy to produce and nobody checks it. Which is the argument for ignoring it entirely and reading the five names instead.