Pricing & monetization
Kaegan wholly owned Lumen5's self-serve purchase experience: activation, onboarding, pricing and packaging, experimentation, and the credit systems that its AI features ran on.
The setup
AI features broke the old SaaS pricing playbook. Every video generated has a real cost underneath (voice synthesis, rendering, model calls), and those costs vary by model. Price too generously and margins evaporate quietly. Price too cautiously and nobody upgrades.
His grounding here predates the AI era. If you were an indie founder in the mid-2010s building on Stripe, you probably used Baremetrics, and he spent over three years there. The finance-analytics buyer and the language of metrics, forecasting, and ROI are very familiar to him, and he read the entire Stripe API documentation so he could speak the same language as those customers. That instinct for going a few layers below the UI turns out to be most of what pricing work actually is.
What he owned
- Two AI credit systems, each designed around how its feature was used. AI voiceover ran on minute-based limits tied to plan tiers, while AI media used per-generation credits that could be topped up. Both mapped automatically to underlying model costs, so margins held across an ever growing number of models.
- Paywall placement and experimentation. The paywall and pricing experiments ran around the voiceover transition and contributed to the same result the rebuild is known for: paid purchases up roughly 40% compared to text-only videos.
- The free-tier video limit: cut videos created by free users by about 30% with no measurable dent in conversions. A pure cost-control lever that didn't touch revenue.
- Plan and pricing design for the voiceover era: minute limits, tier boundaries, and the upgrade paths between them.
- The experimentation infrastructure itself, set up in his PM years, so every one of these calls was measured instead of argued.
How he priced the voiceover era
The minute limits are the clearest example of how he thinks about this. In his words:
When we introduced AI voiceover, I created the AI voiceover minute limits for each plan. Our philosophy was to come up with a limit based on published videos, allowing users to make as many edits and revisions as they wanted during the editing process, despite that costing us every time they did.
The reasoning behind it: tight limits make people hoard. Hoarding means fewer iterations, and fewer iterations means a worse final video, which degrades the very thing the upgrade is selling. So the limit counts published videos, and the editing in between is free even though every revision costs the company money.
How it played out
Monetization stopped being a quarterly scramble and became a system. Pricing conversations moved from opinions to margin math, paywall changes shipped as experiments with readable results, and the credit models scaled through each new wave of AI features without a re-architecture. That mattered most as the model landscape kept moving: swapping providers became a cost question, not a pricing crisis.
It also changed how he thinks about the craft. Pricing is a product surface, not a spreadsheet. The paywall is often the single most-viewed screen in the product, and it deserves the same design attention as the editor.