Last week I finally shipped one of the biggest upgrades to my analytics engine since launch.
For months I had been trying to answer a deceptively simple question: how exactly do people move through a document?
Not just which pages they open, but where they start, what they reread, and where they finally drop off.
Most tools stop at page views or reading time. But those do not tell the full story of attention flow.
So I built something new: a Sankey chart that maps every user's document journey, page by page.
Now I can see when someone starts on page 3 instead of page 1, loops back to reread a section twice, or leaves right after the pricing page.
It is both fascinating and humbling, like watching how people actually think through your content.
To make this work, I had to rebuild a large part of the tracking engine:
session stitching, page transition mapping, weighted attention paths... the whole thing became more complex than expected.
But now, the analytics feel alive. It is not just numbers, it is a moving map of user intent.
What excites me most is not the chart itself, but what it reveals.
Some users jump directly into the middle of a deck, others linger on a single page for minutes before deciding to share it.
Patterns like these can completely change how I design content in the future.
As a solo founder, shipping this felt like crossing a tiny mountain, painful halfway through, worth it at the top.
If you are building your own product, here is a small reminder:
sometimes the most complex features start from one simple "what if?" and end up redefining how you see your users.
And if you are curious to see how this works in action, you can try DocBeacon at https://docbeacon.io. It is the analytics feature I wish I had when I started building.
Honestly, this feature nearly broke me halfway through. Rebuilding the tracking engine took much longer than I expected, but it also forced me to really understand how messy user sessions can be in reality.
If anyone here has tried visualizing user paths beyond simple funnels or heatmaps, I'd love to hear how you approached it. The data structure side of this problem is surprisingly deep.