We eat our own dog food at Chirp AI. It works!
I wanted to share an interesting experience we've had since officially launching our product earlier this year. At Chirp AI, we've been 'eating our own dog food', and it's been working quite well.
Okay, so we’re not literally eating dog food, because as an AI and tech company, we obviously don't sell dog food—we sell agents as a service. So what do I actually mean by saying that then?
“Eat your own dog food” (sometimes shortened to dogfooding)typically refers to when a company uses its own products or services internally, either before releasing them to customers or continually as if they were customers themselves. The idea behind this practice is that it helps improve quality, identifies potential issues before release or before customers encounter them, demonstrates confidence in our own product, and importantly, allows us to better understand and empathize with customer feedback and challenges.
At Chirp AI, here's how we dogfood our own product:
- We deploy our agents (the exact same ones customers would use) as directly callable demos on our website. This helps us collect real usage data and feedback about our agents' quantitative and qualitative performance in conversations with potential prospects.
- We sometimes even have our agent perform outbound cold sales and outreach calls instead of our sales team. This lets us see if it can qualify a cold lead into a warm lead for future human follow-up. It's been an excellent way to gather insights on where our agent needs improvement, essentially putting ourselves in our customers' shoes. We’ve so far had the agent perform 32 cold outreach calls, and it’s successfully qualified 4 of them as potential warm leads to follow up – that’s a 12.5% strike rate, that’s not bad at all for a curated cold call list achieved by our own sales agent when we know that cold calling typically has a very low lead qualification rate – less than 10% is what the internet generally reports. The top line result is definitely interesting and certainly merits a future article to unpack the results and what else we’re observing in the AI agent cold calls —subscribe to my Medium blog to watch out for that on
- The agents we use for outbound sales calls and website demos run on the exact same software, data transformation and processing, AI models, and infrastructure as what our customers would leverage —it's completely identical and exactly the same thing as what our customers pay for.
- We use our Agent Chirp call analytics dashboard just like any other customer would.
- When an agent finishes a call, all call details and data are processed in near-real-time and made available to customers via an internal web portal. This portal provides insights, analytics, and statistics about each call, including full call transcription and summaries available immediately after the call ends.
We use this dashboard religiously, just as we'd imagine our customers might. Every time our agent finishes a demo call or attempts to qualify a cold lead, we eagerly jump into the dashboard. We check if our agent converted the cold lead, review performance metrics, fine-tune conversational performance, and enhance non-functional aspects like response speed.
So how have we benefited the most from dogfooding?
- It's helped us identify additional insights or metrics customers might want in our agent analytics dashboard.
- It's helped us uncover minor bugs under specific scenario workflows.
- It's shown us what the agent lacks in certain conversational scenarios, enabling it to be more successful in hitting a particular KPI.
- It's enhanced our empathy for customers—critical for building a successful company.
- It's allowed us to better anticipate internal product or service FAQs that customers and leads might ask.
- It's highlighted areas to focus on improving non-functional aspects of our agent.
- It’s given us valuable insights into where the agent succeeds and struggles during cold lead qualification conversations, helping us refine its configuration and persona further.
As an AI and tech company, I strongly believe one of the best ways to improve your product and better understand and empathize with your customers (the most important entity at Chirp AI) is to dogfood your own product—even if it's just small parts of it. Anything is better than nothing. Dogfooding has helped us identify issues or enhancements that might otherwise go unnoticed during typical controlled testing, enabling us to evolve our product alongside genuine customer pain points and feedback, which we can relate to much better through this kind of practice.
