How to measure an AI sales agent without fooling yourself
Measure an AI sales agent on what you can check. Count how many shoppers it showed products to, how many added to cart and bought, and what kind of help each order got. Then read the conversations. Treat "AI-influenced revenue" as a record of which orders went through a chat, because it doesn't prove the chat caused them.
Why "influenced" numbers run high
Big figures get quoted without a definition. After Cyber Week 2025, Salesforce reported that AI and agents influenced 20% of global orders that week, about $67 billion in sales, without saying in the release what counted as influence. A product recommendation shown once and a cart built in a conversation can both end up in a number like that.
Shoppers who start a chat are often closer to buying than the average visitor. In the Mindtree survey of in-store shoppers, people who talked to an associate were 43% more likely to buy, and part of that is simply who chooses to ask. Your chat users may convert better than your site average whether or not the agent helped.
What Agentmatica's Insights shows
The shopping funnel counts five steps from your widget: products suggested, product clicks, added to cart, checkouts started, and orders.
Orders the agent touched are labelled by how it helped. Checkout built by the agent means the agent built the cart and the shopper checked out through its link, so the whole order counts. Bought a suggested product means the shopper added an item from a card in the chat. Recommended, then added means they added a product the agent had shown them using your store's own button. For those last two, only the lines the agent touched count towards its revenue.
Sales by Agentmatica then shows revenue, orders, average order value, items per order and the agent's top five products. Order tracking needs our Shopify app or the WooCommerce plugin.
These are records of what happened in and after a chat. We don't run a holdout group, so Insights can't tell you how many of those orders would have happened anyway.
Numbers worth watching
Start with the split between labels. A rising share of Checkout built by the agent means shoppers are buying inside the conversation. A big Recommended, then added share means the agent's suggestions matter but the store's own pages close the sale.
Watch store-wide items per order before and after you turn on pairings, over the same weeks of the year if you can. It's crude, but it isn't limited to chat users.
Look for conversations where the agent couldn't answer. Each one is a product question your pages don't cover.
A weekly routine
Once a week, read ten conversations that ended in an order and ten that didn't. In the first group, check the agent recommended something that fit. In the second, look for the moment the shopper went quiet, and whether a page, a Sales note or a pairing would have changed it. Change one thing, and compare the funnel the following week.
Questions store owners ask
What is AI-attributed revenue?
It's the value of orders linked to an AI agent's involvement, such as a cart the agent built or a product it recommended that the shopper then bought. How a tool defines that link decides the number, so check the definition before comparing figures.
Does Agentmatica prove the chat caused the sale?
No. Insights records which orders went through the agent and how it helped: building the checkout, a product added from a chat card, or a recommended product added on the store's own page. It doesn't measure what would have happened without the chat.
Which metrics should I check first?
Check the shopping funnel from products suggested to orders, the share of orders where the agent built the checkout, store-wide items per order over time, and conversations where the agent couldn't answer.
See what your agent sells
Agentmatica's Insights shows the shopping funnel, how each order was helped and what the agent recommended. Every plan includes it, Free too.
Try it free