title: WhatsApp bot analytics: what to track and how to measure success description: >- Learn which WhatsApp bot analytics metrics matter, what good looks like, and how to connect conversation data to real business outcomes like leads and bookings. date: '2026-07-15' author: Rebecca Pearson authorAvatar: /blog/authors/rebeca-avatar.webp category: Resources cover: /blog/whatsapp-bot-analytics/blog-thumbnail-blank.png readingTime: 6 tags:
- WhatsApp bot analytics
- WhatsApp bot metrics
- WhatsApp AI agent
- WhatsApp automation sourceUrl: 'https://www.codewords.ai/blog/whatsapp-bot-analytics'
You've built a WhatsApp bot. It's live. Conversations are coming in. But how do you know if it's actually working? WhatsApp bot analytics tell you whether your agent is doing its job — and, more importantly, where to improve it. You can't improve what you can't measure, and the metrics that matter aren't always the obvious ones.
TL;DR
- Eight metrics cover almost everything — response time, completion rate, handover rate, message volume, cost per conversation, return rate, sentiment trends, and error rate.
- Connect metrics to business outcomes — a 90% completion rate is meaningless unless you know how many of those completions turned into bookings, leads, or sales.
- Your conversation data is your best training resource — use it to improve your system prompt continuously.
Why analytics matter
A WhatsApp bot that sends replies is not necessarily a WhatsApp bot that's working. It might be sending the wrong replies. It might be escalating to humans for questions it should handle itself. It might be costing three times more per conversation than it needs to. It might be missing the peak hours when customers need it most.
Analytics give you the visibility to catch these problems early and fix them before they affect customers — or before they cost you more than the bot saves.
The eight metrics that matter
1. Response time
How long does it take for your bot to reply after a message arrives? Target under three seconds for routine messages. Spikes above ten seconds indicate a performance issue — possibly a slow integration, a model timeout, or an overloaded webhook.
Track response time as a distribution, not just an average. If 95% of responses come back in two seconds but 5% take 30 seconds, your average might look fine while customers are actually having a bad experience at the tail end.
2. Completion rate
What percentage of conversations reach a successful outcome? "Successful outcome" depends on your bot's purpose: a question answered, an appointment booked, a lead captured, an order placed.
A completion rate below 60% suggests your bot is struggling — either the questions it's being asked are outside its scope, or the flow has dead ends that customers can't navigate through. Above 80% is solid. Above 90% is excellent.
3. Handover rate
How often does the bot trigger a human handover? This is a signal about the quality of your agent's knowledge base and its ability to handle edge cases.
A handover rate of 10–15% is healthy for most businesses — it means the bot handles the vast majority of conversations but appropriately escalates the genuinely complex ones. A rate above 30% suggests the bot is over-escalating. A rate below 5% could mean the bot is attempting things it shouldn't.
4. Message volume by day and hour
When are your customers most active? Volume patterns reveal peak hours, quiet periods, and day-of-week patterns that should inform your staffing decisions for human handovers.
Volume anomalies are also worth watching. A sudden spike might mean your bot has been shared somewhere (good), or that a message from a previous campaign just reached a large audience (manageable), or that someone is attempting to abuse the system (needs investigation).
5. Cost per conversation
If your underlying AI model charges by token, cost per conversation is a critical metric. A support bot that averages 20 messages per conversation and sends the full conversation history to the model on every turn is using far more tokens than a bot that uses a summarised context.
Track cost per conversation and compare it against the value each conversation creates. A bot that costs £0.05 per conversation and deflects a £15 support ticket is extremely good value. One that costs £2 per conversation and only occasionally closes a lead needs optimisation.
6. User return rate
Do customers come back? A customer who has a good experience with your WhatsApp bot and messages again — for a follow-up booking, a new order, or a different question — is the strongest signal you have that the bot is working.
Track the percentage of conversations that come from returning users (customers who've messaged before) versus new users. Growing return rate over time is a healthy sign.
7. Sentiment trends
If you're logging conversation content (appropriately, with privacy controls in place), you can run sentiment analysis on incoming messages to detect trends. Are customers expressing frustration more often than they were last month? Are certain question types consistently generating negative responses?
Sentiment analysis doesn't need to be sophisticated. Even a simple classification — positive, neutral, negative — applied to conversations gives you a trend line to track.
8. Error rate
How often does the bot fail entirely — no reply, an error message sent to the customer, or a system crash? Track this separately from completion rate because an error is a different kind of failure. A conversation that doesn't complete might still have been handled gracefully. A conversation that produces an error message is a broken experience.
An error rate above 1% in production needs immediate attention.
What good looks like
| Metric | Needs work | Good | Excellent |
|---|---|---|---|
| Response time (p95) | >15 seconds | <5 seconds | <2 seconds |
| Completion rate | <60% | 70–85% | >90% |
| Handover rate | >30% or <3% | 10–20% | 8–15% |
| Error rate | >2% | <1% | <0.1% |
| Cost per conversation | >£1 | £0.10–0.50 | <£0.10 |
Setting up basic logging
You don't need a sophisticated analytics platform to start tracking these metrics. A Google Sheet with a row per conversation — timestamp, conversation duration, message count, outcome (completed/escalated/errored), and response time — gives you enough data to spot trends.
For message volume by hour, a simple tally logged to a spreadsheet and visualised with a basic chart is sufficient to start.
As your volume grows, consider a dedicated logging service. CodeWords connects to external logging tools via Composio — you can route conversation events to a database, a BI tool, or a service like PostHog with a few lines of configuration.
Using conversation data to improve your system prompt
Your most valuable analytics resource is the conversations themselves. Reading through 20–30 recent conversations — especially the ones that ended in handover or error — will reveal patterns that no dashboard metric can capture.
Common improvements that come from conversation analysis:
- Questions the bot consistently answers poorly → add the answer explicitly to the system prompt
- Questions the bot escalates unnecessarily → broaden the scope in the system prompt
- Conversations where customers repeat themselves → the bot isn't maintaining context, check memory configuration
- Conversations where the tone feels wrong → adjust the persona instructions in the system prompt
Build a habit of reviewing conversations weekly in the early months. As the bot stabilises, monthly reviews are usually enough.
Connecting analytics to business outcomes
Ultimately, bot analytics are only meaningful in relation to your business outcomes. The metrics above are leading indicators. The business outcomes are what you actually care about:
- Leads generated — how many WhatsApp conversations turned into a qualified lead or a sales call?
- Bookings made — how many conversations resulted in an appointment booked?
- Support tickets deflected — how many questions did the bot answer that would otherwise have required a human?
- Orders processed — how many conversations resulted in a completed purchase?
If you're using CodeWords, you can build a pipeline that logs conversation outcomes to a CRM, a Google Sheet, or a reporting tool. Tell Cody: "When a conversation ends in a booking, log the customer's name, phone number, and booking details to my Google Sheet." Cody builds the logging step automatically.
Tie your bot metrics to these business outcomes quarterly. That's the true measure of whether your WhatsApp agent is working — not the completion rate, but the revenue, the saved time, and the customer satisfaction it generates.
Related reading