What has the experience been like building on Plaid directly, including costs and data quirks?

Someone with fintech background (previously paying Plaid bills of roughly $100k/month at a startup with far more connections) asked about the experience of using Plaid directly. The presenter said Plaid requires going through an audit process to get set up, but it wasn't too bad once through it. A notable quirk is that Plaid reverses transaction signs — deposits come in negative and credits come in positive — which required deciding on a 'data contract' between their database and the Plaid connection. Their approach: clean and normalize data before it hits their database (map Plaid categories to custom categories, reverse signs) so transformation only happens once — following the 'DRY' (don't repeat yourself) principle. They mostly use balances as the core data package and described their use case as very simple relative to the broader Plaid universe, focused just on banking transaction information. Banks sometimes send messy data, but it's manageable to clean.