From a QA standpoint, what I value most about Loman is how predictable its behavior is. When I run the same call flow multiple times, the AI responds consistently, which makes regression testing genuinely feasible. A lot of voice AI products are frustrating to test because the output drifts from run to run, but that hasn’t been my experience here. Loman also handles edge cases better than I expected, including interruptions mid-order, customers changing their mind, unusual menu combinations, and background noise on the caller’s end. It degrades gracefully rather than breaking the flow. There are still occasional misheard items, especially with uncommon names or heavy accents, but overall the pass rate on our test suite remains high. Review collected by and hosted on G2.com.
The main issue I keep running into during testing is the occasional mishearing—most often with uncommon item names, heavily accented speech, or when the caller speaks quickly while stacking modifiers. It isn’t a frequent failure overall, but it’s the category of bug I end up logging the most. I’d also like quicker feedback when I’m stepping through test call flows; response latency is generally fine, but it can spike a bit under load, and that shows up clearly in our stress tests. Both feel solvable, and the team has been receptive, but these are the rough edges that are worth calling out. Review collected by and hosted on G2.com.
