Building a Free Sidecar Product Using ChatGPT

A product-led growth (PLG) case study

As a product-led startup, we’re always looking for new organic marketing channels: viral loops, SEO, and lately, building free external tools that can help drive traffic, sometimes referred to as sidecar products (as coined by Kyle Poyar, check out his case study as well).

Side products are a fantastic concept, but they do have their own challenges:

1. They take time to build. Dev time. Expensive, potentially de-focusing dev time.
2. They have to be relevant to your main audience.
3. They need to bring something new, meaning that no one out there has built them before, which is rare for small niche products since they’re so easy to build.
4. You still have to market them, this time using potentially expensive and de-focusing marketing resources.

In this post, I’ll share how we dealt with these issues building Watermark Wiz — our first-ever sidecar product at Tagbox.io.

Watermark Wiz’s origin story

Our main product at Tagbox is a “Google Photos for businesses” — a media collaboration tool for companies with lots of photos and video. Many of our customers come from the events industry, since they tend to have a truckload of photos.

As it turns out, many of them would like to have their photos watermarked, so that when they share them with the event attendees, they’re branded. Many attendees also share the pics on social, adding another motivation.

While we try to be responsive to customers’ needs, obviously not everything can be built, and this was unfortunately prioritized down for a while. As an interim solution, I was looking for free products I could offer our clients, so they can watermark before uploading to our platform.

While the search “free watermarking tools” yielded a bunch of great-looking options, none were actually free — they all either put their own logo on the photos, or had you watermarking them one-by-one, unless you upgrade to a paid plan.

Gee whiz, I thought (well, I might have used another term): this can’t be that difficult to build, can it? That said, as co-founder of a startup that’s actually doing pretty well, I didn’t have a lot of time to go through the actual process of spec > design > development like I would with most products.

However, for a while I wanted to test ChatGPT’s actual limits in building real-life applications, so this seems like a good opportunity to do exactly that.

Keep in mind that I do have some development background. I’ll try not to be too technical, but I do want to quickly describe the process I used, in case it can benefit others:

Building the app

[This will be slightly technical — skip if that’s boring]

I started a new React project (which took about 5 minutes using Vite), and broke down the basic tasks, asking chatster to hand over the code:

Step 1: adding photos

1. Allow the user to upload photos and then display them
2. Add a “continue button”

Step 2: adding and positioning the watermark

1. Allow the user to choose a watermark and display it over the original photo
2. Allow the user to drag and resize the watermark
3. Generate the thumbnails and preview them
4. Allow the user to download

It took about 3 hours to get the main functionality working. Here’s a video I sent Oz, my co-founder. I was such a proud sidecar dad!

Unfortuntely, as any developer will tell you, debugging often takes more time than the initial development, as was the case here. Not everything ran smoothly — the dragging wasn’t smooth, the logo often went out of the boundary of the original photo, and the app would randomly break on mobile.

Debugging with ChatGPT was partially helpful, and adding GitHub Copilot to the mix helped a bit too — but for the harder issue, I was pretty much on my own.

To put it simply — for non-developers, building a real, functioning app using Gen-AI is just not there yet. Even for this simple example, I would still be prompting it if I didn’t know how to code myself.

But for creating the basis, it was actually great, and probably saved me the better part of a week in just writing the basic code.

An app is born

Finally, within 3 days, I had a working app, and as an extra, I GPT’ed a nice logo and some marketing texts for the site. You can check it out here.

As I mentioned in the beginning, building the app is the easy part — the thing is, you still have to market it. With the ‘free-but-not-really’ tool already having lots of SEO juice, it will likely take a while to get on Google’s Radar for this.

We’ll do the basics — launch on ProductHunt, email our list, post on social. If this gets any traction, we’ll likely build some more tools.

This is definitely the beginning and not the end of this experiment, I’ll keep you posted in a few months on the results!