Build an AI Custom Automation

This guide shows you how to design and set up an AI-powered automation in Planhat — from defining your goal and gathering data to configuring the AI step and deciding what actions to take with the output.

14 seconds

Planhat

Designing your AI custom automation

In this guide, we’ll walk you through our Framework to build a custom AI automation in Planhat. Each organisation will have their own specific needs, so use the Framework to create an automation that works for your business. **Step 1: Decide the outcome and goal** - Define the problem you’re solving or the outcome you want the automation to achieve **Step 2: Identify the inputs** - Identify the data that the AI need - Decide what should trigger the automation - Understand where to collect this data from **Step 3: Design the AI step** - Design the prompt - Select the right model **Step 4: Decide the actions** - Determine what you want to happen with the output **Step 5: Test and iterate!** Let's have a look at each of these steps in more detail.

Step 1: Decide the outcome and goal

Before setting up any automation, take a moment to clarify **what problem you’re trying to solve** or **what outcome you want to achieve**. This step is about defining *why* the automation exists and *what success looks like* before you start configuring anything in Planhat. AI automations can be powerful — they can summarize information, flag risks, identify opportunities, and even trigger follow-up actions. But without a clear outcome, they risk becoming clever experiments that don’t deliver real value. Start by asking yourself: - What insights or summaries would make a CSM’s or AE’s job easier? - What information do they usually spend time gathering manually? - What recurring questions or updates do they need before taking action? - Is the goal to save time, improve accuracy, enhance communication, or enable better decisions? First, define what you want to *happen at the end*. For example: **Outcome:** CSMs spend less time writing meeting summaries. **Automation goal:** Summarize customer call notes using AI and save the summary on the Company record.

Tip! You can use the Templated AI Automations in the App Center as inspiration for what you can do and build on them by using them as a starting point.

Step 2: Identify the inputs

The AI step is only as effective as the data it receives, so being deliberate about which data points matter is crucial. Think about what data the AI will need to successfully run. Some examples of valuable data points could be: - **Recent conversations:** Capture the most recent customer interactions to give the AI full context about tone, sentiment, and recent activity - **Health score trends:** Identify changes over time to detect early signs of churn risk or engagement improvement - **Open tasks or unresolved tickets:** Provide operational context on pending issues that may affect customer satisfaction - **Feedback or survey responses:** Enrich the AI’s understanding of customer sentiment or product perception Think of this as defining the **“inputs”** for your AI.

Now, to start building your automation you need to choose a **trigger** that defines when your automation should run. You can set the automation to start through a **Manual Action**, allowing users to trigger it directly from the object’s profile whenever needed. This is useful for on-demand actions like generating a summary after a meeting. Alternatively, you can trigger the automation based on **data changes** — for example, when a customer’s health score drops, a new note is added, or a field is updated. This makes the automation respond automatically to key events. You can also schedule automations to run **recurringly** (e.g., daily or weekly), though it’s important to note that each run will consume AI credits, so use recurring triggers only when the output needs to be refreshed regularly.

Now you need to determine where to collect the data you identified earlier from. By Using “Get” steps you can pull different information directly from Planhat. - You can retrieve data from various Models such as **End User**, **Company**, or **Conversation** - Configure filters to ensure only relevant records are fetched - You can get a **single** object (e.g., a specific customer record) or **multiple** (e.g., all active customers in a segment)

Step 3: Design the AI step

The AI step is the heart of your AI automation—where data is transformed into insights or actions. First, choose your **Provider** — this appears only if you have multiple connections (for example, your own LLM account and Planhat’s managed one). Next, pick the LLM **Model**. Different models come with different strengths and costs, so match the model to your goal — a lighter one for quick summaries, a more advanced one for deep analysis. Finally, write your **Prompt** — the actual instructions sent to the AI. This is the most important part: the clearer and more specific your prompt, the better the response. Use **dynamic references** (like a company name or recent notes) to make outputs personalized and relevant. We use the following framework for the prompt: - Role & Objective - Data set - Reasoning steps - Instructions - Context This mostly covers all the steps the AI need to give you a precise and targeted output, but feel free to experiment with this yourself.

Step 4: Decide the actions

Once your AI step has produced insights or content, the next question is: **what should happen with that output?** This is where your automation turns intelligence into action. You might start by **saving the output back into Planhat** so your team can easily access and build on them. For instance, you could have the automation **Create** a Document Page with a summary of a customer’s recent QBR, or **Update** a custom field like “Renewal Risk Level” or “Expansion Potential” based on the AI’s assessment. Keeping this data structured means it can be reused in future workflows or reports. You can also use the output to **notify your team right away**. Imagine the AI reviews customer notes and identifies signs of low engagement — your automation could instantly **send a Slack message** or **in-app notification** to the CSM, suggesting a follow-up call. Or if the AI spots strong product feedback, it might notify the Product Manager directly with the key points. Sometimes, you’ll want to **pause for a human checkpoint**. A **Wait** step can hold the workflow until someone reviews or approves the AI’s suggestion — for example, before sending an automated follow-up email to a customer. Finally, your automation can **Branch** intelligently based on what the AI discovers. If the AI labels an account as “high churn risk,” the workflow could trigger an escalation sequence - assigning tasks, alerting managers, or logging a renewal risk. If it instead identifies a “growth opportunity,” it might create an expansion task for the AE or draft a personalized success email for the CSM to send. Together, these actions transform AI outputs into meaningful, automated next steps - saving time, improving response speed, and making sure insights don’t just sit in a log, but drive real customer outcomes.

Step 5: Test and iterate

Now that you have all the building steps that together form an AI automation you need to start testing and tweaking. Try running the automation to check your outcome on a Test Company. Not fully happy? Go back and update the prompt until you are fully satisfied with the output, remember the more details in the prompt the more precise the output will be.

Now you know the framework on how to build a custom AI automation in Planhat! By first defining the **outcome** you want, then identifying the relevant **inputs**, designing the **AI step** and finally specifying the **actions -** you are setting yourself up for success! 👉 **Ready to get started?** Head into the App Center in Planhat and apply this framework step-by-step to create your AI-powered custom automations. Don't forget to check out our other guides with more custom Automations in our How-to-library, or enroll in the "AI step in Automations" course in the Dojo.