How I AI: Reid Robinson's Zapier Workflows for CRM Automation, Meeting Prep, and Feedback Loops

In this episode of How I AI, Zapier product manager Reid Robinson demonstrates Claude and Zapier workflows for CRM updates, meeting preparation, and a reviewed internal knowledge loop.

MCPs are still one of the most overcomplicated ideas in AI. Reid has a much simpler framing: they're simply "app integrations for your AI tools." It's about giving your favorite AI, like Claude, access to the knowledge and actions trapped inside the thousands of apps we use every day. He walked me through how he uses that access to automate tasks he genuinely hates, like CRM updates and back-to-back meeting prep.

One of the most useful workflows we covered creates a feedback loop between customer support conversations and an internal knowledge base. Every interaction becomes fuel for better answers the next time around, with a human review step before anything gets published. Reid also shared a framing I loved for discovering automation ideas: "If you could run ChatGPT in your sleep, what would you do?" That question alone unlocks a surprising number of workflows.

automating CRM updates with Claude and Zapier MCPs

Automate CRM Updates with Claude Projects and Zapier MCPs: Turn meeting notes into reliable CRM records by giving Claude a tightly scoped Zapier MCP toolset and project instructions that define lookup order, field meanings, and when to create or update a record.

Almost nobody likes updating their CRM. Whether it's Salesforce, HubSpot, or a custom solution in Coda, it's tedious work that somehow never goes away. Reid built an agentic workflow that turns most of it into a copy-paste task by combining Zapier's MCP server with the instructional layer of Claude Projects.

building a custom toolset with Zapier MCP

Reid starts by using the Zapier MCP server to create a custom tool collection for Claude. Instead of wiring up tools one by one, Zapier bundles actions from more than 8,000 apps into a single secure connection. For this CRM workflow, his toolset includes actions for:

• Coda: To search and update his project-specific CRM docs.
• Google Calendar: To look up meeting details.
• Slack: To find related internal conversations.
• Evernote: To log notes in specific notebooks.
• Glean: To search across internal company knowledge.

What I like about this setup is that the tool collections are reusable. Reid has different MCP toolsets for different jobs, like sales workflows versus personal tasks, and can connect each one to different AI clients through a single URL.

the Claude Projects "secret sauce"

Here's the part that really stood out to me. Many people use Claude Projects to provide knowledge or context, but Reid uses it to provide instructions on tool usage. He created a project specifically for his CRM updates that tells the model how and in what order to use the tools from his Zapier MCP.

Detailed Claude Project instructions make the tool sequence more predictable when several actions have similar names. Reid still reviews the proposed updates before accepting them.

For example, his project instructions might specify:

1. First, search the Coda document to see if a record for this contact already exists.
2. If a record is not found, use the internal lookup tool to find more information about the person.
3. Based on the meeting notes provided, create a new entry in Coda with fields for Next Steps, Opportunity Details, etc.
4. Populate the fields according to these specific definitions...

putting it to work for post-meeting notes

Once everything is configured, the workflow itself is refreshingly simple. After a meeting, Reid pulls in a transcript from a tool like Granola. He opens Claude, selects his "CRM Logging" project, and drops in the notes. Claude follows the project instructions, accesses the Zapier MCP tools, finds the right contact, summarizes the meeting, and updates the Coda database. A task that used to take 15 minutes becomes a quick paste-and-review flow.

the "always-on" meeting prep assistant

Create an Automated AI Meeting Prep Assistant with Zapier: Prepare for customer interviews automatically by triggering on the right calendar events, gathering product and CRM context before the meeting, and placing a concise source-linked brief directly in the notes page.

The first workflow is agentic and conversational. This one is much more deterministic. Reid uses a traditional Zapier workflow to handle meeting prep in the background so he stops showing up to customer calls with zero context.

triggering the research

The workflow kicks off whenever a new customer interview lands on his calendar. Because it runs asynchronously, it can handle longer research and processing steps that would feel clunky inside a live chat interface.

fetching data with the right model

Once triggered, the Zap runs through a series of lookups:

1. It takes the guest's email address and queries an internal lookup tool built on Databricks to pull company usage data, past sales interactions, and related context.
2. The output from this tool is an HTML file, which Reid converts into a format that's easier for the model to process.
3. He then sends that file into a step using Google's Gemini model. I've now heard this from enough guests that I think it's just true: Gemini is unusually good at handling files, whether that's PDFs, audio, video, or converted HTML like this.

Reid chose Gemini specifically because it handled the file-heavy context more reliably and produced cleaner summaries than the other models he tested.

delivering the pre-meeting brief

The final Zap step appends Gemini's summary directly to the Coda page for that customer interview. By the time Reid opens his meeting notes, the background research is already there. No tab frenzy five minutes before a call.

building a self-improving customer feedback system

Build a Self-Improving Customer Feedback Knowledge Base: Keep a support knowledge base current by mining resolved conversations for reusable answers, checking for existing coverage, and publishing only approved, source-linked FAQ updates.

This next workflow is a perfect example of my challenge to teams: "What would your perfect team with infinite time do?" A perfect team would analyze every single customer interaction, identify knowledge gaps, and update the help docs immediately. With AI, that's now possible.

Reid built a feedback loop that turns customer conversations into a constantly improving internal knowledge base and chatbot.

the analysis Zap

The process starts with a Zap that triggers whenever a support ticket closes or a chatbot conversation ends. The workflow analyzes the transcript and does two things:

• It identifies the core question or issue the user had.
• It extracts the solution that was provided.

the human-in-the-loop review

Next, the AI checks if this question-and-answer pair is already covered in the existing knowledge base (which is stored in a Zapier Table, similar to a Google Sheet). If the knowledge base is missing this information, the Zap doesn't just automatically add it. Instead, it proposes a new FAQ entry and sends it to a human review step.

This is the part I think a lot of teams skip too quickly. Reid doesn't automatically publish new FAQ entries. Instead, he gets a notification to review the proposed answer, edit it if needed, and approve it before it enters the knowledge base. That human checkpoint keeps the quality high while still automating most of the work.

closing the virtuous cycle

Once approved, another Zap adds the FAQ to the primary Zapier Table that powers an internal chatbot for teams like sales and PMM. Because the system continuously learns from real customer conversations, the chatbot gets more useful over time without requiring someone to manually maintain documentation every week.

AI for personal joy and productivity

The personal workflows were some of my favorite parts of this conversation because they felt genuinely useful, not just clever demos.

The family calendar assistant

Reid solved the classic problem of keeping a physical kitchen calendar synced with Google Calendar using a surprisingly practical workflow. He takes a photo of the physical calendar, sends it to Claude, and uses a dedicated Claude Project with instructions like:

• Add events to our shared family calendar.
• If an event is at my son's school during business hours, automatically block off driving time to and from the location on my work calendar.

Claude then uses the Zapier MCP to handle the find, update, and create actions in Google Calendar. It's a very good example of AI solving an annoyingly human logistics problem.

Interview prep as a love language

When his wife was job searching, Reid used NotebookLM to generate personalized interview-preparation audio. He supplied the company career page, job description, and competitor material for each interview.

You are preparing Anna for this interview. Make sure it's specific to Anna...

The tool generated a custom audio briefing she could listen to before each interview. According to Reid, she was repeatedly told she was the most informed candidate in the process. This may genuinely be one of the more convincing examples I've seen of AI as a love language.

Making music with your kids (and Suno)

Reid has also been using Claude with the music generation tool Suno to make songs with his four-year-old son, Leo. He prompts Claude with the day's events, including the required poop and fart jokes, then feeds the lyrics into Suno for a full song. What started as a fun family activity has also turned into a surprisingly effective way to teach older kids how prompting works.

How Reid chooses the workflow

Reid uses MCP tools for interactive work inside Claude and conventional Zaps for slower, predictable background processes. The distinction is based on how much judgment and back-and-forth the task needs.

A useful starting question is what a well-resourced team would do after every meeting or customer conversation. Automate the repeatable steps, keep approval in front of consequential writes, and inspect where the workflow fails before expanding it.