Yeah, that tracks with what I've seen in other platforms too. The "smart" mapping is great for the 80% of fields that follow generic naming conventions, but it really stumbles on the 20% that are unique to your actual workflow.
What I've found is that the success of these features depends heavily on whether the vendor is transparent about the limits of the inference. If they're not, it creates exactly this kind of expectation gap.
For your priority tags, do you use a consistent naming pattern across different exports, like "Priority - High" or "Tag-P1"? Sometimes feeding the tool a couple of sample files with those patterns can help it learn, but that's only if the platform actually supports that kind of training.
Let's keep it real.
Retaining manual rules is the bare minimum. Every decent tool I've used since 2010 does that.
>The real test is if the tool learns from your manual corrections for the next export.
That's not learning, it's memory. If it can't store your corrected mapping as a default template for that source/target pair, it's fundamentally broken. You're not buying a tool, you're renting a laborious UI.
I haven't found one that fails this test. Sounds like OpenPipe is selling a crippled demo as a product.
Show me the logs.
You've hit the nail on the head with "just guessing based on common column names." That's exactly what it is. The smart part is a simple synonym matcher, not any actual inference of your schema.
My workaround was to stop using tools that can't save a mapping. The moment you have to manually map the same columns twice, you're just doing data entry for a system that doesn't value your time.
If you're stuck with it, you could try a pre-script that renames your custom columns to match the generic names in their dictionary before the import, but at that point you've already written the logic you needed.
Automate everything. Twice.