How Breakout's AI Agent Works on Your Website

An AI agent for your website is software that works your visitors the way a good SDR would if they could watch every session: it figures out who the visitor is and why they're there, starts a conversation, answers their questions, shows them what they need to see, and books a meeting or brings in a rep when the time is right.

How AI website agents differ from chatbots and other AI SDRs

Traditional chatbots are based on decision trees: you decide in advance what visitors are likely to ask, give them a few buttons to pick from, and the bot takes them down that path. Ask something unplanned and the bot gets stuck or just asks for an email — a poor experience, and tedious to maintain as your product and messaging change. Most earlier conversational marketing tools, like Drift, worked this way.

AI SDRs replaced the decision tree with a language model grounded in your content — they answer unscripted questions, qualify in the conversation, and book a meeting, and most AI SDRs on the market today answer questions reasonably well. But a good sales conversation is more than answers: a good SDR pulls up the case study that matches the buyer's industry, walks through a demo when asked how something works, and shares a comparison when a buyer is weighing options. Most AI SDRs still do all of this in text, inside a chat window.

The other place they look alike is how the conversation starts — most wait for the visitor to open the chat, or open with the same greeting for everyone ("Hi, how can I help you today?" or something page-based and AI-generated). Most visitors ignore it: they read a few pages, compare you with someone else, and leave. However good the agent's answers are, they only reach the small share of visitors who start a conversation.

Breakout's agent is built around both: starting the right conversation with each visitor, and showing them the assets, demos, and comparisons they need once it starts.

Proactive nudges

A nudge is a short message the agent picks for a specific visitor to start a conversation, chosen from what the agent knows about them: who they are (company and, where identifiable, the person, with firmographics and ICP fit), where they came from (UTM parameters, ad campaign, referring site, email or webinar), what they're doing on the site (current page, pages already read, repeat visit or not), and what your CRM says (target account, existing customer, open deal, account owner).

A few examples: a visitor arriving from G2 or a competitor comparison is likely evaluating options, so the nudge offers a side-by-side comparison. A repeat visitor from a target account gets pricing or a case study from their industry — if the account already has an owner, the nudge can offer a conversation with that rep by name. A visitor who clicked an ad sees a nudge that continues the ad's message, written for their persona. A visitor from Europe sees the report published for that region.

When the visitor clicks a nudge, the agent turns it into a question and answers it in the chat, so they don't have to type anything — nudges can be text or video.

Most sites already have pop-ups, but they show the same offer to everyone and visitors close them without reading. A nudge is picked per visitor, so it's about something they actually care about. Ketch's Director of Demand Generation describes using that context to "nudge them and pique their curiosity with the right thing at the right time."

How the agent works

It knows who's visiting. The agent runs on Breakout's de-anonymization to identify visitors at both account and contact level, combined with what they do on your site and the context in your CRM and marketing automation — past webinars, asset downloads, open opportunities, customer status. Nudges, answers, and routing all run on that context.

It answers from your knowledge base. Every answer is grounded in what you give the agent: your website (point it at the pages that matter and exclude the rest — old blog posts and news pages tend to confuse it more than help, so exclusion rules for paths like /news are recommended, and a properly curated knowledge base produces better conversations than a larger one), documents (positioning docs, sales collateral, FAQs, competitive comparisons, sales call transcripts — a document can be marked internal so the agent uses it to inform answers but never shares it), and demo tours, videos, and images, pulled into the conversation when they answer a question better than text. On top of the knowledge base, you write instructions in plain language, the way you'd brief a new SDR — no decision trees to build.

It surfaces all types of content and actions. Alongside the chat, the agent offers Blocks: book a meeting, take a product tour, watch a video, talk to a person, fill out a form — each an action a visitor can take without starting a conversation first, with Blocks chosen per page. Ketch runs four or five at once and treats them as shortcuts past their own navigation, since visitors rarely dig for content however clearly it's laid out. The agent knows what the visitor did in each Block — if someone watched a product tour, it can answer questions in the context of what they just saw.

It qualifies, routes, and books. When a visitor wants to talk, the agent mirrors your existing form one-to-one and qualifies them against your criteria, using enrichment and the conversation itself, asking only what it can't already infer. Qualified visitors see the right rep's calendar straight away, with routing following account ownership, territory, round robin, or any field on the Salesforce or HubSpot record — the agent can also step in after a regular website form is submitted and show a calendar to qualified leads. Visitors who don't qualify still get their questions answered, and existing customers get sent to support instead of the SDR team: one team estimated that in some months, up to 20% of inbound leads reaching their SDRs were existing customers who didn't realize their company already had access.

AI and live reps together

Not every conversation should be handled by AI. Breakout runs AI chat and live chat in the same agent, so you decide which visitors get a person.

Live handoff. When a visitor asks for a person, or matches criteria you set (a tier-one account, for example), the agent brings in a rep in real time — to the first available rep, round robin across a team, or straight to the account owner. Reps get alerts in Slack, Teams, or email; outside business hours, the visitor gets a calendar instead.

AI auto-pounce. With older tools, reps watch a live feed of visitors and start chats by hand — workable at low traffic, rarely scalable. Auto-pounce solves this: based on your instructions, the agent picks the visitors worth engaging, opens the conversation using what it knows about them, and sends it in a rep's name. To the visitor, it looks like a rep is reaching out; if they respond, reps step in and take the conversation forward.

Translation. Live chat is translated, so a rep can talk to an international visitor in the visitor's own language.

Our rule of thumb is that AI should handle around 80% of conversations, with reps taking the 20% that need a person — the right split depends on who you sell to, and there's value in reps engaging high-value accounts, but human bandwidth should be used thoughtfully.

What customers have seen

• A Fortune 500 customer has generated $3M in pipeline with Breakout.

• Ketch replaced Drift with Breakout, went live within days, and generated more than 150 leads from anonymous traffic in under three months.

• Beautiful.ai, an AE-led sales team with no SDRs, made Breakout their top-of-funnel layer: lead volume rose in the first quarter, they closed three new opportunities in under three months, and fewer product questions reached sales.

• HackerEarth evaluated Drift and Qualified before choosing Breakout — 21% of their Q1 pipeline came from Breakout, and 10 to 22% of conversations led to a captured lead with no human involved.

Setting it up

The agent itself takes three to four days to set up; getting data flowing cleanly into your CRM and reps set up on Breakout takes another week or so. Most of that time goes into decisions rather than configuration — what goes into the knowledge base, what qualifies a lead, how routing should work, who covers live chat — and teams that have those written down move faster. For the first few weeks, Breakout works as an extension of your team to build the workflows with you.

Breakout integrates with Salesforce, HubSpot, Microsoft Dynamics 365, Attio, Pipedrive, Marketo, Slack, Microsoft Teams, Office 365, Google Calendar, and Chili Piper.

Frequently asked questions

How is this different from Qualified or Drift?

Drift was built around rule-based playbooks. Qualified has gone all in on AI with its AI SDR, and while it handles conversations well, it doesn't do a great job showing demos and other visual assets, and parts of the platform still run on rules someone on your team has to build and maintain, making setup tedious. Breakout is AI-native — point it at your content and write instructions in plain language instead of building flows, and the agent is usually live in three to four days — plus you get proactive nudges for every visitor, and live chat and auto-pounce for the conversations that need a rep.

What happens when a visitor asks something the agent shouldn't answer?

It follows your instructions. For topics like pricing, roadmap, or security specifics, you decide what it can share, and it can ask the visitor to book a meeting or hand off to a rep in real time.

Do visitors have to talk to the AI?

No. Blocks let them book a meeting, take a product tour, or ask for a person directly without engaging with an AI agent.

Can it handle support questions?

It recognizes existing customers from your CRM or the questions they're asking and can point them to your support channel. It can answer product questions, but it isn't a support tool, and running customer support through it isn't recommended.

Can we run it on more than one website?

Yes — each domain can have its own agent and knowledge base.

How is it priced?

Pricing is based on your website traffic, expected lead volume, and the reps involved in the handoff.