Success metrics that matter
Most confusion about ShopGuide's dashboard isn't about picking the wrong number — it's about treating two different numbers as if they measured the same thing. Here's what each one actually counts, using ShopGuide's real definitions rather than an industry benchmark we'd have to make up.
What each metric actually counts
- Unique users — distinct shoppers, by persistent identity. One person, five visits, still one user.
- Sessions — distinct browsing visits. One shopper can rack up plenty of these over time.
- Interactions — sessions where the shopper actually sent a message, as opposed to a session where the widget loaded and nobody touched it.
- User messages — total message volume. A depth measure, not a reach measure.
- Chat orders / chat-attributed revenue — completed Shopify orders linked back to a chat thread, when a purchase follows a chat interaction.
They're kept separate on purpose. Collapse sessions into interactions, or interactions into orders, and you get comparisons that don't mean what they look like they mean — a month with more sessions but flat interactions is more people seeing the widget without engaging, not more engagement. See Common mistakes for the specific version of this to watch for.
Two ways to count a session
The dashboard has a toggle that switches the session count behind its ratios between Shopify Sessions and ShopGuide Sessions. They are not two measurements of the same thing, so expect a big gap, not a small one.
- Shopify Sessions use Shopify Analytics sessions without bots, only from the countries in your Shopify Markets. You can turn the country filter off with Count only sessions from countries you sell to in Settings (on by default). If your store has not granted ShopGuide the reporting access it needs, the dashboard says so, shows these ratios as unavailable and asks you to switch to ShopGuide Sessions. It never swaps in a different number quietly.
- ShopGuide Sessions use every session ShopGuide tracked, page views included, bots not filtered. That is typically far larger than the Shopify count.
Pick one as your reference point and stick with it across reporting periods rather than switching mid-comparison.
The ratios that use the selected count as their denominator are Chat View Rate, Session Interactions and Sessions per User, plus Session to Order. User Interactions and Chat Conversion (purchases divided by interactions) do not depend on the toggle. Next to the chat figures, the dashboard also shows non-chat orders, revenue and Non-Chat AOV. Read the difference between chat and non-chat as something observed among shoppers who chose to chat, not as proven lift.
AI agents and AI-referred visits
The AI Agents tab in Path Explorer shows two different things. People who arrive from an AI assistant are normal human visits: they count in your sessions, and their carts and purchases are linked like any other visit. AI agents and crawlers are automated, so they are bots. They are left out of Shopify Sessions and show up only in the tab's agent leaderboard, by agent name. Do not compare the two with chat metrics, and read AI-referred conversion as observed, not proven.
Human visits and AI-referred humans count in sessions; AI agents and bots appear only in the AI Agents tab.
Revenue is linked, not modeled
Chat orders and chat-attributed revenue aren't a statistical estimate — they come from directly connecting a completed Shopify order to the shopper's chat thread. That makes the number defensible: every order counted is a real, traceable one. It also makes it conservative. A shopper who chats, leaves, and comes back later without a traceable link to that thread doesn't get counted. Treat this number as a floor on chat's actual influence, not a ceiling.
Knowledge gaps as a quality signal
Separate from volume and revenue, the number and mix of open items in the Knowledge gaps tab is a real leading indicator of chat quality, and it does not depend on shoppers bothering to leave a rating. Gaps are grouped by what you would do about them (Add product info, Clarify a policy, Customer demand, Fix your store) and ranked by how many of those chats did not end in an order. A cluster around one topic tells you exactly what to add next, since recurring phrasings of the same question are stacked into one row rather than scattered across duplicates.
Where a real comparison happens
A/B testing is the one place you get closer to a controlled experiment: it attributes both chat engagement and completed orders to each variant of a launch message or prompt, so you're comparing two configurations against real revenue instead of inferring cause from a trend line. See A/B testing and A/B test results — and give a test a full business cycle before treating an early lead as real.
Next steps
If a number looks odd, ask Path Guide in Path Explorer to explain it: it shows which data it looked at under each answer.