Why a File Export Costs More Than a Proper Interface

When two quotes for the same automation are far apart, the gap is rarely in the business logic. It is almost always in how data gets into the systems involved and back out again. And the most expensive route is the one that sounds easiest in the meeting, the offer to simply export the data as a file.

CSV looks harmless. You open it and see rows, columns and readable text, and every tool claims to support the format. That is exactly why a file handover gets agreed without anyone asking a second question. It is often the most consequential decision of the whole meeting.

A file is not a format, it is an agreement

There is no binding standard that tools have to follow. Separators, character encoding, quoting, line endings and date notation differ between systems and sometimes between versions of the same system. A documented interface states what is a number, what is text and which encoding applies. With a file, the reading system guesses.

Someone has to replace that guessing. For every field it has to be decided how it is read, what an empty value means and what happens when a value does not fit the expected pattern. That work is not hard, but it accrues per field, not per file.

Two typical pictures show how quiet the damage is. A postal code such as 01067 is treated as a number by a spreadsheet and loses its leading zero the moment someone opens and saves the file before export. An amount written as 1.234,50 in German notation is read by a system that treats the dot as a decimal separator and arrives as 1.23. In both cases a technically valid value ends up in the target system. No import fails, no log turns red.

The cost sits in the handover, not in the reading

Reading a file is built quickly. The project is everything that belongs around the handover.

An automation has to recognise whether the file is complete or still being written. It has to prevent the same export being processed twice because someone triggered it again. It has to notice when an export fails to appear at all, because a file that never arrives raises no error, it raises nothing. It has to cope with someone adding a column in the source system and shifting the order. And it needs a place where faulty rows land instead of quietly disappearing.

With an interface, several of these points never arise. That is why the way in and out of the systems is more expensive in many projects than the logic in between.

Why the error only surfaces weeks later

An automation that breaks is a good event. It reports itself, someone looks, the run is repeated. File handovers fail differently. They run through and write wrong values.

You notice it by detour. Mail comes back undeliverable, invoice checking finds discrepancies, the back office fields more queries than before. By then the wrong values sit in the target system, and correcting them costs more than the original build, because later records already depend on them.

Every file handover therefore needs a checking layer. Expected row count, mandatory fields, permitted value ranges and a reconciliation of totals against the source system. Cutting that item from a quote is saving in the place with the worst return.

What you can settle before asking for quotes

Ask the vendor of every system involved whether a documented interface exists, what it costs and whether your licence already includes it. Often one exists and was simply never activated because nobody had asked. A module unlocked once is almost always cheaper than maintaining a file route for years.

Also clarify whether a test environment exists and who in your company administers the credentials. Without a test system every project takes longer, because nobody is allowed to experiment in production.

And if it does become a file, get the structure agreed in writing. Encoding, separator, column order, date format, and a commitment that changes will be announced in advance. That commitment is the real difference between a file route that runs for two years and one that stops after the next update.

When a file is still the right answer

There are cases where the detour pays. A legacy system without an interface that will be replaced anyway. A process that runs once a month and whose result a person reviews. A data set that is migrated once.

The difference lies in frequency and consequence. A file that moves prices or stock levels every day deserves the same safeguards as an interface. A file that delivers a quarterly report does not.

The recommendation

Treat the sentence about exporting data as a file as a cost item, not as a solution. Ask the vendor about the interface first. If there is none, have the quote separate what reading the file costs from what the safeguards around it cost. Only both numbers together give you an honest comparison against an interface, and only that comparison carries the decision.

Sophera Consulting looks at which systems are involved and how they release their data before quoting, and puts file structure, validation rules and error handling in writing before naming a fixed price. The entry point is free in the automation check.