CWI#61 - 📊 Practical With Analytics & AI 📊

Being sceptic about AI is fine. Getting insights on the practical stuff and what you can do with it T
CWI#61 - 📊 Practical With Analytics & AI 📊
By Dries Bultynck • Issue #61 • View online
Being sceptic about AI is fine. Getting insights on the practical stuff and what you can do with it TODAY to help solve marketing and growth related issues, is ACE! And that’s what I’ve learned last week. On top of that, some insights I’ve got by attending MeasureCamp in Brussels last saturday. Interesting week. Hope the topics below translate to you in any form of value. If not, let me know and how I can fix that for you.

The Search World lost a Comforting Wizard
But first… allow me to start with some sad news earlier this week. Eric Ward might not ring a bell? Eric was to many the best link builder out there. Earlier this week, he left behind his family and kids at the age of 58
I’ve been fortuned to briefly work with Eric on a client back in the days with bSeen (SEA/SEO company from Marnik D'hoore & Stefan Audenaert, a few years back). 
I was amazed by his level of professionalism and extremely easy to communicate with. Easy going. Gentle & supporting. Transparent. I think comforting might be the best word to describe the feeling. A very rare combination in this world. 
Good bye Eric.
Thank you for the inspiration.
MeasureCamp: Practical Analytics Summary
My first measurecamp and definitely not my last one. Nice to meet and see people again within the twitter/analytics space. Over 40% of the attendees were from abroad! Didn’t expect that. wow. Russia, Denmark, UK, Australia, USA, France, Germany, Romania, … 
The board got filled rapidly for this unconference. So lots to choose from. Missed out on a few interesting topics apparently. I skipped the technical talks as I don’t want to get distracted from building my own GTM Framework(s). More on that within a few months, I hope.
It really made me happy to hear Peter O'neill talk about the misconception and bullshit (thanks to Adam Greco - ex, Demystified Analytics) attribution really is. We simply can not do it properly due to the lack of data and what kind of efforts translate into the uplift of shift of attribution (channel + marketing or other efforts). A very hard puzzle too solve that has no value for your customer!
If you’re interested, feel free to scroll through the tweets to get a grip of what was cooking during that day. I think the slides will come online later this week. I’ll share them on my twitter & linkedin. Here’s already a first one by Yves-Marie Lemaître from Ensighten
Key take aways from the talks (which can help you):
  • Help the CEO and support him with data to answer questions
  • Make presentations as simple as possible. Your boss doesn’t have the time to go into the details. Give hem the numbers and suggest the next question to solve or to ask.
  • If you don’t get questions, suggest a list of questions to work on
  • Attribution is almost impossible to solve and will definitely upset people when you try (ego, accountability, etc.)
  • If you can’t translate a metric or analysis related to a customer action, your a bad analist
  • Don’t upset people, listen to them and help them to solve their problem
  • There is more gain into cost-cutting than selling more. That’s fine too. And it should be a part of your solution
  • When you’re sitting more than 8h behind your desk, you’re not talking to people
  • Build-in at least 1h per week to have an open office hour and talk to people
  • Allocate one day per week to do the number crunching and focus on the doing!
#measurecamp hashtag on Twitter
Visualization is a skill you need to practice
Graphs & visualization. Highly underrated. At MeasureCamp, another analist talked about pie charts being hated in the industry but he didn’t knew why. Well… Pie charts are one of the worst items to show data that has any form of relation. It shows no relation between the data and has a discomforting level to understand (ex: 4 categories at 25%, 25%, 24% and 26% for instance) within one blink of an eye (rule of thumb for exploratory viz). Always avoid pie charts. Use bar charts instead!
Grafiekjes, dat blijft toch moeilijk | Michel Vuijlsteke's weblog
About AI ...
It was mentioned in the course and mentioned by Steven Van Belleghem in his last video. But here’s the main thing. All headlines are about the biggest and unachievable (yet) sorts if AI by average companies. General AI. The far-out most advanced type of AI you can get within the next 10 years or so. So… we skipped that part as it needs computing power (and the knowledge) only a few on the planet have access to, at this moment. Pls… check the article. You’ll understand. 
AlphaGo Zero: Learning from scratch
So… in the course we focused on types of vertical AI. The kind of techniques you can use and apply to marketing and growth related questions, TODAY!
Here’s the major break down to use towards your manager or client:
There are three steps:

  1. Descriptive Analytics (what happened):
    Suggest and do an exploratory analysis first that answers at least one specific question. Report on the numbers and suggest the needed actions in order to go and do step 2. Mainly: get & clean-up the data. The shitty part about this, is that it’s 60 or 70% of the time for this three step process. So that sucks. Preparing a small sample with a 1000 records would be a nice start to build the case.

  2. Predictive Analytics (what will happen):
    Once you have the go from management, it’s up to you. Dive into the prediction part using a few different algorithms and get more insights. Testing the best algorithm on a sample and then applying it to the main database adds extra value to the data. It’s an extra person in the room ;) On the what will happen part, take that with a grain of salt. Predictive modeling can also be to what part of cluster of what parameter is influencing the output the most. Correlation, not causation.

  3. Prescriptive Analytics (what should happen):
    In this part, you’d usually estimate the outcome. This is the actual predicting part of what should normally happen although you’re not sure. Example: CLTV should be 18 months, but could be that for some reason I’m leaving as a client due to a bad delivery of my order. You’ll never know, you only can guess the future.
So… depending on the type of business your in, meaning, B2B or B2C, different algorithms will be used. B2B is the most easy part as the most variables are fixed. Variables such as CAC, Pricing of your product and thus Churn & CLTV. For B2C, that’s a whole different ball game. There are so many variables in B2C, especially e-com, which happens to be the most communicated AI related business opportunities. But… there the hardest to solve too. No one talks about that, don’t they? :)
Shit is fucked up out there. 
B2C data is more complex, more personal and more single customer and product type driven than B2B. Every product, every category, every price, every size, every colour, every channel, every client, etc. is different. And, than on top of that, you have the behavioural data of Google Analytics. 
If you need any help figuring out what to do with B2C, hit me up by hitting the reply button. That’s it.  I think I can help you to get to those first steps & come-up with a first case and improvement. 
Any how… getting into this thing is extremely high on my agenda, no matter what and I’m going to use it everywhere I can, whenever I can. 
Passed the radar this week
Google (GOOG) built earbuds that translate 40 languages in real time like the Hitchhiker's Guide's "Babel fish" — Quartz
Minimum Viable Conversion Optimization: If You Only Have One Day a Month, Do This!
See you next week.
Did you enjoy this issue?
Dries Bultynck
By Dries Bultynck

Remarkable reads, spotted momentum & behavioral patterns in media, digital, retail, economics, health, climate, etc.
Often hollistic, sometimes very specific.

I'm Dries & the internet is the best thing that ever happened to me. more about me here:

If you don't want these updates anymore, please unsubscribe here.
Powered by Revue