The AI Analytics tab in WAX brings together all the performance and efficiency indicators of your AI agent.
Here's how to read these indicators correctly, whatever the channel: WhatsApp, RCS or SMS. The KPIs and the way to read them are exactly the same from one channel to another; only the data changes.
1. Understand where the data comes from
Before interpreting a number, you need to understand where it comes from. The data shown in AI Analytics relies on two mechanisms:
UTM tracking: by default, WAX adds a UTM parameter to every link sent by an AI agent (WhatsApp, RCS or SMS). This tagging is what links a click to a session, then to a conversion on the website.
Without properly set UTMs, a conversion can happen without being linked to the conversation that generated it, which pulls revenue indicators down.
Attribution: attribution decides which conversation gets credit for a conversion. WAX attributes a sale to an AI conversation when it happens within a set time window after the interaction (72 hours by default, can be changed).
Two things to watch:
A conversion can be influenced by several channels (email, ads, AI): the attribution model you choose directly affects the revenue figures shown.
A conversation that goes well for the AI doesn't always lead to a measurable conversion if the customer buys later, on another device or without going back through the tracked link.
💡 Key takeaway: a low or falling revenue KPI should always be read in light of tracking quality before being seen as a sign that the AI agent is underperforming.
2. AI Agent Performance
This section measures how well the AI agent handles conversations and generates attributed revenue.
Resolution Rate
Share of conversations fully handled by the AI, with no human intervention and no drop-off. It's the key indicator of the agent's autonomy.
A drop can point to a change in customer requests, a flow change, or a decline in the quality of AI answers.
An increase doesn't guarantee better customer satisfaction: an AI can resolve a conversation without the customer being fully satisfied. It's useful to cross-check this figure with the Escalation Rate (see below).
ROI
Ratio between the revenue generated and the cost of the AI over the period.
Revenue
Total sales attributed to AI conversations over the period, based on the attribution model and UTM tracking in place. Always compare it with the previous period to judge a trend, never as an isolated absolute value.
Cost
Running cost of the AI agent over the period.
Conversations
Total volume of conversations handled by the AI over the period. It's the denominator of several other KPIs: always check a change in this number before interpreting a change in revenue.
3. AI Efficiency
This section complements the first one by measuring the quality and economic value of each conversation.
Revenue / Conversation
Average revenue attributed per conversation handled. It brings performance down to the scale of one interaction and lets you compare several clients or campaigns with each other, regardless of volume.
Escalation Rate
Share of conversations handed over to a human agent. This rate mirrors the Resolution Rate and should always be read together with it:
A rising Escalation Rate together with a falling Resolution Rate confirms a real loss of AI autonomy.
A rising Escalation Rate with a stable Resolution Rate may simply reflect a change in the mix of requests (more complex cases sent to the AI).
Average Order Value (AOV)
Average basket of the orders attributed to AI conversations. A change in AOV isn't always linked to the AI itself: it can reflect a catalog change, an ongoing promotion, or product seasonality on the client's side.
Hours Gained
Estimate of the human handling time saved thanks to the AI, calculated from the volume of conversations resolved without intervention and a reference average handling time. It's a useful indicator to show the value of the AI to support teams or management, beyond revenue alone.
4. Reading across channels
All these KPIs are calculated with the same method, whatever channel the conversation happens on. There's no separate dashboard or calculation logic for WhatsApp or RCS: AI Analytics aggregates the data across channels.
A few channel nuances to know, which can explain performance gaps without calling the AI's quality into question:
WhatsApp: conversations rely on templates approved by Meta and a conversation window (24 hours) that can affect the Resolution Rate if the customer doesn't reply in time.
RCS: the channel's availability depends on the recipient's device and carrier compatibility; a lower volume of conversations on RCS may simply reflect RCS coverage that is still partial, not a less relevant AI.
💡 Key takeaway: when you compare the same KPI across two channels, always take these technical specifics into account rather than treating it as a strictly like-for-like comparison.
5. Best practices for reading your results
Always compare one period with another, never an isolated number (WAX always shows the change versus the previous comparison period, to the right of each KPI).
Cross-check KPIs rather than reading them one by one: Resolution Rate + Escalation Rate, or Revenue + Conversations, tell a more reliable story than a single number.
Check tracking before drawing conclusions on performance: falling revenue can come from a UTM or e-commerce integration issue, not from the AI itself.
Adapt your reading to the channel: the same KPI doesn't necessarily mean the same thing operationally.
Look at the chart's trend rather than a single point: one-off peaks are often less meaningful than a trend that holds over several days.