what are cohorts?
an ancient Roman military unit, comprising six centuries, equal to one tenth of a legion
a group of people banded together or treated as a group
I love definitions like that (#2), let's break it down.
a group of people — ok pretty self explanatory, we can group people by age, location, status, education, activity and more.
banded together — sounds a bit like fate brought them together. I understand that someone did this banding to the group of people, still makes sense.
treated as a group — bingo! treated is the key word here.
Who treats? A group of what?
if we want to treat something as a group we have the obvious ways of doing that, the objective groups like demographics, groups like activity types, purchasing groups, and more, lets call them the objective way as everyone is doing it and everyone should. Subjective way will be to treat/group things that are only relevant to you and technically only you can measure.
Why Cohorts and not simply — Group?
Someone once told me when i studied for the SATs, if there’s a word — there’s a specific meaning to it, and group is too general to describe what we actually can achieve with cohorts. Cohorts helps us better define a group, create cross-groups, filter, adapt and most importantly experiment.
If you’re in marketing, you probably heard that word and its ok if you’re not entirely sure what exactly does it mean, thats what we’re going to cover now.
The best way i choose to describe cohorts is buckets. Let’s take for example Jane, 35, who is really active is social networks and a lawyer, we can imagine that she’s a member of several dozens groups in FB, Instagram profiles she follows etc. We can create a bucket called Women lawyers and it will look like this:
Thats a cohort — we’ve created a bucket with very basic demographics and while we can fine-tune it to the max, we don’t want the bucket to turn up empty. If you make it too specific and based on true/false conditions, you risk of not getting hits on that criteria, instead you need to be able to expand and measure things that are not that clear at first, like what happened before/after you did something (liked, commented, shared). We need to add trackable things that will coincide with the first bucket, and will make us see data in various forms.
Facebook, Google, Twitter, Amazon, Snap, etc: Let’s see how the big ones do it
When you scroll FB and see the suggested content, you engage with it — you like, comment and share, or you ignore it and move to the next post. But FB learns a lot from the content you’re not interacting with. What are they tracking? here are a few ideas:
How long did the content stayed on your screen (viewable)
What did you see immediately before and after
How many of your friends (there are several circles of closeness) interacted with the same content
How active are you in general, by how, i mean exactly how — how many likes have you made, how many comments, how many logins, etc.
and all this is deep drilled within itself and all is tracked and of course- all is scored. That scoring mechanism is what makes or breaks FB engagement and as a result, its business.
for each create a bucket, for each bucket decide how it will coincide with other buckets and on top of all those buckets you can run virtaully endless queries.
Data is “stupid” unless you ask the right questions
These of course are just an example, but what i admire about FB is that they realized that first you need to collect all the data then you can start figuring out what to do with it and how to use it. FB is collecting everything! naturally they have tons of data and segmenting it is a challenge because you can’t ask the same questions over time simply because times and tech changes too rapidly.
Ok, lets make it harder now.
here are a few types of buckets FB might create:
Specific engagement: How many engagements you had with this type of category (politics, fashion, food, Star Wars) — measure each specific engagement and score it (comment is worth more than a like)
Tone of engagement: Today FB is using super advanced AI NLP algorithms to extract the feeling associated with the text written.
3rd party FB connect: Since you login with FB to many other media outlets and services, they can basically track what you’re doing there
Other media sources: partnerships with other online services— they also get tons of external data about your commerce patterns, activities and more
Now we can have 2 buckets:
Notice i am still using the first bucket, only adding to it a lot more segmentation, but the nice thing here is that they can be used together or separately. And to make it way more complicated if we add 5 more buckets each measuring super targeted actions and then run any number of combinations. imagine for a second how much data can you have at your disposal…
The important thing to understand here is that FB knows you better than you know yourself, because unlike you, FB keeps track of everything and in a way that is highly manipulative and flexible. FB uses it to pinpoint the things they show in your feed and measure how you are reacting to it. Supplement to that things you’ve: purchased, visited, traveled to, saw on TV, listened to and even eaten, and you get something tp work with. The beautiful thing (from FB’s POV) is that the more they hit (and they hit) the more you are happier and Dopamine kicks in to get you hooked without even realizing it.
Same can be said about Amazon and the others. eCommerce seems simpler to crack because you’d think that you need to cohort only purchase behavior, but with conversion rates in eCommerce at 0.5–3% — they can learn way more from the rest.
Which cohorts to create?
The reason i shared the FB angle was to open our minds about cohorts and what they can be in 2018. Any cohort basis starts with what we want to track,and the short answer is EVERYTHING. Because we can.
Free tools like Google analytics gives you all the basic stuff you need, and here is a great post to get you started. Other tools: Mixpanel Retention, Free Cohort Visualizer ,Cooladata and more.
Here are a few questions you need to ask yourself when starting with cohorts:
What can i know about my users that i cant derive from my data? (meaning what other sources of information can i use to correlate with my existing data collection)
Will i be able to use the cohort data to change or adapt my marketing strategy?
Will i be able to understand the cohort analytics in a way that would make me see better and clearer what’s working and what’s not?
What metrics i want to improve using cohort analysis? in order to change something we need to fully understand it
Eventually when you use cohorts in the right way, you can not only improve your results and marketing, you can find new features to develop, new markets to go into or strengthen, how much to charge for your services/products and more…
On a personal level, i believe its important to be aware that big companies today use cohorts in ways we cant imagine nor even know, I use it to remind myself that what i’m seeing or buying was funneled to me because someone cohotred me into a bucket predicting there’s a 84.7% of me actually buying that item or. speed up the consumer behavior purchase decision process.
A note about millenials
It’s easier with millennials (or Gen-Y’ers) because their life is already digital and tracked and like an organism it evolves and adapts, so its crucial to keep track of everything not just because you can but because that’s a good bet. Read more about it here.