The assessment was done by people who will not integrate it
Software selection is usually run by the department that will use the product, and integration capability is not something they can reasonably assess. The result is a product chosen on features that then costs several times the expected amount to connect, or cannot be connected in the way the business needs at all. Include someone who understands integration in the evaluation, give them the shortlist and an hour with each vendor, and treat their findings as a scoring criterion rather than a footnote.
Salesperson assurances were never written down
A verbal yes during a demo is worth very little six months later when the person has moved on and the capability turns out to be on a roadmap. Anything you are relying on should be in the contract or at minimum in an email that forms part of it: the objects covered, the tier that includes access, the notice period for changes, the availability of a test environment, and the terms of a data export. Vendors who are confident will agree readily, and reluctance to put an assurance in writing is itself informative.
The pilot only tested the user interface
A proof of concept that exercises the screens tells you whether people like the product and nothing about whether it will integrate. Where the integration is important, make a small technical trial part of the evaluation: connect to the test environment, read and write the objects you actually need, and see how the documentation, the errors and the support hold up. A day of this frequently eliminates a shortlisted product, and always for a better reason than any demo would have given you.
Nobody asked where the data is stored
For Australian businesses handling personal information, knowing which country a system stores data in, and which other providers it passes information to, is part of meeting obligations under the Privacy Act 1988 and the Australian Privacy Principles. Overseas storage is not prohibited and it does carry specific requirements. Ask early, get the answer in writing, and ask the same question about any sub-processor the vendor relies on, because that detail is rarely volunteered and is difficult to establish afterwards.
Exit was never discussed
The natural time to negotiate data portability is before you sign, when the vendor wants your business, and the natural time to need it is years later when the relationship is ending. Ask what a complete export contains, whether it includes attachments and historical records, in what format, how long it takes and what it costs. Get it into the contract. Without that, an unsuitable product becomes one you cannot leave without an expensive extraction project, which materially weakens your position at every renewal.
The cheapest product had the weakest interface
There is a consistent pattern where the lowest licence cost pairs with the most limited integration capability, and the difference is recovered several times over in integration work, manual workarounds and the processes that never get automated. Compare on total cost including the expected integration effort rather than on licence price. A product costing more per year with a well documented, generously limited interface is frequently the cheaper choice within eighteen months, and demonstrably so if you cost the alternative honestly.