OP here -- the numbers are in no way bogus, all sourcing information and methodologies are listed on the page itself (see the bottom left * on each 'slide').
Would love to understand what gave you that impression, as we work hard to build content like this, and hate to see it shrugged off.
"Custora's research shows that stores on average lose roughly 1.75% of revenue every year due to unsubscribes."
This is remarkably similar (in amount) to loss through shrinkage at bricks and mortar stores, which is estimated at 1.7% (according to one source I bothered to find) [1] - but while retailers take steps to reduce shrinkage, there is a cost/benefit balance in play and a certain amount is inevitable, maybe the same is true of unsubscribe
I'm more curious about how that 1.75% compares from other sources of lost revenue; how does it compare to the cost of processing returns, or fraud, or just basic transaction costs, &c?
Maybe it's significant, maybe it's noise, but without anything to compare it to, who knows
Maybe it's retconning loss figures to a model they already have. If they can fit online expectations to meatspace loss figures, then they don't have to change their models.
Stock buy backs are used to reduce the number of outstanding shares available - which results in shares increasing in value due to earnings per share now increasing. It is one way to "mask" slowing growth.
Is the data collection methodology described somewhere? I'm questioning the accuracy given an error on our data (GoDaddy is not our registrar, namecheap is).
It's also a bit suspect that namecheap is not listed as anyone's registrar after so many fled GoDaddy for namecheap a few weeks back.
OK -- the data is definitely out of date for at least one company (ours, custora). I wonder how common this is? Perhaps the cache was not cleared before generating the output?
I forked the repo, ran it on custora.com, and eNom was (correctly) listed as the registrar.
I'm re-running the full ycombinator list now, and will update when finished.
This may seem like nitpicking, but I am always worried when I look at a website and the blog hasn't been updated for months. Even a short "Hey guys! We're working on stuff!" would assuage my fears that you've turned into a zombiecorp. =)
Whoops! Thanks for the feedback! That's an old page that was supposed to be replaced on our latest redesign.
The simplest explanation is pattern matching: we analyze customer and transactional data to understand how different customers behave. Using this understanding we can make predictions for how each user will behave in the future.
We use all of that analysis to power actions - take actions on the right user at the right time, optimizing for CLV.
That says a little bit more. Do you treat it as a reinforcement learning (RL) problem, or as a classification problem? It seems like a sequential decision making problem, so RL is appropriate but AFAIK there is no RL algorithm that generalises over states while still retaining some error bounds. I suppose you could brute-force a Bayesian solution via MCMC.
There are two big problems that we deal with. First is estimation of customer lifetime value. We use a latent attrition model, which is the 'pattern matching'
The second is figuring out which promotions/emails go to which people. This is a supervised learning problem. We train the model with users past responses to discounts and their past behavioral states (which are the posterior probabilities from the latent attrition model). Then we use this to predict how users in those states will respond to similar promotions in the future.
Agreed - seems like the terms could potentially defer the top-quality companies (although most would be happy to have the $ in their pockets), while the start-fund structure makes it attractive to any stage.
I'd hope the money is elective though, as that would certainly solve the problem.
We've just updated the post. The correct date is: Thursday, January 31st.