We should have released AI Agents differently

The release of our AI agents is the real first major product update to Herodesk since our launch.

Don't get me wrong, we've developed and released hundreds of features and improvements since launching just over two years ago.

The core helpdesk suite is almost unrecognisable today compared to then (which is a good thing, btw.).

But the AI agents are in another league when it comes to product updates.

It's taken us almost 6 months to create the first version, and it adds a whole new dimension to our product suite.

We launched it on October 2nd. It's been a few months now, so I've reflected a bit on what went well and what we'll probably do differently next time we have a big product update.

Here are the three biggest lessons I wish I’d known before launching a major AI product.

1) Don't build hype around a specific launch date

In the weeks leading up to October 1st, I built all the hype I could about the upcoming launch. Newsletters, SoMe posts, ads - and many more things, and lots of them all.

The point and intention were obvious: to get as much attention as possible for the new release before it went live.

As we approached October 1st, however, I could see that it was going to be tight. Really tight! We cut back on some features that weren't "need to have" (more on this later) and ended up working (too much) overtime for two weeks to meet the self-imposed deadline.

Still, we couldn't, and the launch wasn't until the day after.

I honestly don't think anyone noticed, but it annoyed the crap out of me that I had promised this new feature on October 1st, and we didn't deliver on that date. In addition, it put a lot of unnecessary pressure on me and the team to work overtime to deliver on this date.

So first "note to self": Don't build hype around a specific date unless you are 100% sure you can deliver.

2) Launch in stages and release gradually

We went from internal test/verify/accept to public release. Quite a jump - and a too big one.

In hindsight, we should have gotten more feedback on everything from the GUI/control panel to which features are must-haves and basic "how does it feel to work with" in general.

Some learnings:

We underestimated two concrete features that we initially thought could wait, but turned out to be must-haves from the outset.

We should have done a controlled soft launch to a small, selected group of customers, who could have spent some time testing, configuring and using it, and given us a lot of feedback.

This would have had the added benefit of building a closer relationship with those customers and making them ambassadors.

Then, we should have launched as a "closed early release" to everyone interested (we had +50 on the waitlist) and repeated the process with them.

And only then, once we were 100% sure about the features, the interface, the setup and everything, should we have made the public release.

By releasing to the public right away, we were also insanely busy implementing all the feedback we got in the weeks following the release.

3) Be careful with sales assumptions

Our AI agents are an add-on that you must pay for in addition to your base Herodesk subscription. So, of course, I've made some assumptions about how many we can sell and included that in our budgets.

In October (the release month), I estimated we'd sell 16 agents.

 We sold (drum-rolls): 2.

Ouch.

More than 100 were created and deployed. We have a freemium business model for AI agents. But only two upgraded to a paid subscription. 

On the one hand, you must, of course, make some assumptions about how much you think you'll sell when releasing a new product.

I mean, that's the whole point of doing it...

On the other hand, you have zero experience to base your assumptions on. They are basically grabbed out of a combination of thin air and gut feeling.

I think that if we'd followed 1) and 2) above, we'd sold more in the first month.

We've caught up with some of it now and are still on track to deliver fine numbers on this project, but somewhat later than initially expected.

--

So yeah, all in all, expect me to implement these learnings next time we have a big release.

 If you have similar or other experiences from big product launches that I should keep in mind going forward, don't hold back 🙌