I'm in charge of Product at Ai, and as such I'm compelled to come up with several new ways to use our platform that our potential customers aren't yet clamoring for. This is not as easy as it sounds, because our potential customers clamor for a lot -- from our sports and fantasy knowledge to our financial products to real estate and personal fitness.
These days, if you can dream it, it most likely has data, and we can mine that data and report on it in any format you want. We can make it read like it was written by Bill Simmons or Bill Shakespeare, and we can crank out personalized articles at the rate of (so far) up to 1,100 articles per second.
So by the time I emerge from my dark, windowless office with the next great AI product idea, we've most likely already had someone from some company somewhere ask us if we could do just that. In most cases, they've already sent sample data.
This is what happened with SiteAI. By the time we were through the first iteration some weeks ago, we had already received a number of calls from rather large interested entities asking if we could parse, analyze, and wax poetic on web analytics. It made the answer easier, sure, but a little piece of me died inside.
Besides that, Robbie Allen actually thought of it first. Or so he says.
Words Are EasyThe thing is, just making data sound human is something we've already conquered, on a massive, very public scale. Throughout the 17 weeks and the millions and millions of Fantasy Football Recaps we delivered to Yahoo users, we never once -- not once -- got any complaints about redundancy, duplication, or wooden-sounding content.
This is not because there was no redundancy or duplication. One of the lessons we learned early on is that when you have tens of millions of anything, you're going to have redundancy. You show me fifty-million snowflakes, I will find many more than two that look exactly alike to the human eye.
What we do is manage that redundancy, and we automate that as well.
Look, any kid with a book of Mad Libs can tell a unique story, regardless of whether or not it makes sense. Any kid with a handle on Natural Language Generation can also tell a unique story, regardless of whether or not it's compelling.
The Insights? Not So MuchThat's not to say that building SiteAi was easy. It came down to a lot of hard work from a lot of talented people to get to the information that made the articles sing (Adrian Atkinson, Brian Sewell, and Cole Faloon led the way on this one, but literally everyone at AI pitched in).
Web analytics data are nearly unbound. A site can have unlimited different purposes, and there are ridiculous variances in traffic from site to site. A drop of five visitors from day to day is no big deal, unless you average six visitors a day. But then, guess what? That's still no big deal, in the grand scheme of things. Wading through hundreds of rules like that was the hard part.
We also, for the first time, steered completely away from a traditional narrative format for a project, and this is because it was overkill. Some stories are best told in narrative, others are best told in list. SiteAi uses a hybrid, a list that reads like narrative, but doesn't get bogged down in paragraph flow.
So What's Next?One thing we've done very well is saved the user from having to log in to an analytics dashboard every day and stare at infographics, tables, charts, and lists until their eyes bleed, trying to figure out if changes from yesterday or last week or last month are good or bad or indifferent and what caused them.
As the owner of several websites, I find this invaluable.
It's that last part, though, those reasons why the traffic changed, that makes SiteAi stand apart from an analytics dashboard, and makes it the first product of its kind for analyzing website traffic data. We have a good handle on this now and will iterate on it for future versions of SiteAI -- so that you'll not only know what your traffic was and why, but how you can capitalize on those changes.
In plain and pretty English.