Verified purchaser
Not quite ready
I had full stacked on this hoping it would be the system that I wanted it to be, but ultimately had to refund it. Bearing in mind, I didn't get it to make outbound calls, but to handle incoming calls. Setup was fairly easy and getting through the account process to setup a number was straight forward.
The problem comes to the AI answering the calls. I tested it several times. I had put a comprehensive prompt in, the AI doesn't do a great job at dealing with the calls in regards to interpreting the information and responding.
It initially answers the calls fine, but it can't quite understand the person on the other end, it doesn't recognise clearly the persons name and keeps on repeating back the wrong name. You also have to keep repeating back your message and other details, then there are times with long pauses and a lot of the time the call will just stop. The whole call process is very frustrating and feels very disjointed. If this were to be used daily I could see it causing reputational damage to your business.
That being said, there is potential and it could be a good system but it's not yet ready to be put into a real world situation. If they can work out the bugs and allow BYOK, it could be a nice system to use.
All the best for the future.
Avi_Dialora
Dec 19, 2025Hi, thank you for taking the time to write such a detailed and fair review. I genuinely appreciate the way you laid this out.
What you’re describing is a pattern we’ve seen when a single, comprehensive prompt is used for inbound calls. While it feels logical to give the AI “everything it should know,” voice agents behave very differently from chat agents. Overly long or dense prompts often lead to exactly the issues you experienced:
• Difficulty correctly capturing names
• Repetition or misinterpretation of details
• Longer pauses while the model resolves competing instructions
• Calls feeling disjointed or stopping unexpectedly
In real-world deployments, the best-performing agents use short, structured prompts with explicit turn-taking and name-confirmation logic, rather than one large instruction set. When tuned correctly, those issues drop dramatically—even for inbound use cases.
That said, your concern about reputational risk is completely valid. No business should deploy something that feels unreliable, and I respect your decision to refund rather than force it.
You’re also right that there is still work to be done. We are actively improving:
• Speech interruption handling
• Name capture accuracy
• Latency consistency
• BYOK flexibility is already there
If you ever decide to revisit this in the future, I’d be more than happy to personally help you configure an inbound agent the right way so you can judge it based on a properly tuned setup, not a frustrating first experience.
Thank you again for giving Dialora a serious try and for the constructive feedback. We’re taking it seriously.