The activation rebuild
Lumen5 had tens of thousands of signups a month, and getting them to a finished video was the whole game. Activation is the metric Kaegan owned longest, and the one the front door got rebuilt around.
The setup
Self-serve at Lumen5 was a firehose. Volume was never the problem; how many of those people got a video out the other end was. He frames the job the same way every time he is asked:
The big one that we would always be working on is time to first value. How quickly can we get a new user from, hey, I'm here to solve a problem, to something that shows them either the full solution or that we're close. And for us, it was really about how quickly can we show them a video that is on brand and is what they're looking for.
Most of the people who signed up saying they wanted a video never got one out the other end, and that gap is what the work was pointed at. The upside of a funnel that size is that it settles arguments. A change could go to a slice of new signups, run against a control, and come back with an answer instead of an opinion.
Rebuilding the front door
The clearest version of the work is the one he tells on calls, start to finish:
Lumen5 promised to use AI to turn long-form content into video. The approach that we used before that was very primitive. It would just rank all of the sentences according to importance, and we would choose the most important ones, and that would become the video script. It was the best we could do at the time. So when ChatGPT rolled around, we were like, oh, we can actually summarize these and write them into a video script. We could tell immediately that it solved a problem that we'd already spent five years trying to solve in different ways.
So the zero to one there was to completely rebuild the video creation experience for our users, basically starting from scratch. That involved first creating a very minimal flow: somebody comes in with a blog post, we show them a script, they can make some edits, and then they can turn that into a video. This performed way worse than the existing flow because it was just so bare bones. We ran it as an A/B test throughout, and we would just iterate and iterate and iterate until we could get to feature parity and activation parity with the previous flow. At which point we were able to replace the old flow with this new one, and then iterate and iterate and iterate so that activation was actually better than what we had before. We eventually increased activation by about 30%, which is huge for us, because we'd have tens of thousands of signups a month, and that's 30% more people that are getting a video at the end of it that weren't getting one before.
Parity first is the unglamorous half, and it is the part he leads with. Replacing the front door of a product this size means the new flow has to be provably no worse before anyone gets to argue it is better.
What he owned
- The activation and conversion funnel end to end, on a product with tens of thousands of monthly signups.
- The experimentation infrastructure itself: the system for rolling out and measuring A/B tests, and the Mixpanel dashboards reporting product and financial outcomes. He built it, which is why these calls got measured rather than argued.
- Two back-to-back projects that replaced the new user experience: the LLM script builder, then the rebuild of the creation flow around AI voiceover.
- An AI onboarding flow that scanned a new user's website, imported their brand colors and fonts, and matched them to a theme, so their first-ever video came out on-brand automatically.
- The unglamorous half: onboarding, undo/redo, and faster rendering, each of which moved activation on its own before the rebuild did.
- A two-month growth squad of two, staffed with engineers pulled from existing teams, pointed at activation and retention.
- The free-user video limit, which cut videos created by free users about 30% with no measurable impact on conversions.
The wins nobody asked for
The biggest activation wins came from watching what people did in their first session, not from what they asked for:
A lot of people when they're using the product for the first time would take some sort of action, oftentimes a delete type action, then drop off. And that built the case for undo redo. We were like, oh, it's a very simple product, this isn't a big priority. And it wasn't anything that people were specifically asking for in those words. That had a pretty immediate impact on activation, and the usage of undo redo was one of those hockey stick graphs, probably one of the highest adopted features that we'd ever released.
The brand scan came from the same place. Businesses wanted the first video to look like it came from their company, and getting there meant hunting down hex codes and hunting through a template library, so a lot of them just didn't. The flow scanned their website, pulled the colors and the fonts, took a snapshot to work out the motif (a triangle brand, a circle brand, a gradient brand), and matched it to a theme.
The bet that didn't pay
Owning activation also means owning the quarter that goes nowhere. Off the back of the voiceover work, the team bet on engagement loops: if one video from a blog post works, why not five?
We got really low adoption on basically all of these features. Anytime we tried to show people multiple videos when they tried to create one, in a lot of cases it would reduce activation. Most of them had a content workflow where they're like, hey, we write a blog post, it's about this one thing, then we're going to turn that into a video. That's our main piece for the week. And even though LLMs are really good at coming up with stuff, they don't work well when they don't have context. We deal with oil and gas companies, for example. They write something about some new plant, and then we would try to come up with an idea for another video. We'd be like, oh, you want to talk about solar wind farms now? And they're like, we're not working on a solar wind farm at the moment.
The big learning was, more of a good thing isn't always a good thing. And another was, if we're going to do things, make sure we have the right context to actually give people something that's relevant to them.
How it played out
The rebuilt front door lifted activation about 30%, on a funnel taking tens of thousands of signups a month. The script flow cut the time it took to get a first video by about 30%. The free-user limit took roughly 30% of free video creation out of the cost base without a measurable dent in conversions. The first of those is the headline; the other two are what made it stick.
The habit outlasted the projects. Parity first, then lift, is how the voiceover rebuild shipped a year later, and that story is the next stop on the line.