Building an AI Agent for PO Verification | Domo BUILD 2026

Good morning, everyone. I'm Amanda Hall, Chief Financial Officer of Massman Companies, and I want to start with a question. How many of you spent part of your day doing work you know a computer should be doing? This might be copying information from one system to another, reconciling spreadsheets. Maybe it's reviewing documents by line. Now I have a confession to make. In my entire career, I've never met a buyer who woke up excited to reconcile purchase order confirmations. Nobody dreams of comparing PDFs to ERP records all day. Yet companies everywhere pay highly skilled people to do exactly that. And as CFO, I don't worry about whether our people are working hard. I know they are. What I worry about is talented people spending time on work that does not create value. And that's where our mantra came from: automate what you hate.

At Massman, we help manufacturers automate packaging lines. We build machines that put products into cartons, cartons into cases, and cases onto pallets. Every day we ask our customers, why are people still doing this manually? It turns out that's a pretty good question to ask ourselves too at Massman. One of those processes was purchase order confirmation review. A purchase order confirmation is sent by a supplier to confirm details such as pricing, quantities, delivery dates, and shipping information. The buyer then has to reconcile that confirmation against the original order to ensure everything matches. It's important work, but important doesn't necessarily mean manual. We had highly capable employees spending valuable time validating documents instead of solving problems, which led us to the simple question, what if AI could do it for us?

That question became a hackathon project that we affectionately named the PO Confirm A Bobulator 3000. In hindsight, we should have thought of an easier name to say, so meet the PO Confirm a Bobulator 3000. This tool uploads supplier confirmations, extracts the relevant information, compares it against NetSuite records, identifies discrepancies, and prepares approved transactions for processing. Rather than requiring a buyer to review every confirmation, the goal is to create an exception based workflow where people focus only on the records that need attention. And as well noted, based on our current transaction volumes, we estimate this could save 2400 hours annually, roughly the equivalent to, you know, 1 to 1 1/2 full time employees while improving processing speed and reducing manual effort. The most impressive part of this is we did this in less than 24 hours.

So why is this important? As CFO, everyone expects me to focus on the savings. And yes, saving 2400 hours annually is meaningful. That's real capacity that we've returned to the business. But that's not what excites me most. The second benefit is speed. Instead of discovering discrepancies days later through manual review, issues can be identified and addressed much earlier in the process. This means faster resolution, fewer surprises, and a better experience for both our teams and our customers. And the biggest opportunity is really what can we do with the time that we get back instead of spending hours validating documents. Our people can focus on supplier relationships, exception management, problem solving, and decision making. The goal isn't simply to save time. The goal is to redirect time towards higher value added work. And so that's really what automate what you hate means. Let technology handle the repetitive work so that people can focus on the work that creates value. So that's an overview of the tool that we built. We brought David Sawyers from the Domo team here to help look under the hood and tell us how we can make this happen.

Thanks, Amanda. Hi everyone, my name is David Sawyers. I'm a principal technical consultant here at Domo as well as a member of the AI solutioning and architecture team. Amanda just walked you through what this tool does for the business. I get the pleasure of spending a few minutes explaining the how, specifically the Domo building blocks that make it work. Because every single piece of this application is a native Domo offering, there's really four moments where the magic happens.

The first is on file upload. Everything that happens inside this custom app is built with Domo's Pro Code asset library and acts as the user interface for the whole solution. When someone drops a vendor PDF into the app, it goes straight into a Domo document collection, Domo's native file library. The moment it lands there, a Domo workflow fires automatically. That workflow is going to be the orchestration engine for everything that follows.

The second is extraction. The workflow hands that file to a Domo AI agent built inside Domo's AI library. That agent's first job is to read the PDF using an image to text tool. It reads every page and pulls out all the structured fields including PO number, expected delivery date, ship to address, and every single line item including quantities, rates, amounts, and part numbers. This allows for a self correcting behavior. It's not pattern matching against a template. The agent is actually reasoning through the document the way a person would and it knows when something looks off.

The third is comparison. Once it has those values, the agent runs a SQL query against a Domo data set that's synced live from NetSuite. In this instance, it's going to be our source of truth. It compares every field line by line. Any quantity mismatches, rate differences, or part number discrepancies are surfaced and flagged with a severity level. It writes the full comparison result to Domo's AppDB, which is what the app reads to display the variance report. This is served back to the user that's ingesting those PDF files.

The fourth is vendor communication. When the reviewer is ready to reach out, they trigger an email workflow. Domo's AI text generation drafts the email automatically based on the specific discrepancies that were surfaced. A human reviews it, approves it, and then Domo sends it, keeping a human in the loop. But all the heavy lifting is already done. And just remember, this entire solution was built over about a 24 hour hackathon. So the full stack you have is an asset library, app, document collection workflow, an AI agent with image to text and SQL, AppDB, and AI text generation for the outbound communication. All of these are native Domo offerings, and all of these features allow you to automate what you hate.

‍