Building AI Agents for Post-sales Workflows

Introduction

Your department is stretched thin, every CSM is handling dozens of accounts, and you know you need more hands on your team. Unfortunately, adding headcount doesn’t seem to be anywhere on the horizon. So what do you do?

You get a digital teammate. Or multiple. And it doesn’t require you to go argue with C-suit for more budget. Because today, anyone can build an AI agent with a little bit of preparation. In Module 1, we covered some basic chat-based agents like the “Simulation Agent” and the “Expert Agent”. Now that you’re more familiar with AI, we’re going to dive deeper into building advanced agentic workflows in this module.

These AI agents are intelligent partners that work across tools, think and act alongside you. They can monitor account health in the background, spot early signs of churn, and trigger follow-ups before issues snowball. They can personalize onboarding, guide customers through underused features, and even prepare QBR summaries while you focus on strategy.

While there are many different tools you can use to build AI agents, in this course we will be using Zapier. This is because it’s a simple and intuitive tool that anyone can use. However, if you’re a bit more advanced when it comes to automations, you can apply the same principles and logic flows to build these agents in tools like Make or n8n.

In this module, we will cover:

• The critical factors to consider before building an agent
• How to use Zapier to build virtual teammates to take the busywork off your hands
• The step-by-step process for building three practical agent workflows for Customer Success
• A library of agent ideas that you can try out for yourself

The best part of building your own agent is that you can customize it for the use case you want. Which means, after getting familiar with the fundamentals in this module, you get to automate countless workflows that help you do more with less!

Chapter 1: Before You Build

Just like with prompts, we have to make some preparations before we get into the actual process. A really successful agent is the result of thoughtful setup: defining what the agent should do, gathering the right data, and understanding the builder environment you’ll be working in.

Start with a problem, not a tool

Before we start actually fiddling around with an agent builder, it’s important to pause and think about why we’re building an AI agent in the first place. 

The most effective AI agents are built to solve a specific, recurring problem in your workflow. Instead of asking, “What can I build?”, ask, “Where am I losing the most time or context as a CSM?”

Maybe your onboarding process feels inconsistent across accounts. Maybe follow-ups after calls take too long to write. Or maybe it’s the constant scramble to find customer updates before a renewal meeting. These are the kinds of repetitive, insight-heavy tasks where an agent can be impactful. 

A good starting point is to define your goal in one clear sentence. Ideally, it should be something measurable and practical, like:

“Reduce time-to-value for new customers by proactively identifying and addressing common onboarding friction points.”

Once you’ve articulated your “why,” the rest of the process becomes easier.

Scope your agent’s role

Where many people go wrong is when they try to build an all-in-one agent that handles everything. From support to expansion to onboarding, you name it. But an agent that tries to do too much often ends up doing nothing well.

Instead, start small. Choose one use case and design an agent that does it exceptionally.
For example, a Proactive Onboarding Assistant might:

• Analyze questions from new users
• Categorize those questions by onboarding stage
• Provide helpful, encouraging responses with links to relevant knowledge base articles
• Escalate high-frustration or blocker cases to a human CSM

That’s it. One focused workflow that gives you clear value from the very start. 

Gather your resources

An AI agent is only as smart as what you feed it. Before you begin building, collect:

• 5–10 real examples of customer questions or tickets related to your use case
• The corresponding answers or playbook snippets
• Links to your help center or internal documentation
• Any other material you feel is important for the agent you’re building to know

These materials act as your agent’s “curriculum.” The clearer and more structured they are, the better your agent will perform.

Choosing your agent builder

There are now many platforms that let you build AI agents without writing a single line of code. OpenAI’s Agent Builder, Anthropic’s console, Microsoft Copilot Studio, and Zapier Canvas are some of the go-to tools nowadays. 

They all follow the same core principles, even if the interfaces look different. Each one typically includes three key layers:

• Instructions: Defining the agent’s role, tone, and decision-making process.
• Knowledge: Supplying it with the right files, data sources, or FAQs to ground its responses
• Logic and Actions: Setting the conditions and integrations that allow it to act, like sending an email, logging data, or triggering a workflow.
  

For this module, we’ll use Zapier as our example because it’s one of the most accessible and easy to use tools available today. It’s designed like a canvas, so you can drag, drop, and connect nodes to build logic flows while seeing how the agent behaves in real time.

But the goal isn’t just to master Zapier’s interface, it’s to learn the process behind building agents. Once you understand how the moving parts work together, you can apply that knowledge to any platform of your choosing. 

What if I don’t have the access levels required to connect systems with each other?

Not all CSMs have the correct permissions to build the agents we present in this module, especially if they belong to a bigger organization. But not to worry! You can use the content we cover here as solutions that you can bring to your manager or other teams to launch a collaborative project to improve AI adoption in your organization.

For tips on how to do this, stay tuned for our 4th Module on Building an AI powered CS org.

Now that you’re prepared to build, let’s make a move from theory to practice.
In the next chapter, we’ll open up Zapier and walk you through its interface so you can see each option, setting, and configuration step-by-step.

Chapter 2: Introduction to Building AI Agents In Zapier

If you use tools like Google Sheets, Slack, Hubspot, or Gmail, Zapier can make them talk to each other. And it’s really easy to use! 

This chapter will walk you through some of the basics of Zapier, and get you started on the foundations you need to build your own agents. 

How Zapier Works (The Only Three Concepts You Need)

1. Triggers - The event that starts your workflow.
   
   • Example: “A new row is added to my Customer Tracker Sheet.”
     
2. Actions - What happens after the trigger fires.
   
   • Example: “Send an email draft to the assigned CSM.”
     
3. Filters and Paths - Conditions that decide whether to continue or branch the workflow.
   
   • Example: “Only continue if health score < 70.”

In each action step, Zapier automatically shows you fields you can “map.” That means you can insert dynamic data (like Customer Name or Email Address) pulled from earlier steps.

Useful Tools Inside Zapier

Zapier Copilot

Zapier Copilot makes Zapier even easier to use, allowing you to use natural language to handle the bulk of the designing. It’s as simple as describing what you want to automate, and Copilot will generate a starting workflow for you.

How to use it:

1. When you’re in the Zap editor, click the Copilot icon in the right-hand sidebar.
2. Type your goal in natural language. For example:

“When a customer completes onboarding in Google Sheets, send a thank-you email and add them to a mailing list.”

1. Copilot will propose a full multi-step Zap that you can review and edit.

This feature is relatively new, so don’t expect it to be perfect. It’s good to build up your own knowledge of how agent building works so that you can fix any of the issues that might show up in a Copilot-built workflow. 

Pro tip: If you’re not sure how to structure your automation idea, ask Copilot to “explain” or “suggest” rather than “create.” It will outline options before building; this is a great way to learn Zapier’s logic while staying in control of your design.

Zapier Templates

Zapier Templates are pre-built workflows designed around common business use cases. They can be a massive time-saver, so make sure to check if there are templates that are similar to your own use cases. 

How to access them:

1. From the Zapier home page, click “Templates” in the left-hand sidebar.
   
2. Use the search bar to find relevant templates. For example, “customer onboarding,” “churn alert,” or “meeting follow-up.”
   
3. When you open a template, Zapier shows you each step, the apps involved, and what triggers or actions it includes.
   
4. Click “Try it”, connect your own apps, and adjust any fields or filters.
   

Templates are a great way to learn by example, since they show you how successful Zaps are structured.

What’s next

Now that you have a good grasp of how Zapier works, we’ll start using it to create real Customer Success agents.

In the next chapter, you’ll see some of the basic principles of Zapier applied in practical Customer Success use cases. 

In this chapter, we’re going to walk you through step-by-step on how to build an Agent that lets hands off deals from sales to Customer Success. You can follow along by opening Zapier and trying it out yourself!

Prompts

Transcript

 In this video, we are going to walk through building a hand of assistant for a customer success workflow. What this agent will do is when a deal closes in HubSpot, it'll automatically generate a summary of the deal, send a slack message, tagging the account owner and create a task in Asana. For a follow up, we are gonna start in the Zapier canvas.

First, we're gonna choose the trigger. This is what will start the workflow. In this case, we want the zap to trigger when a deal closes in HubSpot. So let's select HubSpot here, and it will choose the trigger as updated deal stage.

Next, let's hit continue here. We want to choose the pipeline that tracks deals. For deal stage, we want to select closed won.

This means the workflow will trigger only when a deal is marked as closed. Let's hit continue, and if you need to, you can test your trigger. Now we need to generate a summary of what happened with the account. We'll use AI by Zapier to generate a summary, pain points, and next steps based on the deal data.

Let's select our next action and select AI by Zapier. Here in the prompt field, we'll define what the AI should do. You can use a prompt similar to this.

Once this is done generating, it'll preview the data output of your step. Then you can hit finish here. Next, we need to send a message to Slack tagging the account owner. But first we need the Slack user ID to make sure we're tagging the right person. Let's add another action. Select Slack and make the action event "find user by email".

We'll select our Slack account and we'll hit continue. In the email section here, you can insert the data from your updated deal stage in HubSpot. Once you fill that you can continue and then add another action. Let's add another Slack action. In this action, we're going to choose "send channel message".

Here, we can pick the channel that we want to send the handoff message to, and we can add the message data from our AI summary. There's some additional settings like adding an image that you can explore. But for now, let's move on into adding a task in Asana.

Let's add another step and select Asana. The action event will be "create a task".

Here, you can choose the workplace that you want to use. You can add a task name by mentioning something like follow up on, and then adding the deal name through the "updated deal stage" hubSpot option. For task description, we can take the same AI summary.

There are a few more options that you can explore, but once you're satisfied with those, you can continue and test the step. Once you're happy with all the steps, you can test run the entire zap over here. This is what the final result will look like when I run the zap.

Those are the basics of making a handoff assistant in Asana, and we'll talk about a few more agents in the following videos.

In this chapter, we’re going to walk you through step-by-step on how to build an Agent that lets hands off deals from sales to Customer Success.

Output fields

nps_score

pricing

positive_points

negative_points

Suggestions

Prompts

‍

Definitions

Sentiment
The overall sentiment of the feedback (Positive, Negative, or Neutral)

top_themes
Most common themes and their frequency

Nps_analysis
NPS score analysis and trends

key_insights
Main insights and patterns from the data

recommendations
Strategic recommendations based on findings

executive_summary
Brief executive summary for presentation

Transcript

 In this video, we are going to build a customer feedback insight agent. This agent analyzes all the responses you receive on your customer feedback survey and compiles them all into a high-level report so that you don't need to manually read all the responses yourself. The Zap will analyze the feedback text using AI, extract insights, and then automatically insert those insights into a Google Slides deck.

That means you don't need to copy paste feedback into chat-GPT every time a customer submits feedback. Instead, this agent will give you structured feedback in a visual format that's ready for meetings like quarterly business reviews and product discussions. Let's set the trigger as a ‘scheduled by Zapier’ step.

The trigger event will be custom frequency.

Let's choose the frequency type as monthly. We’re going to set the interval to three. That will make the zap activate every three months.

You can set the start date and the time of day to whatever you like. Then we can go ahead and test the trigger.

Next, let's add a Google sheets step.

The action event will be ‘get many spreadsheet rows’.

For the spreadsheet, we want to take the Google sheets that's linked to your Google form. We are using a simple template for a customer feedback survey. Your actual customer feedback survey can be as detailed as you want. We’ve already submitted some test responses. You can view your responses in Google Sheets.

Here are the test responses in a Google sheet, and this is the Google sheet that you'll want to add in Zapier.

For row count, you can go up to 1,500 rows. Let's add a hundred just for this test.

The next step will be an AI step to analyze the feedback.

Let's add some input fields to give the AI some information from the feedback.

We can get the data for these input fields from the relevant columns in our Google sheets.

Now let's add a prompt that will allow us to analyze our feedback.

You also want to add some output fields, so the AI generates their insights in a format which we can use in our Google Slides.

Now we can generate a preview.

And then we can finish this step.

The final step is going to be a Google Slides step.

Let's have the action event as ‘create presentation from template’.

You can name the new presentation whatever that you want, and then we need to choose a suitable template. The template that we are going to be using is going to look like this.

The information that the AI generates is going to appear within these double curly brackets.

Now let's map the data onto the correct places.

Now we can test this step. As you can see, the analysis has been inserted in the correct places in the Google Slides deck.

This gives you a good foundation for a customer feedback slide deck. You can make whatever changes and refinements that you want from here. This is how you make a customer feedback insight agent in Zapier.

This chapter is a walkthrough for an agent that will help with your quarterly business reviews. Use it to make prepping and creating slide decks for QBRs automatic and accurate!

Prompts

‍

Transcript

 In this video, we are going to build a quarterly business review prep agent. This agent will automatically gather customer metrics from HubSpot, generate key insights using ai, and then fill those insights into a Google Slides QBR deck. The first step is to start with a scheduled trigger. Let's search for ‘Schedule by Zapier.’

Here, for the trigger event, let's select custom frequency.

For the frequency type, we're going to select monthly and we're going to set the interval to every three months. We can set the start date and the time of day to whatever you like.

Once that's done, you can test the trigger.

Now we are going to pull metrics from HubSpot. Let's add a HubSpot action.

For the action event, let's choose ‘Find deals’.

For the first property name, let's select deals closed won.

And for the value, let’s put ‘true’. For additional properties to retrieve, we can take the customer metrics.

The way your customer data has been recorded in your HubSpot account may be different. So be sure to check your HubSpot account to see what properties you need to be using. Next, we're going to generate insights using AI. Let's add an action, and select AI by Zapier. For the prompt you can use something like this.

You can bring in the relevant information from the HubSpot step.

Now we can generate a preview of this step. So this step will generate risks and concerns.

Let's add another AI step to generate discussion points.

Let's add a final AI step to generate key wins.

The final action is going to be a Google Slides action.

The action event will be ‘Create presentation from template.’

Let's give this presentation a name. And let's choose a template for this presentation.

My template currently looks like this.

The AI generated insights will be included inside these double curly brackets. Here in Zapier, we can start mapping the data onto the Google Slides step.

Now let's test this step and see what the new presentation that has been created looks like. As you can see, details of the company have been inserted. Here are the AI-generated summaries on different slides.

This is how you make a quarterly business review preparation agent in Zapier.

Chapter 6: AI Agent Library

This chapter offers a collection of different agents you can build in Zapier to automate high-impact Customer Success functions. 

1. Cross-Team Briefing Agent

This agent alerts Product and Sales automatically when an account becomes “At Risk,” giving them immediate visibility into the customer’s top issues.

Trigger

HubSpot → “Deal Property Change” → Set Property Name as the risk status property found in your Hubspot account

Steps

1. Add a Filter step to continue only when the account meets the condition: “Risk Status = At Risk.”
   
2. Add Zendesk → Find Tickets and search by customer domain, requester email, or external ID. Sort by most recent and limit to the latest 3 tickets.
   
3. Send these 3 tickets into AI by Zapier to generate a short, structured briefing summarizing the top issues.
   
4. Post a message into the Product channel in Slack using “Send Channel Message.” Include:
   
   • Customer name
   • Reason for At Risk
   • Link to tickets
   • AI summary (if used)
     
5. Add another Slack action for the Sales team. Include the same briefing and tag the AE based on the HubSpot owner email if you have email-to-Slack ID mapping in Zapier Storage or Google Sheets.

2. Account Sentiment Analyzer Agent

This agent performs an end-of-week sentiment review of all customer interactions and logs the trend in Notion for long-term visibility.

Trigger

Schedule by Zapier → Every Week (e.g., Fridays at 6 PM).

Steps

1. Trigger at the selected weekly time.
   
2. Pull recent call transcripts from Gong, Chorus, or wherever transcripts are stored (Google Drive, Notion, etc.). Use filters like “last 7 days” or match by account name.
   
3. Use Gmail → “Find Email” to fetch customer conversations. You can limit the search using a search string like:
   from:{{customer_email}} newer_than:7d
   
4. Pass transcripts and email text into AI by Zapier to produce:
   
   • Overall sentiment
   • 2–3 key themes
   • Risk level (low, medium, high)
   • Brief justification
     
5. Use the “Create Data Source Item” to make or update a page in a Notion database called Account Sentiment Trends, including the week’s sentiment, themes, and relevant links to calls or email threads.
   

3. Renewal Risk Dashboard Agent

This agent builds a daily, automated renewal risk score using Stripe payment signals and Zendesk support activity, and syncs the results into a CS platform.

Trigger

Schedule by Zapier → Daily.

Steps

1. Trigger every morning using a scheduled Zap.
   
2. Add Stripe → “Find Charges” or “Find Subscriptions” filtered to show payment failures or past_due states. Extract:
   
   • Number of failed attempts
     
   • Next retry date
     
   • Subscription status
     
3. Add Zendesk → “Search Tickets” and filter tickets created in the last 7 days. Count unresolved or high-priority tickets.
   
4. Use Google Sheets → “Lookup or Create Row” in a “Renewal Risk Staging” sheet. Store for each account:
   
   • Payment issue flag (yes/no)
     
   • Number of failed payments
     
   • Ticket volume (last 7 days)
     
   • Ticket severity
     
5. Use Formatter or Sheet formulas to calculate a simple risk level (for example: “High” if any payment failure + high ticket volume).
   
6. Send the final risk level into a CS platform using: Webhooks by Zapier → POST → CS Platform API endpoint to update health/risk fields on an account.
   

4. Churn Attribution Agent

This agent takes each churned customer, categorizes the churn reason, and logs it in Notion for long-term reporting.

Trigger

HubSpot → Deal Stage changed to “Closed Lost” or Company marked “Churned.”

Steps

1. Trigger whenever HubSpot updates the record to a churned or closed lost state.
   
2. Pull the churn reason using Closed Lost Reason, Churn Reason, or a custom property.
   
3. Add AI by Zapier to classify the text into a single churn theme such as:
   
   • Pricing
     
   • Missing features
     
   • Product fit
     
   • Support experience
     
   • Competition
     
   • Internal restructuring
     
4. Use “Create Data Source Item” to create a new page in a Notion database called Churn Attribution Log, capturing:
   
   • Account name
     
   • ARR lost
     
   • Churn date
     
   • Churn reason
     
   • Theme
     
   • AE and CSM

5. Churn Review Agent

This agent produces a full churn review package combining exit survey responses and recent call data.

Trigger

HubSpot → “New Contact Property Change”  

Steps

1. Use a filter to trigger the Zap once the churn property in Hubspot switches to true.
   
2. Use Google Sheets, Typeform, or your survey tool to look up the customer’s exit survey by email or account ID.
   
3. Pull the last 3 call transcripts from Gong, Chorus, Zoom, or Drive. Filter using date ranges or naming conventions.
   
4. Send both pieces of data into AI by Zapier to generate a structured churn review summary with:
   
   • Context
     
   • Primary churn reasons
     
   • Missed signals
     
   • Recommendations
     
5. Create a Google Doc or Notion document from that summary.
   
6. Email or Slack the Head of CS with the document link and a brief automated overview.
   

6. Onboarding Kickoff Agent

This agent kicks off the entire onboarding workflow the moment a deal closes.

Trigger

HubSpot → Deal moved to Closed Won or Lifecycle changed to Customer.

Steps

1. Trigger when the deal stage becomes Closed Won. Map account name, ARR, contact email, CSM, and start date.
   
2. Create an onboarding project in Asana using “Create Project from Template.”
   Fill in project title, owner, due dates, and onboarding milestones.
   
3. Use Gmail → “Send Email” to deliver a personalized welcome message to the primary customer contact. Include:
   
   • CSM introduction
     
   • Kickoff call scheduling link
     
   • Onboarding overview
     
4. Post a summary message in Slack to the CS team channel with:
   
   • Account name
     
   • ARR
     
   • Assigned CSM
     
   • Link to Asana project
     
5. Add onboarding milestones into a CS platform using Webhooks → POST. Include key dates such as kickoff, go-live, and first-value milestones.