Why Hex isn’t charging extra for “AI”

About a year ago, we introduced the first wave of our Magic AI-powered data analysis features for Hex. You can generate SQL, edit Python, build beautiful charts, or even develop whole analyses – just by asking. It knows about your data schemas, past usage, and adapts to the way you work. In other words, they put all the power of Hex, just a prompt away.

When we first rolled these out, we weren’t sure how much these features would cost to provide, nor what to charge for them, so we made them free during the “beta” while we figured it out.

We learned a lot in the last few months and now, as we get ready to launch the next wave of Magic features, we’ve decided that we’re not charging extra for AI features at all.

This is, perhaps, unintuitive; AI add-ons have been a nice revenue juicer for other SaaS apps, and people would probably pay if we asked. But we think it’s the right approach, for a few reasons.

This is just how software is going to work

First, charging extra for AI implies that it’s an add-on, and not central to the experience.

While Hex was useful and popular before AI, our Magic features are becoming useful enough that we already can’t imagine using the product without them - and that’s only going to become more the case as we introduce new capabilities, and the underlying models improve. We’re only scratching the surface of what’s possible with AI in analytics workflows.

In fact, we believe that in the not-so-distant future, we won’t want to sell a version of Hex without AI. A Magic-less version of Hex would be a compromised experience - and why would we want to sell something that’s not the best?

And really, this is just going to be… software. Charging extra for AI is like having an add-on for “cloud” in 2014. That might have been a meaningful – and potentially monetizeable! – distinction for a short time, but within a couple of years it was just table stakes. We believe that AI will follow a similar arc. Anyone starting a new SaaS app today is building it assuming AI, and don’t have add-ons or non-AI tier. This is where the puck is going, so we might as well go there.

We may always offer the ability to toggle certain features off – if, for example, a customer has particular security or compliance issues – but we think those concerns are also similar to those about “cloud” in the mid 2010’s, and will follow a similar amelioration and acceptance arc over time.

Inference is cheap, and getting cheaper

Second, the underlying cost isn’t super high - and it’s going down.

When GPT-3’s API was first introduced, it cost around $0.08 per 1K tokens. Today, in January 2024, that cost has plummeted to around $0.002 – a 40x decrease. And that’s not even really a fair comparison! Today’s GPT 3.5-Turbo is way faster and more advanced than the version from 3 years ago.

Competition is heating up, too. There’s at least a half dozen models that can credibly provide codegen capabilities at or above GPT-3 level, and the overhead and cost to run your own open-source models is going down as well.

That doesn’t mean the cost will be negligible – we expect to spend quite a bit on inference! – but it can be likened to thinking about the cost of running a production database: a ubiquitous, significant cost that adds up, but by no means the main driver of cost of goods sold (nor something you normally charge extra for).

“AI” isn’t a monolith

Finally, it’s not clear to us that it even makes sense to think about charging for AI as a single bundle or way of monetization.

As an example, our Magic Edit feature allows you to use prompts to edit SQL, Python, and charts in Hex. These make the experience for individuals better. Our Data Manager, however, helps admins for teams enrich their schemas with metadata and (in the future) monitor AI usage. Do these make sense as part of the same per-seat add-on?

Instead, it’s better to think about AI as a primitive that unlocks other capabilities, which can each be monetized in the way that’s best for them. As an example, AI-powered features built for larger companies with more complex governance needs may sit on our Teams or Enterprise tiers. Or, AI-enabled features with truly outsize cost could be monetized through consumption fees.

But those are decisions to make for each specific feature, whether it has AI pixie dust or not.

Why not just charge for it anyway?

“That’s all well and good” you might say (perhaps, especially, if you’re an investor!), “but why not just charge extra for it now, while you can get away with it, and then drop it later? It’s free money, right?”

Kind of! Changing pricing is hard. It also adds friction and confusion for our users. Explaining and messaging products and pricing is hard enough, and we’d rather not have to worry about this too.

And again, more and more, we think the best version of Hex has AI. Our users are data professionals who want the best product so they can do their best work – and we want to give it to them. We’d be failing if people were getting an inferior experience just because we wanted to nickel and dime them.

Keeping it simple

Instead, we're taking a novel approach: charge a fair price for a great product.

As of this writing, our Pro tier costs $36, and our flagship Teams tier costs $75, per user per month. Neither of these are cheap, as far as SaaS goes. But for that price, you get a premium product, excellent support, and our commitment to not tack on little surcharges for every feature down the road.

Do we want to make a lot of money? You bet! We’re running a business, and it’s undeniably the point of the whole endeavor. We’d just rather do it by keeping it simple, and building and selling something great.

And this feels like where some others have lost their way. I’m sure we’ve all seen CTAs and prompts for “AI” add-ons. They’re annoying, and can be a tough sell: justifying the ROI of “AI” isn’t easy. Why not just up the seat price to cover your costs plus some, and stand tall and proud that you’re selling the best-possible product? That’s what we’re going to do, at least.

Magic should be free

So, that’s it – we’re not charging for an AI add-on. If you’re using Hex’s Pro and Teams plans, you can use Magic as much as you want, and look forward to some amazing new things we have coming down the pike. We can’t wait to see what you do with them!