The Integration Tax: Counting the Hours Behind Three Connections

Three tool integrations were supposed to save 14 hours a month and cost 9 hours a month to maintain, according to a three-month log. The net gain was real and a third of the size of the business case that approved them.

The integrations connected a task tracker to a chat tool, a document tool to a file store, and a billing system to a reporting sheet. A 40-person company used them to avoid manual copying.

The business case

Each connection was justified by a manual task it replaced. Copying task updates into a channel took about 20 minutes a day across the team. Copying final documents to a client folder took 15 minutes a day. Copying invoice totals into a sheet took 30 minutes a week.

Those estimates came from asking people how long the copying took, which is a method with a known bias toward round numbers.

How the hours were counted

For three months, every hour spent on an integration was logged under four headings: initial setup, fixing a broken connection, changing a mapping, and answering a question about why something did not sync. The log was kept by two people, so a third of the entries were reconstructed afterwards.

The setup cost was excluded from the monthly figure, which is generous and made the comparison cleaner.

What the log showed

Connection one, tasks to chat, cost 1.5 hours a month, mostly answering questions about why a message did not appear. Connection two, documents to file store, cost 0.5 hours a month. Connection three, billing to sheet, cost 7 hours a month, because the mapping broke whenever a plan changed.

The billing mapping had been changed eleven times in three months. Each change took about 40 minutes including verification.

Why the billing connection failed

The billing system described plans as identifiers that changed when a plan was renamed for marketing reasons. The reporting sheet keyed on plan names. Every rename broke the mapping, and marketing renamed a plan roughly once a month.

Nobody had considered that a marketing decision could break a finance report. That is the ordinary shape of integration failure.

What was done about it

The mapping moved to identifiers, which required the reporting sheet to carry an identifier column and a lookup table. That took six hours once and removed most of the recurring cost.

Monthly maintenance fell to 2 hours, and 1.5 of those are the recurring questions about the chat connection.

What the integrations still do not cover

Exception handling remains manual. When a task is deleted rather than completed, the chat connection posts nothing, and the channel quietly disagrees with the tracker. Two people know this and check weekly; nobody has automated it.

Exceptions are where integration projects quietly fail, because the happy path is easy to demo and the exception is not.

The question to ask before connecting two tools

How often does either side change the thing being mapped. If the answer is monthly or faster, the connection will cost more than the copying it replaces. If the answer is yearly, the connection is probably worth it.

What the log could not see

Trust is not logged. Once the billing sync had broken twice, two people stopped reading the reporting sheet without checking the source, which added a verification step that no heading covered. The verification took about ten minutes a week and was invisible because it happened inside another task.

Indirect costs of a broken integration are the ones that decide whether people actually use it.

The integration nobody asked for

A fourth connection was requested during the quarter: a calendar to a status sheet. It was refused, not on cost but on the grounds that the team had just spent three months learning that a mapping breaks when either side changes a name. Calendars change names constantly.

The refusal was the most useful decision of the exercise, and it cost nothing to make.

Three connections were expected to save 14 hours a month and cost 9, before the identifier fix. They now save 14 and cost 2, and the difference is entirely a decision about which field to trust.