SODAX SDKDigital asset flows you can build and deploy with AI.
Promoted
Inquio
See what your bot is getting wrong
82 followers
See what your bot is getting wrong
82 followers
Most chatbot analytics tell you how many conversations happened. Bot Report Card tells you what actually went wrong. Upload your chatbot conversations and get an AI-powered audit that uncovers hidden issues, hallucinations, customer frustration, missed sales opportunities, and risky responses. Every finding is backed by real conversations and includes practical recommendations, so you know exactly what to fix first.
Inquio
Hardest part for us was that the worst answers never look like failures. Frustration and risky responses at least leave a trace someone can complain about. The plausible and slightly wrong ones don't, the person just fixes it themselves and your quality metrics stay green. If your audit can flag those I think that's the whole product, and the rest is table stakes.
Inquio
@asadmalik901 Exactly. That’s one of the biggest problems we’re trying to solve.
The dangerous conversations are often not the ones where the bot clearly fails—they’re the ones where the answer sounds right but is subtly wrong, and the customer quietly works around it.
That’s why we analyze the actual conversation in context, rather than relying only on explicit complaints or predefined failure categories.
We’re still working on making this detection as reliable as possible, so this is a great area for us to keep improving. Thanks for putting it so clearly!
@inquio The thing I'd want before trusting it is your false positive rate on a hand labelled set. Subtly wrong answers are exactly where an LLM judge is confident and still wrong, so a grader that catches them is a real claim, not a feature. Publish that number, even if it's ugly, and I'd take this a lot more seriously than any accuracy line in the copy.
Inquio
@asadmalik901 That's a fair challenge. One important distinction, though: Inquio isn't a groundedness checker, because it doesn't require a ground-truth knowledge base.
For example, if a bot says a savings account pays 5% interest, Inquio can't know from that conversation alone that the correct rate is actually 4.5%.
Instead, we look for evidence of problems across the full volume of conversations. Repeated questions, inconsistent answers, users challenging information, corrections, conversation patterns, etc. can reveal that something is going wrong. We then cluster these signals to identify the underlying issue and trace it down to the individual conversations.
So our claim isn't that we can identify every wrong answer. It is that by analyzing all conversations rather than small samples, it can reliably surface the topics and situations where a chatbot has problems and suggest what to fix.
There are some important nuances too. For example, with security-related issues, we deliberately tolerate more false positives because missing a real incident is much worse.
I'd actually encourage you to try it. And if you have a curated, hand-labelled dataset, I'd be very interested to see what Inquio finds on it. That would be a much more useful test than an accuracy claim in our marketing copy.