Closio for Shopify

How Closio counts revenue

Closio only claims an order when there is a direct link between the shopper and the agent. Here is exactly what counts, what does not, and why the numbers differ from Shopify's own reports.

The 30-day window

An order is "influenced" when the same shopper talked to the agent within the 30 days before the order. The clock starts at the conversation, not at the first page view, and it never extends past 30 days.

How we know it is the same shopper

When a shopper opens the agent, the widget stores a random visitor id in the browser and writes it to the cart as a hidden cart attribute called cl_vid. When that cart becomes an order, Shopify sends us the order with the attribute, and we match it to the conversation.

No name, email or address is used for matching. If the shopper clears their browser or buys from another device, the match is simply missed and the order is not counted.

Offer code match

When the agent gives a discount code, that code belongs to one offer event. An order that uses the code is attributed to that offer even if the cart attribute was lost. One code can only be counted once.

The holdout: how uplift is measured

Not every hesitating shopper gets an offer. On Growth and Pro a fixed share (you set it, default 20 %) is placed in a holdout group that is eligible for an offer but never shown one. Everything else about their conversation is identical.

Uplift is the difference in how often the two groups ended up ordering. Because the holdout is chosen at random, that difference is caused by the offer, not by the kind of shopper who asks for discounts.

Both groups are measured with the same rule: a session "converted" when the same shopper placed an attributed order within 30 days after the offer moment. The holdout never gets a discount code, so we never use the code to decide; it is always the shopper's later order that counts.

What "Likely" and "Confident" mean

Small groups swing a lot: 3 orders out of 12 holdout sessions is 25 %, but one more or one fewer order moves it to 33 % or 17 %. The confidence badge says how unlikely it is that the gap you see is just that kind of noise.

We run a one-tailed two-proportion z-test asking "does the offered group buy more often than the holdout?". "Likely" means there is at most a 10 % chance the gap is noise (90 % confidence). "Confident" means at most 5 % (95 % confidence). "No clear difference yet" means the gap is inside the noise.

Below 100 sessions across both groups we only show "not enough data" and the raw rates, because no test is meaningful on so few. The card also tells you how many sessions each group has; aim for 100+ in each before changing your ladder based on the result. The 13-week trend under the card shows whether the two lines are pulling apart week after week, which is stronger evidence than one number.

Segments: who converts

Insights breaks conversion and attributed revenue down by segment. Each conversation lands in exactly one row of each segment, so the rows always add up to the overall total.

  • New vs returning: "returning" means the shopper already talked to the agent before the range started (same browser), already bought with the agent before, or was logged in as a customer we had seen before. Everyone else is "new".
  • Chat vs voice: the channel the conversation ran on; "mixed" started in chat and switched to voice or back.
  • By product: the product page the conversation started on. The ten busiest products are listed; the rest are grouped, and conversations that started on another page are "No product page".
  • Mobile vs desktop: the widget reports the device (coarse pointer or a narrow viewport counts as mobile); visitors seen before that was recorded show as unknown.
  • Revenue in a segment is the total of attributed orders linked to those conversations, in your shop currency. Orders in another currency are counted as orders but left out of the revenue figure, and the table says how many.

The funnel: opened → asked → proposed → confirmed → ordered

The funnel follows one conversation through five steps. A conversation only counts in a step when it also passed every step before it, so each count is never higher than the one before and the percentage is a real drop-off.

  • Opened: a shopper opened the agent (test chats from your Knowledge page are never counted).
  • Asked: the shopper sent at least one message.
  • Proposed: the agent proposed items for the cart (it never adds anything on its own).
  • Confirmed: the shopper pressed "Yes, add" (or said yes in voice) and the items went to the cart.
  • Ordered: an attributed order followed that conversation.
  • Shoppers also buy without the in-chat cart, for example from the product page after asking a question. Those orders are still attributed, and the funnel lists them separately as "ordered without the in-chat cart" so the last step never claims more than it shows.

Why the numbers differ from Shopify's reports

Shopify attributes sales to marketing channels (Google, email, direct). Closio attributes to a conversation. One order can appear under a channel in Shopify and under Closio here; that is not double counting, it is two different questions: "where did the visit come from?" and "did the agent help close it?".

Totals use the order's total price including shipping and taxes, in the order currency, at the moment the order is created. Later refunds, edits and cancellations are not subtracted.

What is NOT counted

  • Orders from shoppers who never opened the agent, even if the widget was on the page.
  • Orders placed more than 30 days after the last conversation.
  • Orders from another device or browser where the cart attribute is missing and no offer code was used.
  • Draft orders, POS orders and orders imported from other systems.
  • Refunds, chargebacks and edits after the order was created (the figure is gross).
  • Any order already attributed once (an order id is only ever counted once).

Margin kept after offers

Each accepted offer records the margin that remained after the discount, computed server-side from your margin floor and, on Pro, your unit costs. The dashboard shows the average over accepted offers in the period. The agent never decides the discount; the server does.

Check it yourself

On Pro you can export the period's conversations, offers and attributed orders as CSV from the Today page and reconcile them against Shopify order ids line by line. Insights adds the segments, the funnel and the weekly holdout table as CSV, plus a warehouse export (JSON Lines, one conversation or order per line) for your own BI tool. Every figure on these pages is computed from those same rows.