If you scanned the QR code from the video, this is the page that goes with it. The video covered the big picture. This page covers the build — what the workflow does step by step, what you need to set it up, and a worksheet to plan your first run on your own products and vendors.
If you didn’t watch the video, read on. The page stands on its own.
An employee at your company sits down on Monday morning. They open one chat window. They type this:
The Prompt
“Go to our CRM Vendors module. Pull the top ten vendors we’ve bought 4K night vision security cameras from in the last twelve months, ranked by total spend across all purchase orders. For each one, pull the last unit price we paid. Flag anyone we haven’t ordered from in more than ninety days as stale pricing. Draft a current-quote request email to those vendors. Then check my calendar, find an open slot this week, and book a call with whoever offers the lowest current price.”
The employee gets up. Makes a coffee. Handles two emails.
Twelve minutes later, the agent has finished:
The workload that used to take an admin assistant most of a day. The agent ran it on data you already own — no web scraping, no copy-paste, no guessing.
Three technologies have to be wired together for this to run. Each one does a different job.
Google’s Gemini, Anthropic’s Claude, or OpenAI’s ChatGPT — whichever you prefer. The agent is the worker. It reads the prompt, breaks it into steps, queries your systems, drafts the emails, and decides what to do next based on what it finds.
Think of MCP as USB-C for AI. The cable that lets the agent plug into your business systems. Anthropic invented it. Google supports it. Zoho ships an official MCP server for the CRM. With MCP wired up, the agent can read from and write to your CRM as part of the workflow — query the Vendors module, pull purchase order line items, write the quote request as an Activity, log the call as a Meeting against the deal, all without anyone copying data by hand.
The system of record. Your CRM holds the Vendors module, the Products module, the Purchase Orders, and the historical pricing. The agent queries it. MCP moves the data. The CRM holds the result. When the agent finishes, the work product is sitting in the CRM — searchable, reportable, and ready for the next person on your team to act on.
Why this matters
Most agent demos start by scraping the web. That’s the part agents do worst — weird page layouts, anti-bot defenses, paywalls, stale data. Starting from your own CRM data flips the equation. You’re running the agent on the cleanest data in your business — the data you generated yourself. The web only comes in if you actually need to (Option 2 below).
The pricing comparison is the part most demos lie about. Let’s walk through it honestly. You have three options, each one a different trade-off between speed and accuracy.
You can run a rough version of this workflow today. A polished version takes a week. Here’s the order to do it in.
Pick your agent
Google Gemini (Pro or Ultra tier), Anthropic Claude, or ChatGPT. For the calendar-and-Workspace integration shown in the video, Gemini has the deepest native hooks. For most CRM-first workflows like this one, any of the three will work — what matters is that your agent can talk to MCP servers.
Connect Zoho’s MCP servers to your agent
Zoho ships four official MCP servers: Data Insights (read-only), Data Operations (read and write), Modules (schema), and Workflow & Automation. For this workflow you need Data Insights and Data Operations at minimum. Setup takes under thirty minutes for someone logged into both systems. OAuth handles authentication — your credentials never leave Zoho.
Check your data hygiene
Three things have to be true in your CRM. One — your Purchase Orders are linked to a Vendors module record. Two — your PO line items reference Products with consistent SKUs (no free-text product names). Three — line items carry a unit price field, not just a total. If any of these are missing, the agent can’t run the query reliably. Fix this before the first run.
Define your ranking criteria
“Top ten vendors” can mean three different things: highest total spend, most POs by count, or most recent activity. Decide which one matters for your business. The agent will follow whichever criterion you give it. For procurement decisions, total spend is usually the right answer because it weights for vendors you actually rely on.
Pick your pricing source approach
Choose Option 1, 2, or 3 from the section above. For your first run, Option 1 (historical pricing) is the fastest way to see whether the workflow even runs. Once you trust the basic plumbing, upgrade to Option 3 (hybrid) for real procurement work.
Run it once. Watch where it breaks.
The first run will not be clean. Maybe the SKU matching trips because two products have similar names. Maybe a vendor record has no email on file and the quote request fails. Maybe the calendar step trips on a recurring meeting. Write down exactly where it failed. That document is more valuable than the workflow itself — it’s your operating manual for the next six months of refinement.
Fix one thing. Run it again.
Don’t try to fix everything at once. Pick the most expensive failure — the one that wasted the most time — and fix that. Run the workflow again. Document the next failure. Repeat. Inside a month, this workflow runs without supervision and your team has gotten back half a day a week.
Fill this in before you open the agent. Five minutes of planning here saves an hour of fumbling later. Print the page or copy your answers into a doc — the form fields are for thinking, not saving.
The rule at ABR is no fabrication. Here’s the honest map of what’s available right now, what needs the paid tier, and what’s announced but not yet shipping to general users.
A note on the $100 question
The most capable version of this workflow runs on Google’s Ultra tier at $100 per month. That’s four to five hours of an admin assistant’s paycheck. If the agent saves your team a single afternoon a month, you’ve broken even. The real return is the operating experience you build before a competitor forces you to figure it out on a deadline.
Don’t try to deploy this in full production. Don’t bet the company. Experiment.
Pick one product category in your business — a part you order regularly, a service you renew, a consumable you keep restocking. Open the worksheet above, fill in your product and your ranking criterion, and run a version of this workflow through whichever agent you have access to. Use Option 1 on the first run. Watch where it stumbles. Write that down.
That’s the assignment. Thirty minutes of work, and you’ll have learned more about agentic AI in your business than ninety percent of business owners will learn in the next year.
When this workflow runs without stumbling — and it will, sooner than most people expect — your competitor will be Googling “how do I get started.” You’ll be on version four.