1 Year of Byword

Excerpt

7 lessons learned from the first 12 months of byword.ai

Published Date

Nov 15, 2023

Hide in Main Feed

Featured Image

Author

Mack Grenfell

The day this post comes out (November 16th) marks one year since Byword’s public launch.

From what was initially just a landing page that’d email you an article, a lot has happened in the past year!

In that time, Byword has:

Grown to nearly 50,000 users.

Passed 7 figures ARR.

Generated 2 million articles (and over 4 billion words).

To celebrate these milestones, I thought it’d be fun to take a little look back at some of the lessons learned along the way. I hope that some of this will be useful for anyone with an interest in the AI space, or who sees themselves building something similar to Byword.

#1 - Competition is great…

Before I started building Byword, I worked as a consultant - working with a range of startups to put together large-scale AI-written content campaigns. I’d always wanted to productise this work into something that scaled better than I and my time could.

One of the things holding me back though was seeing how much competition was already out there in the market - surely there’s not room for another AIxSEO writing app? It may be a cliche lesson -that what looks like competition is actually market validation- but it’s painfully true.

It’s especially true in context of a new and exciting field like AI, where no single product is ever going to capture all of the demand and use cases that customers have. Every product out there is still figuring things out to one degree or another, and that’ll always leave space for newer entrants.

#2 …as long as you know where to specialise

Once or twice a week I’ll get asked a question that runs along the lines of how is this better than just using ChatGPT?

I’ll give what I think is the honest answer, which is that if you just want to generate an article, you’re better off using ChatGPT. Sure, there’s a bit of extra work to get it to the right length, generate the meta tags, the image, get it onto your site, sort out interlinking, and so on, but it’s not insurmountable.

The problem, is though - what if you want to generate two articles? Well, if you’re using ChatGPT, you’re looking at roughly twice the work. Ten articles? Ten times the work. A thousand? Ten thousand? You get the idea.

With Byword I aimed to build everything around trying to solve this scaling problem. You’ll spend a bit of time getting your account set up at the start, but after that there’s next to no marginal work for generating any number of articles. This is essentially the bet that Byword was built around - that users find value in the convenience of generating articles quickly, easily, and asynchronously.

I call this the bet because pursuing it has meant deprioritising or outright ignoring other avenues to go down. Byword has never focused on giving users the ability to make tweaks to articles, or generate different iterations (aside from a relatively simple rewrite feature); neither has it ever tried to be a writing assistant. These features appeal to a very different class of users to those that Byword is built for - those that want to push the limits by generating content at scale.

#3 Don’t over-engineer it

On a technical level, Byword is quite a simple platform.

It’s largely built on top of bubble.io, a no-code platform that allows users to build web-apps without writing any lines of code. All of the front-end that you see, and the database that sits behind it, is built on Bubble, without me having written any code.

This isn’t to say that Byword is a no-code app however. There is plenty of code running in the background, handling everything from article writing to site integrations, but it’s nothing enormously complex.

If I’m honest, it wasn’t so much a conscious choice to build on top of Bubble. Rather, it was the only realistic choice. As a self-trained developer, there are things I’m good at, and things I’m less good at. I wouldn’t have the skillset to build out every inch of Byword by hand, but I have the ideal skillset to code up all of the AI parts of Byword while Bubble handles the rest.

The reason I’m making this point is to highlight an opportunity that’s often missed. Typically, people with domain expertise (i.e. a knowledge of SEO) don’t have the technical skills to build fully fledged products. No-code products like Bubble (or Byword’s low-code use of Bubble) lower the barrier for entry, allowing founders a way to productise their domain expertise without having to go and re-train as a full stack developer, or hire engineering talent.

This approach has not only meant that we’ve been able to iterate and ship features quickly, but also avoid traps that similar products can fall into, which leads me on to…

#4 Smart scaling

Byword has never taken, and will never take any funding. It’s entirely bootstrapped, and I think this is without a doubt the right path for the majority of generative AI products. The economics of AI today make it incredibly simple to prototype MVPs for minimal cost, which in most cases is really all you need to validate whether an idea is going to work.

As I mentioned earlier, the very first version of Byword was just a landing page where users could submit a keyword and provide their email address - Byword would then email them their article a few minutes later, in exchange for signing them up to the waitlist.

This was very quick and easy to build (it was literally just one page!), but gave a huge signal early on that Byword was something that users had appetite for. In the month that Byword existed in this form, we had over a thousand users sign up, and a number of them saying they were willing to pay as soon as the tool went public.

The traction from the public launch, which in reality was just a function of our waitlist landing page, meant that we were able to grow Byword in a fully sustainable way, without taking on any outside funding.

#5 Proof of work

Going in, I knew that one of the big issues we had to solve for was users’ trust in AI content to actually deliver results. Nobody needed convincing that this could work (who doesn’t want more content for less expenditure?); the key question on people’s minds was whether it actually would.

Probably the biggest advantage of having worked on a range of AIxSEO projects prior to Byword was the ability to showcase their results as proof that this stuff really did work. The primary example of this is of course the Causal case study, which recently passed the milestone of a million sessions per month.

It’s hard to overstate the value that this has had in growing Byword. In such a new space, where there’s a lack of consensus over whether these methods are actually reliable, it’s been invaluable to be able to point back to this as proof.

While this lesson is broadly applicable, it’s especially so in the context of generative AI. The lack of pre-existing case studies means that any new ones have potential to reach and convert an incredibly large audience.

#6 Short customer feedback loops

There are times in Byword’s life where we’ve asked for feedback on a wide scale, but also times where we’ve done so in a much smaller sense.

We often silently launch new features within the product, before announcing them over email. This gives us a chance to see who uses it, and how they use - I’ll often follow up with them individually to get their thoughts and feedback.

This provides a very low-pressure way to develop and prototype new ideas, without all the expectation that comes from large-scale launches (which can often happen some time after the feature is first available).

Working with users on an individual level also tends to elicit a level of honesty and trust that doesn’t necessarily always come with a mass email announcement - we’ll often get a ton of great feedback from users by reaching out 1-1, that we likely wouldn’t have gotten otherwise.

As a case in point, if you've come from the Byword 1 year email, all of the upgrades under A Better Byword have come from these kinds of conversations.

#7 Bring your own key (BYOK)

One of the biggest product bets we took with Byword was on building out a BYOK (bring your own key) model. You might know this as Byword Unlimited - pay Byword’s monthly platform fee, and provide your own AI API keys, in exchange for unlimited content generation.

It’s a model that has played an integral part in Byword’s growth, and has worked out well for a number of reasons:

It helps provide a unique product offering to users who want to scale their content as much as possible, and not be beholden to article credits.

Byword can allow users to do that without affecting service for non-Unlimited users:

Unlimited users being on their own API keys means that they don’t affect Byword’s own rate limits
Unlimited users have their own dedicated server space configured around their own API keys, which means that their activity doesn’t slow down Byword for others

I fully believe that this is an optimal model for a lot of generative AI apps out there, and suspect we’ll see more convergence to BYOK models going forward. If I was starting a new app today, I’d consider even building the entire product offering around a BYOK model!

Wrapping up

These are just some of the lessons learned over the last 12 months. There are plenty more I haven’t spoken of, and no doubt plenty more to come.

If you’re not a Byword user already, I’d love for you to check it out, and see some of these lesson in action. If you are, then thanks for reading, and for being part of Byword’s first year!

Mack