Unstructured → Structured // BRXND Dispatch vol. 060
On the real magic of AI.
You’re getting this email as a subscriber to the BRXND Dispatch, a (roughly) weekly email at the intersection of brands and AI. The next edition of the BRXND Marketing X AI Conference will be in LA on 2/6. If you’re interested in attending, we just put a few early bird tickets on sale for 15% off. If you’re interested in sponsoring, we’ve got a form for that as well. (Oh, and if you want to speak or know someone who should, let me know that too.)
Hi everyone. This week, I had two bits of content come out that I thought I’d share and build on. First off, I had the opportunity to be on Henrik Werdelin and Jeremy Utlety’s Beyond the Prompt podcast, and then, in a more marketing-specific context, I guest curated Faris & Rosie Yakob’s Strands of Genius newsletter. I'd encourage you to check out both, but I also wanted to build on a few themes that emerged across them.
I shared a story on the podcast about a recent dinner where I sat next to the CIO of a large media company. When he asked me how to articulate the value of AI, I responded by asking how much of his job as CIO involved making data from one place work with data from another. His answer: "Kind of all of it."
Arguably, that single problem is the one AI is best at solving (sometimes I joke it’s my “desert island” feature—if I could only use AI for one thing for the rest of my life, I would use it for data transformation rather than writing, etc.). The very first thing that wowed me about AI, where I thought, “I’ll never do this the old way again,” was extracting structured data from the text of a website. Here’s how I described that back in February:
One of my go-to AI use cases is data extraction. I first discovered this when I was building my Marketing x AI Landscape way back when, and I wanted to do price comparisons. I scraped each pricing page from each vendor and then sent it through the GPT-3 API (at the time) to turn it into structured data. This has legitimately opened up a world of use cases for me, and it’s something I do enough of that I’ve actually built a separate product that I might one day make available to others that has the sole purpose of taking a data schema and a URL and doing all the scraping and extraction.
Data extraction—turning unstructured data into structured data—is a subset of this data transformation functionality. The ability to turn any data into any other shape lies at the heart of AI's power. It’s not that you couldn’t do this before. It’s that it took a crazy amount of human effort and often led to brittle solutions. The canonical example here is web scraping. Just a few years ago, when I was busy building tools that parsed pages, I had to look at HTML and find a pattern that would consistently target some specific bit of content or button. If the people behind that site changed how they displayed that information, everything I built broke. Today, you just give it the whole thing and let AI do the work.
We interrupt this newsletter to offer a word of thanks to our sponsors for BRXND LA 2025. Airtable is operations for the AI era. Getty Images is a preeminent global visual content creator and marketplace with a commercially-safe AI platform. A huge thanks for their support. If you’re interested in sponsoring BRXND LA 2025, please be in touch.
Just yesterday, I had the opportunity to spend the day at an Airtable AI Accelerator event where they released some brand new functionality that lets you extract data from documents on the platform. The second half of the day was focused on building, and I made something I’ve been wanting for quite a while: an SOW to payment tracker.
The Problem
SOW gets sent with a payment schedule: You get paid some portion up front, some portion at a milestone, and the final bit on project completion. But how do you effectively track that? If you’re a big company, you’ve almost definitely got a solution to that problem. But I don’t. I’ve used Bill.com for years (mostly against my will) and recently switched to Mercury invoicing (that’s a referral link: I really like them). While that lets me track invoices, it doesn’t tell me when to send them. The challenge here is that you’re moving from a completely unstructured document—the final SOW is almost always a signed PDF—to something much more structured and easy to follow.
The first thing you do is upload the document. Airtable automatically sucks out the textual content of that, so you can work with it. From there, I do some AI work to extract a title, description, and the date of signature. (As an aside, I used ChatGPT to generate a bunch of fake SOWs to test with. If anyone wants to try this themselves, I put them all in a Drive folder you can download from.)
None of this is particularly complicated or impressive on its face, but to think about the challenge of how you would solve a problem like this a few years ago is pretty mind-blowing. Someone would have to take this doc and enter all this structured data into a system. Now, you just throw the final unstructured output in there and have the AI do the work. Not only are employees happy they don’t need to do this busy work, but the AI will always fill in exactly the fields you want it to. It’s a win-win.
The real magic comes in the next step, though. I had the AI extract all the invoices, along with some other structured data, and generate records for each one that needed to be sent.
Once all the invoices are in a system, we can easily track when they’ve been sent and received. My next step, which I started last night but haven’t quite finished, is to start pulling my bank transactions into Airtable (thanks again to Mercury, who have an API). The plan is to match transactions to invoices, so I don’t need anyone to track when the money has been received. (If anyone from Mercury is reading this, I would really love it if you added invoices to the API as well.)
Takeaways
A few things to take away from this:
Boring > Interesting: everyone wants to talk about the Coke AI commercial, and that’s cool. Go ahead and talk about it. But for my money, this sort of “boring” operational work is where people will see a ton of short- and long-term value in AI, and it will be most people’s entree to these tools.
Skip the Input Step: So much of modern work is asking people to fill out boxes in a software system. The vast majority of enterprise SaaS tools exist to solve that problem. One of the things I talk about in the podcast is that I don't quite understand what happens when AI starts to take on that work. As I said, the people are happier because they don't need to do this rote work, and the managers are happier because the boxes are always filled out exactly as they want them to be.
Data Integration is the New UI: While we've spent years focusing on making better user interfaces for data entry and manipulation, AI is shifting the paradigm entirely. The future isn't about building better forms—it's about seamlessly converting unstructured information into structured data, automating workflows, and connecting systems. This has always been the promise, but it never really worked because the data integration was so hard. Now it just kinda works …
If any of these ideas resonate with you, we’ll explore them further at BRXND LA on February 6th. Early bird tickets are on sale for another few weeks at 15% off ($650→$550). After two successful events in NYC, I'm excited to bring these conversations to the West Coast.
As always, if you're working on interesting AI projects or want to chat about any of this, please be in touch.
Thanks for reading,
Noah