How to Measure ROI with the Data Extraction Integration

Every conversation your AI agent has carries a signal of value - a quoted deal size, an order amount, a resolved ticket, an hour of work saved. On their own these signals are buried in free-form chat. The Data Extraction integration turns them into structured numbers you can measure over time, which is exactly the raw material you need to tell an ROI story.

This tutorial shows you how to pick a measure, extract it from every conversation, and use the resulting datapoints to demonstrate value.

The Idea: Pick One Measure

ROI starts with a single number that represents value. Don't try to measure everything - choose one measure that maps to a business outcome:

• Revenue: deal value, order amount, upsell value
• Efficiency: hours saved, tickets deflected, calls avoided
• Quality: satisfaction score, lead quality rating, conversion rate

The Extract integration's job is to read each conversation and pull out that one number, consistently, so it can be charted and compared.

Step 1: Define a Measurable Schema

Create a Data Extraction integration and give it a schema with at least one numeric field. Two properties turn a plain field into a measure:

• collect: true - tells ChatBotKit to record the value as a metric (numeric fields only)
• display - controls how the value is formatted on the chart (number, percent, or currency/USD)

Here is a schema for a sales assistant that estimates the value of each opportunity:

The dealValue field is your measure. Because it is marked with collect: true, every conversation that surfaces a value contributes a datapoint, and display: currency/USD makes the chart read in dollars rather than raw numbers.

Tip: Write a precise description. The clearer you are about what the number means (and its unit), the more consistent the extracted values will be. Design your bot's backstory so the information needed to compute the measure naturally comes up in conversation.

Step 2: Let It Collect

With the Trigger set to automatic, extraction runs after each conversation completes and the collected value is logged as a metric. Over time these datapoints accumulate into the Measured Values chart on your integration page.

Already have a history of conversations? Use the Trigger button on the integration page to apply your schema to the most recent 100 conversations and backfill the chart - useful for establishing a baseline before you start telling the story.

Step 3: Turn Datapoints Into a Story

A single extracted number is a datapoint. A series of them is a narrative:

• Trend - Is the measure growing? The Measured Values chart shows the daily series, so you can point to the line moving up and to the right.
• Value vs. cost - ROI is value relative to what it cost to produce. ChatBotKit attributes token usage per bot and per conversation, so you can place the value your agent generated next to the credits it consumed. When the value line clears the cost line, the agent is paying for itself.
• Mix - Collect more than one measure (for example dealValue and a satisfactionScore) and the chart layers them, formatting each in its own units, so you can tell a richer story than a single number allows.

The point is to move the conversation from "the bot handled X chats" to "the bot influenced $Y of pipeline" - a measure a stakeholder can act on.

Step 4: Surface It Where Decisions Happen

Once your measure is collecting, put it in front of the right people:

• Integration page - the Measured Values chart and the Extracted Items table (exportable as CSV) live on the integration itself.
• Blueprint designer - drop an Extract Chart tool onto your blueprint canvas and connect it to the integration to keep the measure visible alongside the rest of your agent's design.
• Webhook - configure a request URL on the integration to push each extracted value into your own dashboard, spreadsheet, or BI tool for deeper analysis.

Wrapping Up

The Data Extraction integration is more than a way to capture form fields - it is a measurement instrument. Pick a measure, mark it with collect: true, format it with display, and let it run. The accumulated datapoints become the evidence behind your ROI story, turning everyday conversations into a number your business can stand behind.

For the full configuration reference, see the Data Extraction documentation.