About three months after launching Growbro, we had a customer write us a frustrated email that I still think about:
"I deployed your bot two weeks ago. My team says it's working. But I have no idea if it's actually capturing leads or just having meaningless conversations. I'm basically paying for something I can't see."
He was right. At that point, our analytics were embarrassingly basic — a total message counter and a list of captured leads. Nothing in between. No way to understand how the AI was performing, when customers were engaging, or which of your multiple agents was actually pulling its weight.
We spent the next six weeks building the analytics dashboard. Here's what we learned matters.
The engagement funnel: where are people dropping off?
The most revealing metric is what we call the engagement funnel. It tracks three stages:
Conversations Started → Engaged → Lead Captured
- Conversations Started: Someone sent at least one message to your AI agent (on any channel).
- Engaged: The conversation went beyond a single question — the user asked at least 2 follow-up messages, indicating genuine interest.
- Lead Captured: The AI successfully collected at least one piece of contact information (email, phone, or name).
When we built this, we assumed most businesses would show a gradual decline across the funnel. What we actually saw was more binary: either the drop-off happened immediately (people sent one message and bounced) or they went all the way to lead capture. Very few people "engaged" without converting.
This told us something important: the opening message matters more than anything else. If your AI's welcome message is generic ("Hi, how can I help you?"), you get higher bounce rates than if it's specific to your business ("Hey! I can help you find the right plan for your team size. How many people are on your team?").
We now surface this funnel prominently so businesses can experiment with their welcome messages and see the impact within days.
Activity heatmaps: when are your customers actually reaching out?
This was a feature request from an e-commerce customer in India who noticed that most of their sales happened between 10 PM and midnight — after their human team had gone home.
The activity heatmap is a grid showing message volume by day-of-week and hour-of-day, aggregated over the last 30 days. Each cell is color-coded from light (low activity) to dark (high activity).
What businesses learn from this:
- Timing campaigns. If your heatmap shows peak activity on Sundays at 8 PM, that's when you should schedule your WhatsApp broadcasts — not Tuesday at 10 AM when your marketing team happens to be at their desks.
- Justifying AI investment. When a founder sees that 40% of their customer conversations happen between 10 PM and 8 AM, the ROI argument for 24/7 AI support basically writes itself. No human team covers those hours affordably.
- Staffing human agents. For businesses that use the Pause AI (human takeover) feature, the heatmap helps them staff humans during peak hours and let the AI handle off-peak periods autonomously.
The AI leaderboard: comparing agent performance
Many Growbro customers run multiple AI agents — typically one for sales and one for support, or one per product line, or one per geographic market. The question they always ask is: "Which one is performing better?"
The AI leaderboard provides a side-by-side comparison across:
- Total conversations the agent handled
- Leads captured per agent
- Average messages per conversation (a proxy for conversation depth)
- Human takeover rate (how often the AI had to be paused for a human — lower is better)
This enables genuine A/B testing of agent configurations. You can run two agents with different personalities (one concise and professional, one warm and conversational), split traffic between them, and see which one captures more leads after a week.
One of our customers discovered that their "Sales Pro" agent was capturing 3x more leads than their "Support Hero" agent — not because of the prompt, but because the Sales Pro had more aggressive lead qualification questions configured. They updated the Support Hero, and both agents now perform comparably.
Conversation-level metrics we track
Beyond the high-level dashboards, the analytics page shows operational metrics that matter for day-to-day management:
- Total messages: Raw volume across all channels.
- User vs. Bot messages: The ratio tells you how "conversational" interactions are. A high bot-to-user ratio means the AI is doing most of the talking (potentially too verbose). A near 1:1 ratio means natural back-and-forth.
- Unread conversations: How many threads have unresponded human messages — critical for teams using human takeover.
- New vs. Returning users: Distribution showing whether you're attracting new traffic or servicing repeat customers.
Booking analytics
For businesses using the Google Calendar integration, we track bookings separately:
- Total bookings created by the AI
- Status breakdown: Pending, Confirmed, Cancelled
- Booking type distribution: Consultation, demo, service appointment, etc.
This gives service-based businesses (salons, law firms, agencies, coaching practices) a real-time pulse on their appointment pipeline without checking Google Calendar manually.
The part we're still improving
I'll be transparent: our analytics aren't perfect yet. We don't yet offer custom date range comparisons (e.g., "compare this month vs. last month"). We don't have exportable PDF reports. And our funnel metrics don't yet segment by channel (so you can't see "WhatsApp funnel vs. Instagram funnel" independently).
These are on our roadmap. But what we have today gives businesses something they genuinely lacked when using generic AI tools: visibility into whether their AI investment is actually working.
If you're running AI agents without proper analytics, you're operating on faith. Check out what data you can unlock at growbro.ai.
