I hate Black Friday, stop spamming me.

Artur Kiulian
Nov 26, 2017 · 9 min read

A case study on how successful companies use artificial intelligence and machine learning to break through the noise of holiday season.

I hate Black Friday.

So much noise, so much boring noise from every single company on the internet.

Face it, they spam you with no second thought.

Deals. Deals. Deals.

10%, 20%, 30% off.


Unintelligent, Ignorant, Brainless, Foolish, Vapid, Bland and many other synonyms to the stupidity behind that.

It’s almost impossible to filter out all this noise and craziness.

And you would guess marketers are tech savvy and would tailor their campaigns to your interests?

Your previous shopping behavior, emails, messages, right?

Cause all that deep learning artificial intelligence is everywhere nowadays.


Most companies don’t give a damn about that.

As a venture studio partner, I’m fortunate to work in the industry with many great founders/businesses and I see them making the same mistakes over & over when it comes to marketing.

This time was different.

And I’m fortunate to tell you why.

Introduction: Big Brands, Huge Demands

One of our portfolio companies, BoxBlvd, has been known for working with Stan Lee (creator of Marvel comics) and Cartoon Network.

Yes, you know their work, you’ve probably seen at least one Marvel movie coming out recently: Thor, Spiderman, Avengers etc.

So how can technology startup help these huge brands if they are already making billions?

The biggest struggle for huge brands in general is understanding their SuperFans.

The audience that is highly engaged and loves the brand so much that actually buys everything they put out..

We are not talking about 7,000,000 millions Facebook page likes that Stan Lee has.

Those are great for liking posts, sharing memes and social proof.

But that’s where it ends, especially with the absence of organic traffic for pages on Facebook.

Identifying and understanding that super fan audience is tough.

And that’s where machine learning can truly help. But as any successful use of artificial intelligence it needs data.

Brands need initial dataset on what audience to look for, where they live, what they like, how they talk, what makes them smile.

All of that.

To battle this problem BoxBlvd came up with the idea of crafting exclusive experiences and delivering them to the doorsteps of fans every few months.

It worked.

Thousands of avid fans participated, creating a baseline to what a true fan is.

Same thing happened to another client with the launch of a collectible box. Unseen before insights and data on the most engaging fans, willing to do anything for being a part of the exclusive experience.

Scaling Super Fans

BoxBlvd has been super successful at delivering these magical experiences but has been mostly driving the marketing firehose of brands, which is quite limited by the means of Facebook organic reach, Twitter algorithmic feed and other social medial channels, which don’t really work.

So they decided to run a truly intelligent Black Friday marketing campaign that is based on the existing collected data.

Identifying Audience

They’ve started with analyzing existing subscribers had with the end goal to acquire more of such audience.

First, they had to enrich data about the current subscribers with the data from public social media profiles. There are many tools out there that help with that, including AI-powered Hank.

Then they run a demographics classifier on top of all that public information.

The results were impressive and totally made sense to everyone.

So they went deeper and invested their resources into building an affinity analysis model, which would learn what those fans care about.

This analysis gave an amazing understanding of which personalities they enjoy (actors), what content they consume (movies), which entertainment channels they prefer.

Then it was a time to actually understand what kind of personalities those people are.

There are many tools out there to analyze personality from social media and most of them are using traditional machine learning techniques and regression analysis.

BoxBlvd iterated on the existing work in the field and applied it to the enriched datasets.

They used a five-score predictive model to identify personality traits and the results were astounding.

  • Conscientiousness: responsible, organized, persevering. Conscientious individuals are extremely reliable and tend to be high achievers, hard workers, and planners.
  • Extroversion: outgoing, amicable, assertive. Friendly and energetic, extroverts draw inspiration from social situations.
  • Agreeableness: cooperative, helpful, nurturing. People who score high in agreeableness are peace-keepers who are generally optimistic and trusting of others.
  • Neuroticism: anxious, insecure, sensitive. Neurotics are moody, tense, and easily tipped into experiencing negative emotions.
  • Openness to Experience: curious, intelligent, imaginative. High scorers tend to be artistic and sophisticated in taste and appreciate diverse views, ideas, and experiences.

(To those who are curious about the actual implementation, I’m including research paper at the end of the article)

So what’s next after you’ve nailed the understanding of your audience? You need to acquire more of that audience and try to convert existing one into more superfans.

And there are only a few channels that are still predictable in 2017.

Email is dead, is it?

An average open rate for emails within the Entertainment / E-commerce industry is around 15–20%, as the Mailchimp report says.

Which is much higher than what most of the marketing agencies experience doing email blasts to subscribers.

Here’s some real data, not so good as you can see.

But what if you craft the perfect personalized email tailored to the insights from the personality analysis, affinity machine learning model and enriched social media data?

Here’s where the magic of artificial intelligence comes in.

By clustering the existing dataset of subscribers and similar social media profiles BoxBlvd automated personalization.

Here’s what a convolutional deep learning model allowed them to uncover:

  • Clusters by Demographics (who are they)
  • Clusters by Location (where they live)
  • Clusters by Personalities (how to communicate with them effectively)
  • Clusters by Slang, Location & Personalities (how to relate to them)

In addition, it was important to understand when specific person is active during the day, since it heavily affects the open rate in general. Luckily, the enriched social profiles from Hank already had that.

By using these clusters and previous marketing emails the natural language processing model combined with the recurrent neural networks generated thousands of unique email templates.

The result: astounding 35% open rate.

Extremely high rate if you consider the amount of promotional crap everyone got on Nov 24.

Impressive and totally worth the effort.

Anatomy of the successful email template

So what made those email templates so successful?

Everything, basically.

From the email subject tailored to the audience, to the email signature of the person who emailed it.

Each email template was processed to maximize and provoke emotion. Of course, there were some emails that didn’t make any sense but that’s where humans edited them to sound more natural.

Email above includes:

Personal touch from a company representative

No doubt people ignore email campaigns filled with imagery and tons of text, it simply doesn’t work. When was the last time you paid attention to one of these?

Emotional phrases

Slang variations of texts based on personality and location

Yes, neural networks were trained on slang datasets for specific locations and generated those Texas phrases that no one understands.

Email template phrases were transformed by using word2vec word distances and dictionaries of most common phrases.

Personality adjusted phrases

Affinity relation

Fans love specific actors, channels etc, so quite easy for a segment of audience to relate to that.

Facebook ads don’t work

Most of the time social ads are expensive. Especially if marketers don’t understand targeting.

Though Facebook provides the most advanced targeting engine, ever. And it’s a shame not to use it with combination of all the insights BoxBlvd have uncovered.

But how do you go one step further and actually personalize the interaction with the ad? To avoid bounced traffic and meaningless website clicks. Again, artificial intelligence comes to the rescue.

Bots are dead

Facebook messenger bots seem like a dying trend to the most of the people on the internet, I thought that too until I’ve seen these results.

But there are so many clever hacks you can craft with them. Your creativity becomes the limiting factor.

BoxBlvd created engaging ad post requesting the interaction within the comments section to get the discount code for 25% off Black Friday coupon.

Why? What’s the point?

Besides the social proof and engineered engagement, obviously.

Attention is the goal.

Here’s the trick: after user comments under the page post, page can reach out to user by sending a message to his inbox.

This is a direct communication channel straight into the person’s messenger, and guess what…

It converts even better than email.

And the magic is that you can set up chatbot to understand the more complex requests like “When will I receive the box?” etc to actually help manage the customer service requests, as long as you dedicate time and resources to train your chatbot.

Obviously, you could still do that without any AI and the team of outsourced staff somewhere in Asia could manage your page. But what’s the point if you can benefit from automation and conversational interfaces.


We are at the very beginning of the wave of intelligent marketing. Very few marketers truly understand the power of data driven approach and it costs companies lots of money.

Though the fierce competition, noise and lack of attention slowly but surely demands the use of artificial intelligence across the marketing stack.

So lets embrace it, my marketer friends.

Yes, you.

The ones that sent 67 identical looking fancy emails about how I can get 50% off some stuff I don’t give a damn.

Thank you.

PS: Designer friends, please stop doing these banners, they don’t help :D

If you’ve liked this article, please share it with your marketer friends and business owners. Looking forward to seeing what they come up for Black Friday next year.


BoxBlvd, technology startup revolutionizing superfan intelligence for brands.

Stan Lee Box, exclusive subscription box of collectibles from Stan Lee, comes with an autograph.

Hank, lead generation service powered by machine learning to find emails and maximize open rates.

Predicting Personality from Twitter, research paper by Jennifer Golbeck, Cristina Robles, Michon Edmondson, and Karen Turner.

Algorology, special gift for my followers, access to my invite-only newsletter on how to do business with Artificial Intelligence.

This story is published in The Startup, where 263,100+ people come together to read Medium’s leading stories on entrepreneurship.

Subscribe to receive our top stories here.

The Startup

Medium's largest active publication, followed by +479K people. Follow to join our community.

Artur Kiulian

Written by

Serial Entrepreneur, Partner at Colab.la, Author of “Robot Is The Boss” www.robotistheboss.com

The Startup

Medium's largest active publication, followed by +479K people. Follow to join our community.