Skip to content
Chatbotscape
Editorial flat-vector illustration for How to Set Up Proactive Chat Triggers Without Annoying Visitors
Published 6 min read

How to Set Up Proactive Chat Triggers Without Annoying Visitors

Quick answer: Pick two or three pages where a visitor is likely to want help, give each one trigger, set the delay from how long people really stay, and write an offer, not a pitch. Make sure a message does not repeat, plan what happens if the visitor answers, and measure dismissals as well as conversations. Start small and widen only when the numbers say to.

Step 1: Decide whether the chat should speak first at all

Proactive chat suits pages where people stop to decide: pricing, a cart, a booking page. It suits poorly pages where they read or are already finishing a task: a long article, a checkout payment step, a help article they came to read. If your team cannot answer quickly, or the bot cannot handle the follow-up, wait; see when not to use a chatbot. A message that gets a slow or wrong answer is worse than no message.

Step 2: Choose two or three pages

Open your analytics and list pages by visits and by how often visitors leave without acting. Pick the few where help would plausibly be wanted. The vendor pages read for this guide give examples: Zendesk names a shopping cart page and a pricing page; Intercom's example is a pricing page reached from a particular promotion. These are examples, not rules for your site.

Write one sentence per page: "On the pricing page, visitors compare plans, so we offer help choosing." If you cannot write the sentence, leave the page out.

Step 3: Pick one trigger per page

Documented triggers include time on page, number of visits, leaving the page, referrer, country and a date window (Microsoft's Dynamics 365 page lists these among eight scenarios). Start with the simplest: time on page, optionally plus a returning visit, as in Zendesk's returning-visitor example (previous visits greater than zero, excluding people already helped). Combining many conditions makes the rule hard to test and explain.

If your platform has no such feature, Dynamics 365 shows the code route: wait for the widget to be ready, then call its start-proactive-chat function with your message. Their sample code has issues the page does not mention, including a syntax error in one example and a debug alert in another, so test any code before it reaches real visitors.

Step 4: Set the delay from data

Do not copy a number. Intercom's example uses 10 seconds and Zendesk's cart example uses more than 60 seconds; both are examples for their own scenarios. For your pages, look at how long visitors who convert stay before they act, and put the message after the normal reading time, not inside it. Make a note of the reason.

A useful check: on mobile, the widget covers more of the screen. Test the same rule there, and consider whether a smaller prompt or none is better.

Step 5: Write an offer, not a pitch

Short, specific to the page, easy to ignore. Templates to adapt:

Pricing page: Comparing plans? Tell me what you need and I'll point you to the right one.

Cart page: Questions about delivery or returns before you check out? Ask here.

Returning visitor: Welcome back. Want to pick up where you left off?

Avoid fake names or invented urgency, and do not pretend a bot is a person. For a returning visitor, say what you know only if you really know it. More wording ideas are in chatbot fallback copywriting. Test two versions with the method in A/B testing chatbot messages.

Step 6: Stop repeats, plan the follow-up and measure

Stop repeats. Several vendor pages are silent on frequency limits (Intercom, Zendesk and Microsoft's pages describe none), so find out whether your platform stops a message from reappearing after a dismissal, and test it. If it does not, add your own rule, for example one proactive message per visitor per session.

Plan the follow-up. Zendesk's article says proactive chats skip the pre-chat form, so you may not get a name or email and routing can be bypassed. Decide what the bot asks next and when a person takes over; see the handoff design guide. If it is outside working hours, see setting up an after-hours chatbot.

Measure. Count proactive and visitor-started chats separately, plus dismissals, replies, and what happened after. See chatbot metrics and conversion rate.

Check privacy. Cookies for visit counts and location for country triggers may need consent in your region; see chatbot data privacy and the GDPR chatbot compliance guide. This is not legal advice.

Check the arithmetic

An invented example, not a measurement. A small shop's pricing page gets 1,000 visits in a week. The proactive message shows to 400 of them (those who stayed past the delay). 80 dismiss it, 16 reply, and 304 ignore it. 80 + 16 + 304 = 400. The reply rate is 16 ÷ 400 = 4%, and the dismissal rate is 80 ÷ 400 = 20%. If 5 of the 16 repliers then buy, 5 ÷ 16 = 31.25%. Without a comparison group, such as visitors who were not shown the message, you cannot say the message caused those purchases. These numbers are made up to show the method.

Common mistakes

  • Showing it on every page.
  • Repeating after a dismissal.
  • Copying someone else's delay instead of using your own data.
  • Sending a visitor to a bot that cannot answer the follow-up.
  • Counting only the chats started, not the dismissals.
  • Skipping the mobile test.

FAQ

How do I set up proactive chat? Choose two or three high-intent pages, pick one trigger for each, set a delay from your own visit data, write a short offer, stop repeats, plan the follow-up and measure.

What are good proactive chat triggers? Time on a pricing or cart page is the usual starting point; returning visits, referrer and exit movement are also documented by vendors. Which works best depends on your site.

How long should I wait before showing a proactive message? There is no standard. Intercom's example uses 10 seconds and Zendesk's cart example uses over 60 seconds. Use how long your own visitors stay before acting.

Will proactive chat annoy visitors? It can if it repeats or appears where people are reading. The vendor pages read here give no figures, so start with a few pages and watch dismissals.

Can I use a chatbot for proactive messages? Yes, if the bot can handle the replies. Otherwise route to a person; see the handoff guide.

Do I need an agent online? Not necessarily, but decide what the visitor sees when nobody can answer quickly.

Sources