On 15 November 2026 Looma launched Magic Tracking — it sees the cart, the checkout and the purchase of people who used the chat or the search. Here are the first five days, next to three months of conversations we already had.

Until this month we could tell a store how many customers talked to the AI and how many opened a product from the search. What we could not tell was the thing every owner asks first: did they buy?
On 15 November 2026 we launched Magic Tracking. It is a new part of Looma that sees three steps on the store’s own pages — a product added to the cart, the checkout started, the order placed — and marks the orders of people who used Looma’s chat or search within the 30 days before they bought. It is on by default for every store, and the store changes nothing in its code: the same Looma script that shows the chat does the tracking.
So the order numbers in this article are young. They cover the first five full days of tracking: 16–20 November 2026, in several stores. Next to them we put what Looma has recorded for much longer — conversations and searches, here for three months, 1 August – 31 October 2026. These are two different sets of data with two different levels of confidence, and every chart below says which one it shows.
Magic Tracking marks an order as “after Looma” when the buyer talked to the AI, used the search or clicked a product Looma showed at some point in the previous 30 days. It follows the person, not the click: someone who asked a question on Monday and came back to buy on Thursday is counted.
Read that definition twice. It is orders placed by people who used Looma — not orders that would not have happened without it. A customer who searched and then bought might have bought anyway. We come back to this at the end.
Of every hundred people who talked to the AI or used the search, about seventeen put something in the cart and about eight placed an order within those five days. We do not have the same funnel for visitors who never touched Looma — the tracker keeps only daily totals for them — so we cannot put the two side by side.
From 1% to 25%. The stores at the top of the range run both the chat and Looma’s search on the site, so a large part of their buyers pass through them. The stores at the bottom use the chat only. This is the most important thing to take from the chart: the number says how much of the store’s buying path runs through the assistant, not how good the assistant is.
Almost half of these orders were placed within ten minutes of the conversation or the search, and two thirds within an hour. One in seven came more than a day later — the customer asked, thought it over and came back.
In most of these stores the people who used Looma spent 17–36% more per order than the store’s other buyers; in one store they spent 13% less. With 13 to 63 orders per store this is a hint, not a result: one large order moves the bar. We will repeat the comparison when there is a month of data.
Orders are new. Conversations and searches are not: Looma has recorded them since each store connected. So for this part we can take a long period — three full months, 1 August to 31 October 2026 — and the stores that had at least 50 customer conversations in it.
In 7,265 of 10,731 conversations the AI did everything itself: no operator wrote a single message, nobody joined the chat, and the AI did not ask for a human. That is 67.7% overall — but stores differ a lot, so here is the range:
The spread — from 37% to 77% — is mostly about how the store is run, not about the AI. Where operators like to step in early, the share is lower. Where the catalog and the store’s policies are loaded into the knowledge base in full, the AI has fewer reasons to call a person.
We took the time of the first message in each conversation and compared it with a normal working day: Monday to Friday, 9:00 to 18:00.
More than half of the customers write when a store without an AI assistant would answer “we’re closed”. One in eight writes at night. The average wait for the first AI reply across these conversations is about seven seconds.
In 21.9% of the conversations the customer clicked a product the AI showed — 4,278 product pages opened from the chat in three months. And 505 times a customer left an order or a request right in the chat.
1 Aug – 31 Oct 2026 · 3 months
Some of the stores also use Looma for the search box on their site. In the same three months their visitors ran 32,221 searches we have a record of; 98% of them returned products, and 27,251 ended with the visitor opening one.
A caution we owe you: a search is recorded when the visitor either clicks a result or closes the search panel. Someone who typed a query and left the page is not in this count, so the real share of searches that end in a click is lower than these two numbers suggest. What we can say for certain is the absolute figure: 27,251 times in three months the search took a visitor to a product page.
People who ask questions and search are more interested to begin with. To learn what the assistant adds, a store has to compare periods or run a split test — the tracker alone cannot tell.
Magic Tracking started on 15 November, and the order figures cover 16–20 November. A sale, a weekend or one wholesale buyer can move them. Treat them as a first reading.
A buyer who talked to the AI on a phone and ordered from a laptop is two different visitors to the tracker. In some stores it linked nothing at all in those days. The real share is not lower than what we show.
Orders without an order number are not counted. One browser that placed 56 orders in five days — clearly a store employee entering phone orders — is excluded together with every other browser that placed four or more orders. We then built the same rule into the product — see the last section.
The larger ones hold most of the conversations. That is why the charts give the range between stores and the middle store next to the total.
Magic Tracking is already on: it is enabled by default, and there is nothing to install or change on the site. The “Effectiveness” tab of the Looma dashboard shows the orders after Looma, their share of all orders, the average order, the funnel and the products bought — for any period, with the same definitions used in this article. If one of the three steps is not caught on your site, the tracker’s settings let you point it at the right page or button.

Below the totals the tab lists the orders themselves: the order number you can find in your store admin, what was bought, the amount, the buyer’s city, device and traffic source, and what happened before the purchase — the search query the buyer typed or the chat they had, with a mark showing whether the AI led that chat alone or an operator joined.

The employee’s browser from the fine print was not a one-off: wherever managers enter phone orders through the site or test the checkout, their orders landed among the orders after Looma. So the Magic Tracking settings now have a “Who is not counted” block with three rules. A browser that placed four or more orders within 30 days is treated as store staff — the number can be changed. Your team signed in to the Looma dashboard or app is recognized by IP address. And you can list your own IP addresses, such as your office’s. The first two are on by default.

These orders do not disappear: they stay in the store’s total number of orders and are only left out of the orders after Looma, the revenue and the funnel; the dashboard says how many were left out. We would rather show a smaller number you can trust.
Three things in the older data do not depend on the sample. Customers write when you are closed — more than half of them. Most of what they ask can be answered without a person, if the catalog and the store’s rules are there to answer from. And a search that understands the question takes people to products tens of thousands of times a quarter. Whether that turns into orders is what you can now watch day by day.
Looma installs in minutes and starts selling for you 24/7.
Try it free →Looma installs in minutes and sells for you 24/7.