At aerolab we work hard to help our clients connect with their customers in meaningful ways.
But we quickly found out that in very successful campaigns a huge amount of data is generated and makes it impossible to
understand what’s happening. Who liked your campaign?Who wants to buy your product?Who are your most valuable customers?
Even with huge -and boring- Excel spreadsheets it’s hard to see the people that stand out from the crowd, let alone finding
opportunities to make new friends. And that’s the kind of hard problem we love to solve.
What is Love?
The thing that makes Machine Learning the most useful for us, is granting software the ability to learn
from experience and perform actions automatically. This grants our applications a certain level of
intelligence, where they are able to make the best decisions constantly, without needing to have
a programmer on hand to tell it how to do things.
This makes it possible to identify segments of the population that with traditional segmentation criteria might as well be invisible. And most importantly, being able to find out what makes each one of our customers tick, understanding their personalities and finding out things we might have missed otherwise. This gives you a greater understanding of your customers and the effectiveness of your campaign.
And with in-depth reports, it’s possible to identify new segments based not only on basic population data, but on their interests and social networks as well.
This gives you a leg up on your competitors by being able to identify trends and patterns before anyone else, and quickly testing new ways to engage those customers before anyone else can get around to it.
Being able to identify patterns automatically means it’s possible to automate actions as well. This makes it possible to deliver messages to each segment based on the most effective criteria for every single customer.
This means not only less work, but also letting our bots do the ground work for you.
It’s possible to pick the best stories, products or even headlines for every single person and have the system automatically deliver the right messages to them.
This not only makes the process much smoother and hands-free, but it can outperform traditional campaigns very quickly.
This means we can have machines delivering slightly different experiences to every single customer, based on their unique personalities.
The music fans will get something music-related, the football fans will get something more sports-related, and not only that, but we can mix that up with their Facebook and Twitter profiles, so we are able to detect tendencies based on what their friends are doing, sometimes even before they even know they like something.
Keeping 2000 teenagers from having a crisis
FunTime is one of our favorite brands. Not just because they are really fun people, but also because they
know their product like no other company we’ve ever seen. But with great marketing and a great product
come great (and interesting) problems as well.
The problem was seemingly simple:
We had to find the best way to organize the rooms in each trip.
So 2000 girls divided by 4 in each room is 500 rooms and that should be pretty much it.
The usual approach was just letting the girls pick their roomates in a large meeting, but that was causing problems, since it only works for out to a few hundred people or so before it becomes overwhelming. Not only that, but the number of possible combinations is staggering:
There are more ways to organize those rooms than there are atoms in the universe.
We wanted to find a solution that works every single time, even for huge numbers of girls and
-most importantly- made the experience significantly better for everyone involved. Less fights with roommates,
less administrative work, and since your roommates are a key component to making the trip a memorable one, a better way to experience DisneyWorld.
So... we built a social network!
But not just any social network
Yes, you can comment, post photos and “like” things, but behind the scenes lies most of the code that makes the system tick:
A huge Machine Learning algorithm analyzes each girl’s personality profile and
matches them with the people they are most likely to be compatible with.
Everything from music, books, their friends in common or the school they go to goes inside and automagically the system pairs you up with the best possible roommates, including the friends you already know.
And if everyone doesn’t fit in your room, we even try to put the rest of your friends right next door.
There are even internal reports on all the information we collect to help improve the trip based on the things the girls like the most, while protecting everyone’s privacy.
By helping the passengers make the best possible decisions in a completely transparent way, the number of conflicts went down by 90% and FunTime Community (the social network we built) has now grown to become one of the selling points of the product.
The system was such a hit that not only the girls loved it, but the coordinators were amazed at how easy going the groups are now. Coupled with a completely redesigned CRM integration, we've automated most of the communication processes, helping them concentrate on growing their product instead.
Building a thousand brands out of one
Cuponstar is one of the most important players in the local deals market, and we can proudly say that
we’ve built all of their platforms so far, from their first website a few years ago to their current
IT infrastructure that keeps up half a dozen services.
Through their unique approach of distributing discounts via SMS shortcodes, they’ve grown their
customer base with products that go all the way from movie tickets to eyewear, having more than 300 promotions online at any given moment.
When we started taking a look at their customer’s behavioral patterns, we noticed that each one of those
customers tended to buy a specific set of products. We found movie fans, who love the 2x1 discount in tickets,
a lot of foodies, who tend to pick some specific foods from sushi to barbecue, and others who love shopping,
enjoying some great discounts in clothing and accesories.
This caused an interesting problem:
As the number of users and products grew, the brand started feeling generic
and not really tailored to anyone in particular.
This posed the question: How do we make each one of those individuals feel identified with the brand? After looking through the mountains of data we accumulated through every single channel
(from the website through the newsletters)
We arrived at a unique solution:
Making a thousand different websites for a thousand different customers
By applying clustering algorithms to the consumer database we taught the servers to recognize which customers belonged to each segment. The great thing about this is that the system works even when fed seemingly irrelevant information, like where did they their customers get their email accounts
(it turns out that gmail users tend to like going to the movies).
That way we can start building their profiles very early in the game and start customizing the site so they’ll get more
out of the experience, helping them find what they want faster and suggesting things they’ll
love, even if they don’t know them yet.
Even in early testing this has worked wonders at delivering promotions with a much bigger conversion rate than usual. Addressing their wants with the laser focus machine learning delivers is a fantastic resource when it comes to getting a loyal customer base and helping them identify with the brand.
Which is, in a nutshell, what we love doing.
If you want to make something amazing with aerobot or learn more about how we can
use Machine Learning to improve your business, drop us a line: