How To Use Warmly's AI Chat To Drive More Pipeline

Why use AI Chat

🎆 Catching buyer interest when they're hot on your site is like trying to capture the perfect picture of a firework right when it explodes.

⚡️ Or it's like catching lightning in a bottle.

The timing needs to be perfect. You miss it; it's gone.

That's why we created AI Chat.

We've found that proactively engaging with target prospects live on the website can cut sales cycles by 30%+ and increase pipeline 2x.

Especially when prospects are visiting high-intent pages like pricing and case studies, you know they're thinking about you. If you catch them when their interest is piqued, you could skip the form fill → meeting booking → actually have the meeting in a week.

By then the prospect might've gone with your competitor.

Instead, compress the sales cycle by having the meeting right on the site.

"You're free, I'm free, let's have a conversation right now"

Didn't even need their number to cold call.

The main problem is your sales team only has so many working hours. There are potentially thousands of qualified prospects visiting your site every day. It's not always feasible for a human to interact with every visitor.

A machine doesn't have the same limitation.

Arm AI to the teeth with context, and it can deliver the first 1-2 human messages faster than any rep, with the same degree of relevance and authenticity. It never sleeps. It doesn't get discouraged (yet). It scales infinitely.

The moment the AI gets a qualified chat response, it can loop in the right seller.

Let AI line up conversations for you.

Step-by-step instructions to set up AI chat

Step 1: Make sure you're on the right plan

AI chat is only available for customers on our paid plan. If you have any questions, please contact our sales team at sales@warmly.ai.

Step 2: Enable AI Chat under your Settings tab

Step 3: Select the person you want the AI to send messages on behalf of

Step 4: Configure your AI Chat prompts

There are two proactive chat messages your AI can send. Right now you can't configure the delay of when the messages send (we're working on this toggle right now).

Turn on AI Generated for the two chat prompts then click Edit to edit the prompts.

You will see a prompt playground pop up where you can test out prompts for AI generated messages that will automatically sent on a rep's behalf. See the results on the right-hand side

Step 5: Configure your AI chat fallbacks

If the AI is unable to generate a chat message, write the text to that the AI will fall back to

Step 6: Set up the segments and conditions for which the AI Chat is triggered

Under Triggering select + Add trigger and select from the dropdown which action should trigger the AI chat.

For example you can select Lands on page for the AI to message when the visitor lands on a specific page. Then set the conditions for which the AI should proactively chat.

You can also specify the pre-defined segments for which you want the AI to proactively chat. For example, you could create segments based on ICP companies or target accounts that are open opportunities

Step 7: Set up Slack notifications when a visitor responds to your AI

You want to be notified when a visitor actually responds to your AI so you can respond in real-time.

Head over to your Slack Integrations and add a new notification

Scroll down to the Slack notification segment called Respond to Chat and select the Slack channel you want to be notified when a visitor responds to your AI chat. We recommend creating a new channel in Slack dedicated to visitor responses to your AI.

You'll start to see notifications flood in of people responding to your AI chat like in the example below where Brian Luppert from Flip CX was on our case studies page and responded to a message our AI sent on behalf of Katherine, our AE.

AI Chat Prompt Best Practices

Keep the messages human

We recommend you keep the AI-generated messages short, casual, and less than 15 words to emulate real human messages.

Imagine texting a friend (in a professional context). Preserve that style.

Include examples of messages you would send to embed your voice.

Example prompt for the first message:

Say hey, their name (if you have it), and make a comment about the page the person is currently on. Don't say "Need any assistance". 


Keep it to 11 words or less. Keep the tone casual. Lowercase the first letter in the sentence. Don't include any notes, disclaimers, or error messages. Don't include any quotations.


Examples of text to use:
1. hey Steve! Saw you're on our pricing page. LMK if you have questions?
2. hey Steve! Saw you're on our case studies page. Anything interesting?
3. hey Steve! Saw you're checking out our blog page. What do you think?

Always test in the sandbox environment to see what kinds of responses you'll get.

Consider the three levels of visitor reveal when creating the prompt

1. You know the person (~6-10% of site traffic)
2. You know the company (~60% of site traffic)
3. Unknown

Attributes you can personalize if you know the...

Unknown:

1. Most recently visited page URL

Company:

1. Company Name
2. Company Domain
3. Company Location
4. Company Description
5. Company Founded Year
6. Company Employees
7. Company Category
8. Company Industry
9. Company Sub Industry
10. Company Timezone
11. Company Employee Count
12. Company Tech Stack

Person:

Everything in Company plus:

1. Name
2. Email
3. Role/Title
4. Company Name
5. Company Description
6. Location
7. Bio
8. Seniority

These attributes are typically sourced from LinkedIn

Consider the attributes of the person the AI is reaching out on behalf of when creating the prompt

We also feed in context about the person the AI is reaching out on behalf of when sending the message that you can use when writing the prompt.

1. Name
2. Email
3. Calendar Link
4. Role/Title
5. Company Name (your company)
6. Company Description
7. Location
8. Job Description

These are a combination of attributes sourced from LinkedIn as well as inputted by the person in Settings

Additional FYIs

• You have 15,000 AI-generated messages a month. Once you run out the static fallback messages will start to kick in. You can monitor this in settings

• We don't yet pull in CRM fields. We're working on this

Let us know if you have any more questions!