Subscribe
Sign in
Canonical AI
Analytics for Your Voice AI Agent
Start new thread
Tom Shapland

Canonical AI - Analytics for your voice AI agent

by
We help Voice AI developers improve their agents. We map caller journeys. We show you where and why callers are dropping off. We provide audio metrics (i.e., latency) and conversational metrics (i.e., query for cases where callers ask for a representative).

Add a comment

Replies

Best
Tom Shapland
Maker
📌
Hi PH community! I’m Tom Shapland, a cofounder of Canonical AI. We love working with Voice AI developers to help them improve their agents! Most Voice AI agent developers are manually listening to a subset of their calls. Or they’re only finding out about issues with their agent when their customers complain. We thought there had to be a better way to analyze Voice AI calls at scale – so we built Canonical AI! The Cambrian Explosion of Voice AIs LLMs have changed the paradigm for Voice AI. Compared to rule-based systems (i.e., Amazon Polly), LLM-based Voice AI agents understand the intent of the caller and can more often resolve the issue without escalation to a human agent. Moreover, with LLM-based Voice AI agents, developers can build a Voice AI agent more quickly, onboard customers quicker, and iterate on the product faster. Our customers’ Voice AI agents are doing amazing things! It’s so much fun to see the agents achieve the caller’s objective, even in the face of skepticism from the human caller. But the world of LLM-based Voice AI agents is still nascent. We’re seeing tooling emerge for testing Voice AI agents before they’re deployed. We’ve building the analytics platform for understanding, improving, and reporting on performance after they’re deployed. Key Features
  • - Call Journey Maps: Identify where calls are failing in the caller journey so that fewer callers drop off.
  • - Sad Path Analysis: Find the calls that have taken an unexpected direction so you can redesign the conversation or try other technical strategies.
  • - Audio and Conversational Metrics: Easily surface problematic calls with audio metrics (e.g., calls with a high percentage of silence) and customer conversational metrics (e.g., calls where the caller has never heard of the Voice AI agent's company).
Next Step Check out our demo! Analyze calls from a car dealership Voice AI agent in our demo! Try out the platform with your own calls! Get an API key on our website here! If you're not comfortable with uploading calls programmatically (i.e., via our python example, our Vapi integration, or our Pipecat integration, then send us a drive folder of your audio recordings. Chat with us in the comments! We’d love to hear what you think! Please ask questions or comment below!
Contact Umair
@tom_shapland Mapping caller journeys and identifying drop off points is a game changer for improving agent performance. Well done.
Jeff Chen
@tom_shapland Hi Tom! We use a provider underneath the hood (like a twilio equivalent for voice) but would love to do this in our application layer. We usually have customer examples along with our ai examples. Does your product let us benchmark these against each other?
Tom Shapland
Maker
@thisisjeffchen yes! Our platform would help your customers see how their calls are supposed to be going compared to how they're actually going in the wild. It's an epiphany for Voice AI users when they first see what's happening in their calls.
Kwindla Kramer
If you have conversational voice AI features in production, you have pain points that Canonical AI can help you fix. I've been bouncing ideas around with Tom and Adrian since the beginning [*] of the current era of voice AI. They've built interesting, valuable tools and have worked closely with many of the early pioneers deploying voice+LLM applications. It's definitely worth looking closely at what Canonical AI can do today, and worth following the company as they continue to build. [*] sometime last year. :-)
Tom Shapland
Maker
@kwindla thank you for your kind words! We've learned a lot about Voice AI from you and your team. It's been fun being a part of the community with you!
Maxim Makatchev
Super useful part of the layer of AI accountability. I often here from the customers "but AI makes mistakes..." A tool like this could be a great way to show a "canonical" certification to objectively address such a concern.
Adrian Cowham
Maker
@maxipesfix Thank you! 🙏
Tom Shapland
Maker
@maxipesfix Love what you did there with 'canonical' :). Thank you, Max! Looking forward to seeing what you do with the product.
Chris W
Cool idea. Are you mostly tailored to customer service or do you have customers who use Voice AI for other purposes? We're an AI language-learning app, and need some way of working out exactly where users are dropping off in conversations.
Tom Shapland
Maker
@cwbuilds1 We're early in our journey with this product. At this point, all of our customers have Voice AI agents that have a clear objective (For example, the Voice AI is an after hours agent for a HVAC business). We'd love to work with more open-ended Voice AIs. Would you tell me more about what you're building?
Chris W
@tom_shapland That makes sense. We probably need to define some clearer objectives for users in our app anyway. Yeah, it's a way of practicing speaking foreign languages with AI. We don't use an end-to-end voice product, but we've cobbled together TTS, STT and an LLM. Would we be able to implement Canonical with our TTS, STT and LLM system?
Tom Shapland
Maker
@cwbuilds1 Neat! What a great use case for Voice AI! Yes, users just send us a link to an audio recording. We take the rest from there. It doesn't matter how the voice orchestration was done.
Chris W
@tom_shapland Awesome - will give it a go. Thanks and good luck.
Jordan Dearsley
As the founder of a voice API infra company, I know how hard it is for our customers to deploy, debug, and test their voice agents. This is definitely a real problem. Tom & Adrian are an awesome team, and I'm looking forward to seeing how this product develops!
Tom Shapland
Maker
@jordan_dearsley Thank you, Jordan! We love your product and your team!
Varun Jain
Can this be used to provide our users of the voice agent with post call analytics about the conversation? e.g. filler words, time speaking, etc
Tom Shapland
Maker
@varun_jain12 Yes, we provide audio analytics as well as conversational analytics. Users can see time speaking, percent of the call that's silent, interruptions, etc.
Contact Umair
Congratulations on launching this much needed tool for Voice AI developers @tom_shapland. This tool looks like a great asset for any team looking to refine their conversational AI. Great job keep it up.
Haseeb Awan
This sounds interesting. We're testing voicebots and QA and analytics is a big pain point. I'll test it out
Adrian Cowham
Maker
@haseeb Thank you! We're looking forward to seeing your calls come through. Let us know how we can help, don't be shy about reaching out.
Tom Shapland
Maker
@haseeb Thank you, Haseeb! Looking forward to hearing what you think!
Rood Judeley Joseph
Amazing tool, everyone that is building AI voice solutions should definitely be using that
Adrian Cowham
Maker
Tom Shapland
Maker
@rood_judeley_joseph thank you! Excited to see what you're cooking up in Voice AI!
Jannis Moore
Absolutely love the idea of integrated AI Voice analytics! I've been using Langfuse for a while, so it’s awesome to see fresh innovation in voice AI. Keep up the great work!
Tom Shapland
Maker
Thank you, Jannis! I admire your work in Voice AI! Nice seeing you here!
Eric Landau
Congrats on the launch! Looks like a very easy way to get started with Voice AI. Very cool!
Tom Shapland
Maker
Thanks, Eric!
Adekemi Faronbi
Sad Path Analysis sounds like a fantastic feature! Understanding the unexpected directions calls take can provide key insights for improvement.
Adrian Cowham
Maker
@adekemi_faronbi We hope so! We're trying to help all you who are build Voice AI. Thank you!
Tom Shapland
Maker
@adekemi_faronbi thank you! We're thinking of renaming the company to 'Happy Path'
Rachel Bradley
Awesome idea, would appreciate a better understanding of how you built it
Adrian Cowham
Maker
@rachel_bradley2 Happy to share, let's connect and have a chat.
Tom Shapland
Maker
@rachel_bradley2 Adrian is the right guy for that. I'm just the guy that does the launches :).
Huzaifa Shoukat
Huge congrats to the Canonical AI team on today's launch! I love how you're shedding light on the blind spots of voice AI interactions. Here's a curious question: Can your platform also provide insights on how to improve the success rate of resolving caller queries without escalating to a human rep, or is that a future roadmap item?"
Adrian Cowham
Maker
@ihuzaifashoukat Yes! For example, we can user points of frustration, like having to repeat a phone number. Would love to chat in depth about this with you!
Chuck Fields
This is a super cool product. We need to have more insight into what AI is doing behind the scenes. Having this sort of high level analytics is crucial for any business with voice AI agents.
Tom Shapland
Maker
@chuck_fields Thank you for your support and encouragement!
Lucas Campa
A great idea well executed - highly usefulf or voice AI teams like us!
Tom Shapland
Maker
@lucaswcampa thank you, Lucas! Love what you're building at Toby!
Luke Harries
Congrats on the launch! What’s been the most surprising learning your customer has had with your analytics?
Tom Shapland
Maker
Thank you, Luke! We're excited to work with developers using 11labs new Voice AI agents. Some of the interesting problems include... - The AI just stops responding. In one case, the root cause is with a third-party provider's API. The developer built in a backup provider to handle the error case. - The caller hangs up before giving the AI a chance to solve the issue. When developers delay in the conversation design the fact that it's an AI, success rates go up. - The AI calling from the wrong list. The caller's kept saying on one of the sad paths, "I've never heard of your company." Lead qualification is a fast growing use case of Voice AI. But you can only call back leads who have consented (i.e., through an online form) to be called. In this case, the Voice AI developer went to their customer and gave them the insight that they were not curating their lists correctly. A win for the voice ai dev!
nina k
Great team. Tom is a leader in Voice AI.
Tom Shapland
Maker
@ninacali4 You flatter! That means a lot to me coming from you! Daily paved the way for the rest of us in Voice AI.
Richard Song
Congratulations on the launch, Tom! Go YC alum!
Tom Shapland
Maker
@renchu_song thank you!
Alan Wells
This is awesome! The drivers of AI agent performance are often so opaque when running in production at scale. I think Canonical will be one of those things where, looking back, it will be hard to imagine that we deployed voice agents without the types of insights that Canonical provides.
Tom Shapland
Maker
@alanwells thank you! We share that same vision!