Landbot With Causality Engine: What Do the Chats Sell?

The short answer: Landbot will tell you that 3,000 people talked to the bot this month, and the shop will tell you what sold, and no screen in either joins the two. Landbot, the no-code builder for AI agents and rule-based chatbots on websites, WhatsApp and Messenger, reports chats and messages: an Active chat is a user who reached at least the Start Message (Metrics section article). No help article counts orders. That makes this a partial pairing, and an honest one: a bot on the store sends no traffic of its own, so the question is what the bot changes among visitors who are already there. Landbot's A/B Test block answers that for free. Causality Engine, a causal attribution tool that reads the store's GA4 export, answers the channel question around it, and its first finding is free at app.causalityengine.ai/start.

Disclosure: I founded Causality Engine and own it, and this page recommends it. Every claim about it links to Causality Engine's own pages.

Before a bot goes on the product pages:

1. Price a chat against an order. For example, 1,000 chats past the plan's allowance at $0.05 each cost 1,000 x $0.05 = $50, per Landbot's pricing page (checked on 11 November 2026). Set that beside the margin on the orders you expect, not beside the chat count.
2. Use the export Landbot already gives you. Conversations' data lists every user and every field the bot saved, and Send report emails it as a CSV (Metrics section article). If the bot asks for an email, that column joins to the shop's orders.
3. Decide what GA4 should see. A Code block can fire a GA4 event at any step of a web bot (embedded GA4 guide). Pick the step that means something, not "chat started".

What Landbot measures, and what it does not

The Metrics section shows Active chats, New chats, Received messages, Sent messages and Total messages, by bot and by channel, over up to 12 months (Metrics section article). Flow Analytics, under Analyze, shows how users move through a bot and where they leave, with a row for each path of an A/B Test block (A/B Test article).

WhatsApp campaigns add three counts: Message Templates sent, Delivered and Read. Read depends on the recipient's privacy settings, and the times shown are when Meta confirmed each status, not when you sent (WhatsApp Campaigns article). There is no click, order or revenue figure in any of these. Landbot claims no sales, which is unusual in this set, and it means the comparison has to be built rather than audited.

Landbot and the read, compared

What Landbot can hand to GA4

Two setups, two guides. A bot embedded in your own site as Fullpage, Popup, Embed or Livechat uses the page's GA4 tag and fires events from a Code block such as window.gtag('event', 'start chat'); Landbot says this applies on a domain other than landbot.io and is what keeps "a correct source/medium/channel attribution" (embedded GA4 guide). Events do not fire in preview mode, so test in production. A standalone bot on the link from the Share section takes its own GA4 snippet in Custom Tracking Snippets, and that method is marked not compatible with WhatsApp or Messenger (standalone GA guide).

For the shop, embedding is the setup that keeps the visitor's original channel intact. The bot then changes nothing about how GA4 files the session: someone who arrived from a Google ad is still Paid Search when they buy after chatting. A standalone bot is different, because its links into the shop leave one site for another. Tag them. Google's URL builder page asks for utm_source, utm_medium and utm_campaign on every tagged link, and its direct traffic page lists missing UTM parameters as a cause of (direct) / (none).

WhatsApp bots are a third case. Google's default channel group page puts a visit with source whatsapp under Organic Social, since its published source list classes whatsapp as social. Any store link a WhatsApp bot sends needs that tag typed in.

Worked example: a product finder split in two

For example, with numbers chosen for the sums and no real shop behind them: an embedded product finder runs a 50/50 A/B Test block. Path A ends on a 10% code, path B on a plain product link. Each path saves a field named finder_path. In a month, 3,000 chats split 1,540 to A and 1,460 to B, close to even but not exact, which Landbot says is normal for a random split.

The Send report CSV, joined to the shop's orders by email, shows 92 buyers from path A and 70 from path B. That is 92 ÷ 1,540 = 5.97% against 70 ÷ 1,460 = 4.79%, a gap of 1.18 points, or about 0.0118 x 1,540 = 18 orders the code added, before its discount is paid for.

Meanwhile the channel reports run on. Meta claims 610 orders for the month, Google Ads 548, and the shop booked 1,190: 610 + 548 = 1,158 from two platforms alone, leaving 1,190 - 1,158 = 32 for every other channel including organic search and email. The bot appears in none of those figures, because it is not a channel.

That is where Causality Engine's read fits and where it stops. Per its how-it-works page, it splits Direct back to the channels that sent those buyers and sets each channel's caused share beside last-click, with a data-health score from 0 to 100 and a next step. It will not see path A or path B. Read the two answers together: the split says whether the code earns its discount; the read says which channels are worth sending to a page with a bot on it.

The free check, run inside Landbot

The split above is the free check and needs nothing else. The A/B Test block sits under Flow Operations, takes a percentage per path, and works on Web, WhatsApp and Messenger. Flow Analytics compares the paths, and the CSV from Conversations' data ties them to orders by hand. Keep one path as a control that changes nothing, and run it until each path has enough chats for the gap to hold.

When the channel question comes next, Causality Engine, which I founded, takes the GA4 Attribution paths export on the start page; the preview runs in your browser at no cost, and the full read is €99 once with a 30-day refund (pricing, checked on 11 November 2026).

Where neither helps much

• Store purchases need a GA4 purchase event. The read works from purchase paths, and the homepage asks for 12 months or more of GA4 data.
• Small channels stay unread. A channel under 1% of credit on both methods is called too small to act on (pricing for agents).
• Leads that close elsewhere. A bot built to fill HubSpot, Landbot's home ground, ends in a CRM record, not a store order.

Landbot bills by the chat, with each one past the allowance at $0.05 and counted per 24 hours across channels (pricing page, checked on 11 November 2026), and the chat is also the headline unit on its Metrics page, so the number that grows on the dashboard is the number that grows on the invoice.

See also

• Landbot: plans, chat billing and fit
• How to use Landbot: website leads qualified into HubSpot
• Landbot on the marketplace: Landbot's affiliate program terms
• Causality Engine: the read and what it needs
• How to use Causality Engine: export, read, regional test
• Causality Engine on the marketplace: its affiliate terms, given as the owner's
• Landbot vs Tidio: lead flows against a shop help desk
• Conversion rate calculator: orders over sessions, your own inputs

FAQ

Does Landbot connect to Causality Engine?

No, and there is nothing to connect. Landbot can place events in GA4 from a web bot; Causality Engine reads only the Attribution paths file you export from GA4 afterwards (integrations page).

Why do Landbot's hidden fields miss some UTM values?

The Hidden Fields article says they work with lower case only, so a parameter written with capitals is not recognised. Keep tags lowercase, which GA4's case-sensitive values reward anyway.

Can a causal read say whether the bot itself sells?

No. The bot sits on the page, not in the visitor's route, so it never becomes a channel. Use the A/B Test block for that question and the read for the channels that bring people to the bot.