AI Can’t Fix Your Bad Feeds Alone—Here’s Why You Still Matter!

Episode Description

Think AI can clean up messy product feeds on its own? Not so fast. In this PPC Town Hall, Optmyzr CEO Frederick Vallaeys chats with DataFeedWatch founder Jacques van der Wilt about why AI-powered shopping campaigns still need human oversight.

They explore the challenges of AI hallucinations, the rising costs of automation, and the growing role of AI-driven search engines like Perplexity. If you’re running shopping ads, this conversation is packed with insights you won’t want to miss.

Episode Takeaways

Google’s AI-powered shopping campaigns promise automation and efficiency, but AI is only as good as the data you feed it. If your product feeds are a mess, no amount of AI can magically turn them into high-performing campaigns.

Here are the key takeaways from Fred and Jacques’ conversation on making AI work for feed optimization while keeping control and getting the best results.

1. AI needs good data to work effectively

Automation in feed management isn’t new, but AI is taking things to the next level. The challenge? AI relies on structured, accurate data to generate good results. If your product feed has missing attributes, incorrect pricing, or vague descriptions, AI won’t be able to fix those issues on its own.

“If someone searches for ‘Levi’s 501 men’s blue jeans size 32’ and that exact phrase is in the product title, Google has a high level of confidence that this is a good match. AI still needs structured data to work optimally,” Jacques explained.

The takeaway: AI enhances optimization, but advertisers need to ensure their feeds contain clean, accurate, and complete data.

2. AI plays a big role in feed optimization

AI can improve how product feeds are structured, filling in missing attributes like colors, sizes, and categories. But AI isn’t perfect, and human oversight is still required to validate the results.

Jacques shared how DataFeedWatch uses AI to optimize feeds:

• Automapping attributes based on store data
• AI-powered title and description rewriting
• Automated categorization for different ad platforms

However, advertisers should still review AI-generated optimizations to ensure accuracy and relevance.

“Anyone who tells you they’ve built a perfect AI system is lying. It’s always a combination of AI, A/B testing, and human oversight.”

 

- Jacques van der Wilt

3. The shift to visual and conversational search

Consumer behavior is evolving, and AI-driven search is changing how people find products. Instead of typing simple keywords, users are now using voice and image search, as well as AI chat assistants like ChatGPT.

Fred shared his experience researching an iPad: “I had a five-minute conversation with ChatGPT, explaining my needs, and it broke down all the options for me.”

Platforms like Perplexity are even working on AI-driven product search, where titles and descriptions might be rewritten dynamically to better match user intent.

“Perplexity will likely interpret product data in a much more AI-driven way than what’s currently being done on Google, Bing, or other platforms,” Jacques explained.

This shift means that feed optimization isn’t just about keywords anymore, it’s about structuring data in a way that AI can interpret effectively.

4. The cost of AI at scale

AI-powered optimizations aren’t free. When advertisers start applying AI to thousands of products, API costs can quickly add up. Jacques highlighted a key challenge:

“We had users with 500,000 products expecting instant AI-generated optimizations. But every call to OpenAI costs money, and at that scale, it becomes unsustainable.”

To manage costs effectively, advertisers need to:

• Choose the right AI models for different tasks (high-cost models for critical tasks, lower-cost ones for bulk processing)
• Prioritize AI optimizations where they have the biggest impact
• Test and refine AI-generated results before full deployment

5. AI is a tool, not a replacement for human strategy

The key takeaway from this discussion? AI is a powerful tool, but it’s not a magic fix. Advertisers still need to:

• Ensure their feeds have structured and accurate data
• Monitor AI-generated changes for quality and relevance
• Stay ahead of evolving search trends and AI-driven platforms

As Jacques put it:

“AI has no meaning in its own right. It’s just a tool to improve what we’re already doing.”

6. The future of AI in visual search & feed optimization

AI will eventually shift toward understanding images, but it won’t replace structured data overnight. Jacques emphasized that computing power is still a limitation when processing images at scale.

“Processing images with AI requires significant computing power. We’re moving toward AI understanding visuals, but it won’t replace structured data overnight.”

For now, structured product feeds remain the backbone of effective AI-driven search and advertising.

Get the most out of AI-Powered feed optimization with Optmyzr

AI can enhance feed optimization, but it works best when paired with human oversight and strategic adjustments. Tools like Optmyzr help advertisers take control of their shopping campaigns by providing:

• Custom feed audits to identify missing or low-quality data
• AI-driven insights for smart optimizations
• Automated rules to prevent bad data from hurting performance

Not an Optmyzr customer yet? Now’s the best time to sign up for a full-functionality 14-day free trial.

Thousands of advertisers—from small agencies to big brands—worldwide use Optmyzr to manage billions in ad spend every year.

You’ll also get the resources you need to get started, plus expert support from our team to answer questions and help you maximize your results.

Episode Transcript

Frederick Vallaeys: Hello, and welcome to another episode of PPC Town Hall. My name is Fred Vallaeys. I’m your host, and I’m also the CEO and Co-founder of Optmyzr, a PPC management tool.

For today’s episode, we have Jacques van der Wilt, the founder of DataFeedWatch. He started the company in 2008 and has a lot of insights on feed management and how it applies to e-commerce advertising on various platforms.

Data feed management used to be a highly manual process: lots of spreadsheets. Despite the automation that tools have introduced over the years, the rise of generative AI and its ability to work with structured data is making massive waves in how these processes are handled.

I can’t wait to hear from Jacques about how AI is changing feed management, how it’s impacting e-commerce advertising, and to leave all of you with some actionable tips and tactics you can apply to your campaigns.

So with that, let’s get rolling with this episode of PPC Town Hall.

Jacques, welcome to the show. Great to have you on!

Jacques van der Wilt: Fred, thank you very much. Thank you for having me.

Frederick Vallaeys: Yeah. And we’re going to do this episode in English, right? Even though both of us speak Dutch.

Jacques van der Wilt: Yeah, but we want other people to understand what we’re saying. So we’ll go English.

You guys will have to deal with my accent, which is like way thicker than Fred’s, who’s become a real American, but I’m sure we’ll get the story right in proper English.

Frederick Vallaeys: Yeah, that’s right. And if we don’t understand, there’s always generative AI and transcripts and translations as well.

Hey, but Jacques, where are you calling us from today?

Jacques van der Wilt: I am in my office in Amsterdam. Even though Datafeedwatch is an American company, most of us reside in Europe, we’re a remote-first company, and I’m running the company from Amsterdam, with having many people in Poland and several other countries, that’s how the world goes today.

Frederick Vallaeys: So a quick little anecdote. When I had left Google 12 years ago, I started my agency and I sold that agency to a company called SalesX, and they were in Foster City and I was working out of that office. And one day I needed some Data feed management help, and so I looked up online, and I found Datafeedwatch, and I looked at the address, and it’s literally the suite two doors down from the salesX office.

So I walked over, but unfortunately, you guys had already left Foster City and were at a different location. That’s how I first got to know Jacques because I was asking people, Where did everyone go? And someone mentioned, Oh yeah, Jacques is Dutch. So we had that connection right away.

But Jacques, talk a little bit about those early days. What made you start a company like DataFeedWatch?

Jacques van der Wilt: Yeah, that’s a great intro, Fred because Foster City brings back many memories. We didn’t start DataFeedWatch—we started WordWatch, which was an automated bid management system. At the time, Google only had search ads, text ads, and we ran WordWatch for a while. We got funded and all that, and my co-founder said, Let’s go to California and sell it. I said, Dude, we just raised money, we’re not going to San Francisco; it’s too expensive. That was our first Dutch moment; we were being frugal.

We ended up agreeing on Foster City. We first went to San Jose and spent a year there. WordWatch became a classic early-days internet startup story, excited founders raise a million euros, build a product, burn through most of the money in less than a year, and end up with something that wasn’t built for the masses. It was too niche, not as well-developed as it should have been, and ultimately, the company failed.

I ended up restructuring that company and thinking, Okay, what’s next? Around 2011–2012, Google introduced a new advertising format called Product Listing Ads. It was completely new, it had product images, titles, prices, and required something called a data feed. We saw an opportunity because we had customers interested in shopping ads through WordWatch, but they struggled with feeds.

The process was a nightmare. They had to ask their developers for a CSV file with the right data from their store. Before they could even launch a shopping campaign, they were months into development and tens of thousands of dollars deep. After a while, my team was frustrated. They said, This sucks.

So we asked ourselves, What if we built a way to create optimized data feeds? At the time, there weren’t many solutions, and the few that existed were expensive. Since we were a tech company, we figured, let’s just hack something together.

We built a rough prototype: ugly, basic, just three functionalities. But it worked. It connected to stores, downloaded data, let us rename and combine fields, and add static values. That alone solved 80% of the problems we had. We quickly realized, this is better than WordWatch.

So we pivoted. We improved the interface, added more functionality, and in spring 2013, we took it to market as DataFeedWatch. It turned out to be a much better business, and it’s been growing nonstop for the past 11 years.

Now, we have way more functionality, way more customers. About 40% are in the US, 40% in Western Europe, and 20% across Australia, New Zealand, and 50–60 other countries. The need is always the same: advertisers need data feeds for Google, Facebook, Criteo, TikTok, and other platforms. They need optimized feeds because the data in their stores isn’t great, and they need frequent updates, sometimes every hour, because inventory, titles, and prices change fast.

The core business has remained the same, but the features, organization, and support have grown. Even after selling the company to cars.com a couple of years ago, I’m still happy to be leading DataFeedWatch.

And that brings us to today’s topic: AI. I’m a big AI fan in general, spoiler alert. But the real question is, What is AI’s impact on data optimization, e-commerce, and the industry as a whole? That’s the next chapter. AI has given me new energy, new interest, and way more opportunities.

Frederick Vallaeys: Cool story. And so that makes two technology optimists on this call. Let’s see what kind of optimistic stories we can tell you.

As for AI, let’s start from the consumer perspective. Like you said, in the early days, shopping engines were pretty basic. They weren’t fresh and results would show products that were no longer available because there were no data feeds. Then data feeds were introduced, and suddenly, we had images with prices.

Now, we’re seeing AI overviews, tools like Perplexity, and all sorts of new ways consumers shop for things.

So I was recently looking for an iPad. But I didn’t need a business iPad. I just needed something I could use to watch movies on the plane that was a little bit bigger than my phone, and rather than going to the Apple website or going to technology review websites, I decided to just have a voice conversation for about five minutes with ChatGPT. I explained to it what I was hoping to use and it explained to me the different models and the pros and the cons.

And I thought it was such a fantastic experience. And so cool. Talk to us, talk to us about what is the evolution of consumer behavior in e-commerce, and then maybe how that connects back to what we do with feeds.

Jacques van der Wilt: Yeah, I think you already told the story and I think many consumers aren’t as deep into it as you are, every kid in school will use ChatGPT to write better texts and to find better answers and what have you.

And so consumers will increasingly enhance their own product search with way more than just googling the name of a product. So this is going to evolve. And the things that we’re seeing is, of course, that every organization that is like search or social related is seeing this consumer need and is developing like crazy to make it happen.

Yeah, I’m reading about that too. It’s clear that Google’s position as the number one search engine may be at risk for the first time. There’s a new technology that could overtake it.

That’s why Google is pushing Gemini, Facebook has Llama, and ChatGPT, at least for now, has been the 800-pound gorilla in the space.

But many other players are flocking to the market, and to take it back to data optimization, buying products, and e-commerce. I read today that Perplexity is planning something similar to a merchant center. They want retailers to provide product data, and Perplexity will likely interpret that data in a much more AI-driven way than what’s currently being done on Google, Bing, or other platforms.

We’re just scratching the surface and there’s an incredible iceberg below. AI is going to impact everything, especially eCommerce search, and it could completely redefine the industry.

Frederick Vallaeys: Yeah, I completely agree. AI is going to redefine quite a few things about how we work. But the Perplexity development is especially interesting, so let’s dig into that a bit more.

I think what I heard you say is that the feed we give to Google is interpreted quite literally, right? Your title is your title, your description is your description, and in most cases, those elements get rendered exactly as they are in the ad.

But with Perplexity, it sounds like it’s more of a framework that AI will interpret, summarize, and potentially enhance. It could adjust the title to make it more engaging or optimize the way information is presented to drive better results.

Is that what they’re aiming for? And if so, how do we adapt to that kind of environment, where we’re no longer operating with Google’s tight controls, where keywords are keywords and ad text is ad text, but instead providing raw data and letting AI shape it however it sees fit?

Jacques van der Wilt: I think if you send your data feed to Google, they will also use AI to interpret it. But of course, Google has a long history of developing this technology, leading to where they are today with Gemini. Having that experience and an established infrastructure makes a difference.

That also means Google and companies like Perplexity or Bing will have different paths when it comes to AI-driven search and feed optimization. The newer players jumped into the market with AI as their foundation, whereas Google has been refining its approach for years. That contrast will likely influence how each platform evolves.

As for how much AI will optimize product feeds, I don’t think that will be their priority. Search engines, whether it’s Google, Perplexity, or Bing, are fundamentally focused on interpreting the consumer’s query and matching it to the best possible result, whether that’s an organic listing or an ad. That’s where we’ll see AI make the biggest impact first.

For a long time, I’ve been preaching (and still do) that optimizing product titles in data feeds is critical. If someone searches for “Levi’s 501 men’s blue jeans size 32” and that exact phrase is also in the product title, Google’s algorithm will have a high level of confidence that this is the right match. That increases the chances of the ad being shown, leading to more impressions, higher CTR, and ultimately better conversions.

When a consumer sees exactly what they were looking for, they’re more likely to click, leading to a higher CTR and better conversion rates. That creates a direct link between data optimization and campaign performance.

With AI, consumers can express their needs in a more detailed and natural way. Instead of just searching for “new iPad,” as you mentioned, Fred, they might say, “I need something for the plane, bigger than my iPhone, but I don’t necessarily need an iPad.” By providing more context, they give the search engine a richer set of signals to work with.

AI will then be much better at interpreting those signals and determining the best product recommendations. That could mean responding directly to the user’s query or selecting the most relevant ads. Instead of simply pulling 20 potential matches, AI could intelligently narrow it down to the top five most relevant options and display those.

That benefits both the consumer and the retailer. The consumer gets a highly relevant result, making them more likely to return to that search engine, whether it’s Perplexity, Google, or another platform. And for retailers, better matches mean more impressions and higher engagement, ultimately driving more sales.

Frederick Vallaeys: Now, the example you gave about a Levi’s 501, that’s a very structured product. So if you look for a Levi’s 501, then clearly you need to find a Levi’s 501. I recently did a bathroom remodel at my house.

For example, when looking for cabinets, I could think of Ikea as a brand, but beyond that, I didn’t have a specific name in mind. Instead, my search was more about describing what I wanted: a panel with a slight texture, a modern shade of blue, and a specific type of countertop. It was a highly visual search.

Another experience that stood out to me was when I was editing photos of my family, turning us into avatars. My daughter wanted to see herself in the Barbie movie, so I uploaded her picture and asked the AI to make her look more like Barbie. Before making any edits, the AI first described the image and it was shockingly accurate.

It said something like, “There’s a girl wearing a pink t-shirt with a unicorn design that looks like a mix between a narwhal and a unicorn.” The level of detail was mind-blowing.

This brings me to a big question: how important is the product title in this new AI-driven world? If AI can analyze images and extract details like textures, colors, and unique design elements, will titles matter as much? Instead of relying on perfect keyword matches, will AI simply “vectorize” products, allowing it to understand and match items based on their actual characteristics, even if the title isn’t an exact description? Do you see this fundamentally changing how search works?

Jacques van der Wilt: Yo, Fred, I don’t think it’s really an either-or situation, I think you just described an evolution.

If we’re searching for a bedroom setup, we might also upload a few pictures of bedrooms we’ve seen from friends or collected on Pinterest. Or maybe it will all just happen on Pinterest since the images are already there. The search experience is only going to get more visual.

That said, I don’t think we’ll switch from relying on titles to visuals overnight. Processing images with AI requires significant computing power, it takes a lot of energy, expensive chips, and other resources. And with images, that demand will only increase.

Of course, we’re actively exploring ways to use images as a primary source of product information and extract as much data as possible from them. But even then, probably not everything can be derived solely from images. Sending hundreds of thousands of images to a server on the other side of the world for processing takes time.

So, I’m not saying it won’t happen; I’m just saying it won’t happen overnight. The tech teams will find ways to make it work faster, with less energy and lower processing costs. And once they do, I think we’ll eventually get to exactly what you’re describing.

Frederick Vallaeys: Yeah, no, that’s a really good point. Scale is a major factor here.

The experiments you run on the free GPT plan, or even the $20 GPT plan, are fun and interesting, but when you need to deploy AI across 100,000 products, you have to integrate it into an API. And then, depending on which model you’re using, you’re looking at costs of around $60 per million tokens. Those costs can add up quickly.

But one promising development is that while there’s a $60 per million token model, there’s also a $2 per million token option. As a marketer, part of the optimization process is figuring out which large language model (LLM) or vision model is good enough to deliver the results you need without dramatically increasing costs for only marginally better outcomes.

And that brings me to my next question, Jacques: which LLMs and vision models are your favorites? Which ones do you use? And what are some cool ways you’ve seen marketers deploy AI?

Jacques van der Wilt: That’s a lot of questions in one go, I don’t really have a favorite LLM.

Right now, I think ChatGPT is still delivering the best quality results. But as you mentioned, running it at scale can get very expensive. That’s why there are so many other language models that are almost as good but priced differently.

If there’s any arms race happening right now, it’s the race between LLMs. Companies are working frantically to improve their models, and every month, or every two months, we see new versions of nearly a dozen different LLMs, all getting better.

At the moment, interest in AI is split between its promise and its threats. The threats range from concerns about AI in warfare to issues like AI-generated images.

Going back to your example, I’ve created many images using AI by providing a description, and while the results were often impressive, there was always something off. Maybe the guy in the image had three eyes, or two hands on one arm, or he was sitting behind the wheel of a car but facing backward. AI hallucinations are still a big challenge.

That said, we’re just at the beginning of this journey. The pace of improvement will be incredible. In general, people tend to overestimate what AI can do in the short term but underestimate how much it will advance over time. We’re going to see that pattern play out again here.

Wait, what was the question again? Oh yeah, my favorite LLMs!

So I don’t have any.

Frederick Vallaeys: I agree with you. So I use mostly OpenAI’s ChatGPT. Not necessarily because it’s my favorite, but because I see all of them are leapfrogging each other. And eventually they all seem to come to the same place and the same capability. ChatGPT had one of the first solid APIs, so we latched onto it early.

That said, I do keep an eye on what other models are doing. If OpenAI doesn’t release a certain capability promptly, I might explore alternatives. But for anything we do at scale, OpenAI is still our go-to. But we don’t always use the latest models again, because they’re much more expensive.

And for tasks like keyword generation or ad text suggestions, it is not worth 30 times the cost for a slightly better ad when it comes to something like writing ad scripts, I do use the latest models like O1 because it is significantly better. O1 is the first version of GPT that has given me a single shot, correct answer with a single prompt, it actually writes a script that works without issues.

In GPT-4.0, I usually had to go back three or four times to fix errors before getting a usable script. That’s still faster than writing it manually, but in this case, the extra cost for O1 is worth it.

Jacques van der Wilt: Yeah, exactly. When it’s just one script, the cost isn’t a big deal. But when you need keywords for hundreds of thousands of products, that’s a different story.

This was actually one of the first challenges we ran into when developing our AI-powered feed optimization. We’d offer AI-optimized titles, descriptions, and other attributes, and people would sign up, connect a store with 500,000 products or even 10 stores with 100,000 products each and expect instant AI-generated optimizations.

But every product processed meant paying OpenAI a fraction of a cent. And at that scale, those costs added up fast. For a single signup, we could rack up thousands of dollars in API fees just to generate AI values for all their products.

Obviously, that wasn’t sustainable. No one could afford that, not us, not them. So we had to shift to a paid model. And honestly, I hated that. Because if you’re a real customer, someone serious about using the tool for years, not just trying to get free AI-optimized values and bounce, you’d end up getting hit with fees just because you have a large store.

Okay. Imagine getting access to a great tool with powerful AI optimizations, only to be told, “You’re on the free trial, but before you get any AI, you need to pay me $5,000.” That would be a terrible experience. I’d rather not offer AI at all than put customers through that. And the alternatives weren’t much better, forcing users to upgrade to a plan that costs $1,000 more per month just to access AI? That didn’t feel right either.

So even before we started building AI into our platform, we had this exact conversation about scaling. Our goal was to offer AI either for free (which we currently do) or, if that wasn’t feasible, for a reasonable cost.

That’s what drove our approach: how could we create our own AI ecosystem by leveraging our experience, smart development, cloud scaling, and existing technology to make this work? Because while we know AI costs will eventually come down, we don’t know when or how fast and I wasn’t going to sit around waiting for that to happen.

Frederick Vallaeys: I want to talk more about that, but it led me into another thought. Somebody at a conference recently asked me, with the plethora of AI Solutions that are out there. How do you choose one? And have the trust that tool is going to be around in a couple of years so that you don’t make an investment into something that’s going to fail and so where you led me with that question was sure, there could be an AI company that’s giving you all of these capabilities and a hundred thousand titles optimized. But like you said, they’d be losing a huge amount of money on that potentially.

So that might be a business that fails in a year or two. And now you’ve invested all your capabilities, your workflows in that. So how do you think consumers and advertisers should think about choosing AI tools when there are so many choices?

Jacques van der Wilt: It really depends. For starters, on the business that you’re in, if you need some kind of AI capability that you’ll have to like deeply integrate into your operations and maybe also in your processes, your software, and what have you. Then, imagine what happens if that company goes belly up, it’s like losing a limb or something for that company.

But if your business is, let me just speak for myself. If your business is data optimization, and you sign up for a tool that would give you like AI values for the various attributes that you need to optimize. Then it’s optimized, it’s in your Google or Facebook and would have you feed and nothing will change.

And then if your provider goes belly up, you’re going to have to find another one who will then optimize the values for the new products that you add to your feed. So the ability to switch around is probably very important, more important than anything else. Otherwise, it’s like you say you invest deeply in something.

We don’t know what’s going to happen at short notice with this new industry.

Frederick Vallaeys: So my take on that may be a little bit different is you’ve already invested in tools and workflows that your team is now using, that they understand, that they hopefully like. Many of these tools are probably adding the same AI capabilities that AI-only new players are coming up with.

And so the new AI players, they may do a better job of saying, hey, like we’re the newest, we’re the greatest, we’re the latest AI. But think of the switching cost and retraining everything your team is doing, right? Whether it’s data feed management, or it’s PPC optimization software, or it’s your customer management system, all of these things.

It’s just that there was this great quote, our tools use electricity. Our tools use AI, all tools use AI at this point. They will use AI at some point. So you don’t have to go look at the new tool. You should talk to your existing vendor and communicate with them. Oh, maybe I saw this great Demo from someone and they were able to do this and, oh, it’s just simple integration with Open AI.

We’ll do that too for you. And then you don’t have to switch up all your workflows.

Jacques van der Wilt: Yeah, no, that’s a very good point there, Fred. Yeah. And that was also definitely part of a thinking we’ve never looked at AI as something separate. Oh, now we have AI, sounds good. This is the day and age, but it means nothing.

It’s 30 years ago, people would say I’m on the internet. Gee, congratulations, right? And AI is going to work the same way. Like the internet, it’s just a network that connects everything and makes everything easier. AI has no meaning in its own right? So you want to use it as a provider, like we are, we want to use a new technology that happens to be called AI to improve the stuff that we’re already doing.

And I think that there are plenty of tools that are doing it like that. And, maybe those are a safer choice than a new kid on the block that says, Hey, I’m new, we got founded like three months ago, and now we use AI to do everything that you’ve been doing with this other tool for the past five years, I’m not saying that, that can be done, but if it’s embedded, you can.

You can continue with everything you already have and have embedded in your organization.

Frederick Vallaeys: That’s probably a safer choice. So DataFeedWatch then, it sounds like you ended up building your own AI capabilities. Tell us a little bit about what’s cool about AI in DataFeedWatch. What does it enable advertisers to do?

Jacques van der Wilt: Yeah. I have a little story about how, particularly when I got started with AI. I got started like five, six years ago with AI. You know why? I was on roll with Daily Feet Voyage. Everything’s going well.

And then, I believe that many founders are suffering sometimes from this. The thing is going so well and we’re growing so fast and making so much money. What if I lose it all? And then I figured, okay, so what would it take for me to lose like everything? And I thought about that and okay, yeah, we have a data feed optimization tool.

It’s very user friendly and intuitive. People come in and they optimize their feeds, they do the Levi’s 501, blah, blah, blah. All good, but it’s still work. And the customer still needs to think, okay, if I create a title for apparel, what needs to go in there? So then I figured, okay, if there would be a new daily feed tool tomorrow, that would just say, click here and your feed is optimized and ready.

Then I’m like, okay, we’re dead. Because then everyone would go there because it would be like the perfect feed. It would take three clicks. I wouldn’t have to think, I wouldn’t have to work. I wouldn’t have to click and it would be done. And then I figured, okay, let’s elaborate on that.

If that’s the kind of company that would, you know puts out a business, I should become the company and start sketching, like literally wireframing how that should work. The AI embedded AI functionality that we have in Daily Watch today, I actually designed five or six years ago, coming out of fear, basically.

And, maybe it’s not literally the same, but it’s more or less, it’s the very same principle. And then of course, I’m like, oh, this is great. I’m going to be that company and we’re going to get to work. And I hired new people and man, it was so difficult. We did manage to get some basics in place, but yeah, after two years, I abandoned it.

Yeah. And then, Open AI, GPT came along, like the time has come right now. Now’s the time we’ve got to do this. Yeah. We got to work right away. And like now I was able to realize the vision that I had five years ago, so that’s cool. And that’s also the reason why I’m such an AI fanatic.

Frederick Vallaeys: Great advice. I think everyone listening today, whether you’re an agency owner or a business owner, think of what you’re worried about for your business or the grand vision of what you maybe at some point thought about achieving and it wasn’t possible at the time and see how generative AI can get you to that place, right?

Whether it’s an app that you’ve always wanted to build, it’s so easy to build an app now, so much faster than it used to be. Or I’ll get your branding. That has become much easier. So many things that are now possible. It’s no longer about saying I can’t, it’s about saying, I haven’t yet.

And now is the time for you to start.

Jacques van der Wilt: Exactly. If you’re a founder, if you run an agency, or if you run a technology company and you’re not at least exploring what AI can do for your business, then you should start worrying at night, right? The train has left the station and you better be on it or get on it soon.

It’s not too late, but the time is now. And to go back to your question, Fred, we also had to think about what exactly we needed to create. Our focus is enabling customers to build optimized data feeds because the better the feed, the better the data, and ultimately, the better the campaign performance.

So we’re going to keep doing that; that’s our core business, and it needs to be seamless. Even though AI is involved, we want to ensure the customer remains in control. So, where did we start? First, I had this vision: you should be able to click once, and it should just work.

We’ve expanded our automapping so that as soon as you connect your store, we scan everything in it. If you want to sell on Google, we automatically map every attribute Google requires from the data available in your source feed. And every field that can be optimized with AI is already optimized.

Then, we wanted to make it simple: you see your data, and you just click “save.” That’s your one-click optimization. We’ve put that in place.

But AI can do even more. We’re just getting started. One major challenge customers face is missing data. Say you have 100,000 products, and 10,000 of them don’t have a color value. If you’re selling apparel, Google will disapprove those products.

You can fix that manually by creating an Excel file listing colors and extracting values from descriptions. But that’s tedious, and you’re likely to miss a few. Until now, that was the best solution. But today, AI can automatically detect and fill in those missing values for you. We even provide a separate field for all the color values AI finds.

The same goes for sizes and other attributes; you don’t want to waste time on them. Just check the box, and you’re done.

Then there’s categorization. Who enjoys categorization? Nobody. Google has made it easier by saying, “Don’t worry, categories are no longer mandatory, we’ll do it for you.” They use AI for that. But for other channels, having the right categories is still crucial for better performance.

So, AI takes product types from your source feed and maps them to Google, Facebook, or any other platform’s categories. It’s a tedious job, but no worries, just click “save,” and it’s done for you.

Sometimes, descriptions are missing. And, with all due respect to our customers, sometimes they’re not well-written. Some are great, but many aren’t as they’re unclear, overly long, or just poorly structured. One thing AI does exceptionally well is rewrite content, pulling in additional information and making it more concise. That means much better descriptions.

Since we don’t handle images yet, let’s go back to titles, which we touched on earlier, Fred. So we created a system that, because it’s also our experience that matters, right? At the end of the day, humans do matter.

We built a system based on our expertise because at the end of the day, humans do matter. I have a team with 10 years of experience in data optimization. We’ve developed a framework for 200 product categories, where we know the ideal title format.

Then, we tell AI to identify the product and rewrite the title to include the right attributes in the right order. It also pulls in relevant details from descriptions to enrich the title. This is key to optimizing titles for better campaign performance.

Frederick Vallaeys: That all makes sense. So, there are some fields you absolutely have to fill out to avoid feed disapprovals. Then, there are optional fields, filling them in just helps ensure your products are better matched in the right situations.

And then there are fields like title and description. You already have them, but they might be suboptimal. AI can improve them, but how do you measure that?

Jacques, if your original description was a mess and AI rewrites it so that it sounds better when you read it: how do you actually know it performs better?

Jacques van der Wilt: What we don’t do directly is to take the titles that we optimize ourselves and then measure the performance, for example, in the Google campaign in the Google shopping campaign. Because there are so many things defining the campaign result. But we do enable customers to do A/B testing with different titles.

So they can A/B test their original title against the AI titles. If you do that massively, then you should start seeing at least signs or your analytics that, that is indeed happening. But we, other than measuring the actual impact in the campaign, we do use technology and humans, lots of them, to also basically check if those titles are indeed better than they were.

We have a system that we use AI, very original, to check the quality. There’s of course, always a risk that AI will say, hey, what I just created is absolutely the best, right? So yeah, I put in some objectivity there. But at the end of the day, we’re also using more than half of the people that work in my company to check titles and descriptions regularly to make sure that if it’s not perfect, that at least it’s not wrong.

And then if stuff is missing that we modify a system that in the future, it will not be missing. Anyone who will tell you that I got AI and I created the perfect system will lie. That’s not possible, or at least not this year or next year. But it’s a combination of AI, actual A/B testing, and old-fashioned humans to make sure that the quality of the AI-generated stuff.

Is it really in order? And if you want, yeah. Okay. Spoiler alert: now, what we see is that we do have customers that have written the perfect titles, right? So don’t use AI. It was already like really good. You knew what you were doing. So there’s no need to use it.

There might be a small chance that they will gobble it up and it will not be as perfect as you had in your mind. But in general, especially with lots of products, with thousands or tens of thousands of products it’s almost physically impossible to do that optimization title by title manually, and if you had the time, then you, I think you, the human that knows what he’s doing would still do a better job in most cases.

Frederick Vallaeys: And I think when it comes to these GPTs, you can also feed them examples of high performing headlines and descriptions that you’ve had in the past, and high performing is done measured by your KPIs and data out of Google ads that you give that as a data set, right? And you can say, write my new headlines.

More like the high performing ones and the system will automatically pick up on what it is about the structure, the grammar, the call to action. And then it’ll take that concept and apply it to different products, but come out with something that should presumably also carry then that higher performance by following the methodology that has worked in the past.

Jacques van der Wilt: And that’s basically how you’re training your own system to improve. Continuously.

Frederick Vallaeys: No, that’s great. So if people want to try it out, I will put this in the show notes, but tell folks where they can go for that trial. And if they want to talk more to you or hear what you’re thinking about, how do they follow you?

Jacques van der Wilt: Yeah. So you can contact us at datafeedwatch.com. You can also just sign up for a two-week free trial to try everything that I said. I’ll see if you like it before you pay a single dime to us, and just enter my name in LinkedIn and invite me. And then we’re going to be talking.

Frederick Vallaeys: Great. And we’ll put that in the show notes, Jacques van der Wilt, and DataFeedWatch. So Jacques, thank you very much for being here today, sharing some of your thoughts on AI in the context of data feeds. Everyone, thank you for watching today. If you’ve enjoyed this episode, you want to be notified about, please subscribe.

Also, we would love a rating, especially those five-star ratings, we enjoy. If you want different topics, feel free to reach out to me and tell me what we should be talking about, maybe suggest some different guests. I’m always happy to take suggestions. With that, thank you for watching PPC Town Hall, and we’ll see you for the next one.