Automating outreach with an AI-agent

Overwhelmed by 500 replies in your inbox? Spending days just on negotiations? Let the AI agent handle your creator communications!

The AI agent can automate everything from sending the first message to negotiating terms and making offers.

It works using Quick Actions - message templates for each step of your workflow. Then you just define the agent's behavior with simple "If this happens, then do that (using a Quick Action)" rules.

Here's how to create your agent:

1. First, create Quick Actions for all the tasks you want to automate (like responding to interested bloggers, requesting stats, or rejecting a creator).
2. Next, go to 'Creator AI Agent Settings' in your settings.
3. Click the 'Create AI Agent' button.

4. Give the agent a name, specify the stages where it should be active, and select the Quick Actions it will use.

5. Write out the instructions for how the agent should interact with creators.

Here’s an example of an AI agent's workflow:

So, how does this AI agent work?

1. It receives a reply to your outreach campaign.
2. The agent analyzes the conversation.
3. If the blogger is interested, the agent explains the next steps and asks for their stats (by triggering the 'Interested' Quick Action).
4. If the blogger isn't interested, the agent will ask for feedback (by using the 'Not interested' action) and move them to the 'Negotiation' stage.
5. Once an interested blogger provides their stats, the agent flags the conversation for your review and moves them to the 'Provided info' stage.

How to analyze statistics using an AI agent?

It is an experimental feature. To enable it, you need to:

1. Open your AI agent's page.
2. Add a requirement in the agent's instructions to request statistics from the blogger.
3. Add instructions specifying the criteria and conditions for making an offer to the blogger.

Here is an example of what this part of the instructions might look like:

How does this statistics agent work?

1. The creator says that they are interested in collaborating with you.
2. The agent asks them to share their audience statistics.
3. The creator shares screenshots with information about their audience.
4. The agent analyzes the screenshots and compares the data with the criteria stated in the prompt (geo, gender, age > X%).
5. If the stats are acceptable, the agent makes an offer using the ‘Offer’ action.
6. If the stats are not acceptable, the agent rejects the creator using the ‘Reject’ action.

Here is a more complex example of agent’s prompt for data analysis with a check if it includes everything requested:

In this example, we're asking the AI to analyze attachments with statistics (geo and gender distribution). If the stats fully meet our requirements, we send an offer. If there isn't enough data, we use a Quick Action to request the missing information. If the blogger refuses to send their stats, we thank them, explain that we can't move forward without the analytics, and politely end the conversation.

To request the missing statistics, we use the 'Incomplete Stats' Quick Action:

How to make sure your AI agent works as it should?

To test your AI agent you can use ‘Playground’ - a sandbox for safe experimentation with agents and quick actions. There you can simulate real-time conversations pretending you’re a creator, and see how agents answer your messages.

1. Open Settings.
2. Find Creator AI Agent Settings.
3. Click Playground.

1. Choose the agent you want to test.

2. You can change the agent’s settings on the left.

3. And you can change the test creator’s stats - followers, median views, ER, platform, name, bio, etc. - on the right.

4. Here you can also find a conversation summary - the agent’s chain of thought.

5. Write and send a message. You can also load an existing conversation from your conversations.

6. Click Test.

7. Now you can see the action the agent decided to take (see Conversation Summary) and the message the agent has sent.