In most businesses, the integration layer is a person

Software vendors sell products. They do not sell the connections between products, and almost no business runs on a single vendor. What fills the gaps is people. A quote produced in one system gets retyped into an invoice in another. A name from an email becomes a row in a spreadsheet and then a record in a CRM. None of these steps is difficult, which is exactly why they are invisible in every process document.

Added together they are frequently the largest single category of administrative time in a small or mid-sized business. The work is unglamorous, distributed across many people in small increments, and almost never measured, which is why it persists for years without being addressed.

Finding the crossings

The reliable method is to trace one unit of work end to end and mark every point where it moves between systems. Take a single job, order, patient, matter, or ticket, and follow it from arrival to completion. At each step, note which system holds the record and how it arrived there.

A typical trace for a service business looks approximately like this:

1. Enquiry arrives by email or web form
2. Someone reads it and creates a record in the CRM (crossing)
3. A quote is produced in a quoting tool or a spreadsheet (crossing)
4. The quote is emailed, and the CRM is updated to reflect that (crossing)
5. On acceptance, the job is scheduled in a calendar (crossing)
6. After completion, an invoice is created from the quote (crossing)
7. The invoice is entered into the accounting system (crossing)
8. Payment status is reconciled back to the CRM (crossing)

Eight steps, seven crossings, and at every one of them a person is moving information that already exists from one place to another. The record is created once and re-entered seven times.

Calculating what a crossing costs

Each crossing has four costs, and only the first is obvious.

Direct time. Minutes per instance, multiplied by instances per month. If a crossing takes two minutes and happens two hundred times a month, that is roughly six and a half hours. Across seven crossings at similar frequency, forty-five hours a month, which is most of a working week.

Error correction. Manual transcription has a non-zero error rate, and the cost of an error is rarely the thirty seconds to fix the typo. It is the invoice sent with the wrong figure, the call to explain it, the credit note, and the reconciliation. A single-digit percentage error rate on a high-volume crossing generates significant recurring work.

Divergence. When the same fact lives in several systems and is maintained by hand, the copies drift. Eventually nobody is confident which system is authoritative, and reporting requires a reconciliation step before anyone trusts the numbers. This cost compounds quietly and is usually paid at month end.

Latency. Work that waits for a person to carry it across sits idle. A quote produced on Friday afternoon that becomes an invoice on Monday has spent three days in a queue for no reason other than the absence of a connection. For anything customer-facing, that delay has commercial consequences beyond the labour cost.

Which crossings to close first

Not all crossings are equally worth closing. Three factors decide the order.

Frequency beats duration. A thirty-second crossing occurring five hundred times a month is worth more attention than a twenty-minute one occurring twice. High-frequency crossings also tend to be the most rigidly structured, which makes them the most reliable to automate.

Structure beats volume. A crossing where the same fields move in the same shape every time is straightforward and stable. One where a person interprets unstructured input and decides where it belongs is a different and considerably harder problem. Start with the structured ones; they deliver value quickly and their failure modes are easy to instrument.

Ownership of both ends. A crossing between two systems the business controls is tractable. One where the upstream is a third party who sends inconsistent formats introduces a dependency nobody can manage. Those are still worth addressing, but they are second, not first.

What this looks like once closed

The realistic outcome is not that every crossing disappears. It is that the structured ones close entirely and the judgment-bearing ones get reduced to a confirmation step. The enquiry becomes a record automatically, the accepted quote becomes an invoice without retyping, and payment status reconciles on a schedule instead of at month end.

What remains for a person is the part that genuinely required one: deciding what to quote, handling the client who wants something unusual, and judging the exceptions. That is a substantially better use of the same salary, and it is available to most businesses without replacing a single piece of software they currently run.