Since 2016, the interest for Artificial Intelligence has increased more than twofold worldwide.
We can see that in Google trends.
There definitely is amplifying buzz around AI although it’s not a new topic. AI has been around for decades but as Ray Kurzwell says “The singularity is near”.
How much longer will it take to reach this singular point where human and artificial intelligence merge?
With a MarTech stack which now includes thousands of growth analytics SaaS solutions, AI for growth is becoming increasingly complex. Yet, growth marketers need to learn how to make new artificial entities part of their growth arsenal and an integral part of their growth team.
Today, as we iterate on growth, a lot of the growth process is still manual. Let’s take one foundational aspect of growth, for example, high-tempo ICE as described in this short video by Dan Martell, most growth marketers use a spreadsheet to align their teams around growth experiments.
Sean Ellis, one of the fathers of growth hacking, and his team at growthhackers.com, recently launched a software called NorthStar to help accelerate and streamline this cross-functional growth process.
What happens in the near future when AI is more deeply integrated into our growth teams to act on all levers of the AARRR framework?
What is AI?
Before sharing real-life AI growth actions I led in my life as a mobile SaaS startup entrepreneur and as a growth marketer, let’s start with getting a clear definition of what AI stands for as of today.
Dr. Raj Ramesh shared a very easy to understand YouTube video that does a great job at explaining what AI is about for now – and the context around many of the currently trending buzz words such as Machine learning, Deep Learning, Natural Language Processing, Convolutional Neural Networks and more.
What is AI – For People In a Hurry!
To keep it simple, there are two main sections of AI.
One that is based on symbols, symbolic learning and one that is based on data, machine learning.
The former has emerged as early as in the 19th century, as explained in this paper by Inman Harvey. A computer at the time was like a clerk employed in an accountancy or insurance office to perform immense and tedious calculations by hand and by following detailed set procedures.
Fast forward to 1997, and Deep Blue is the first form of symbolic AI to defeat chess champion Kasparov.
We can give a lot of credit to symbolic inroads to help raise the overall AI awareness. Yet, such artificial entities focus on crunching systematically through all the possibilities. Think the finite universe.
Today, data sets are becoming larger and more dynamic. Marketers can collect real-time data at each touch point of their customer journeys. Discover patterns that correlate to their north star metrics to ideate and act on an increasingly larger and more and more data-informed set of growth experiments.
The art of growth marketing is now turbocharged with actionable data learning.
Below are some real-life examples you can act on:
Cross-platform web + mobile AI for automated customer segmentation and messaging
In a previous article on mobile app growth, I had touched on how you can leverage mobile engagement analytics to spur growth.
What I find quite fascinating with an AI driven cross-platform analytics solution such as Pyze, is the simplicity of leveraging behavioral analytics to deliver personalized messaging in no time to each behavioral segment.
The first layer of AI in the Pyze experience is auto-segmentation AI.
Just by dragging and dropping segmentation dimensions, in a matter of seconds, you can make deep segmentation actionable for immediate marketing right away.
The segments in blue, represent the Silver subscribers from the U.K. organized by increasing levels of attrition.
We can act right away on each auto-segment by clicking on it. We are now crafting an in-app message directly in Pyze.
On the left, we have the various elements of the in-app message. We can edit the picture directly in Pyze, enter the title, core message and add deep-linking CTA buttons or even CTA linking to external webpages which can be useful for survey collections as an example.
On the right, in real-time, you see exactly how your in-app message is going to render.
You can do the same if you elect to send a rich push notification. In these advanced push notifications, you can include a full picture and even a video if you want!
The other interesting layer of AI in the Pyze experience happens at scheduling.
Pyze leverages its big data mobile graph to offer the “Best time to reach per user” scheduling functionality. They use machine learning to custom send each in-app message on a per user basis to maximize tap-through rates for both in-app and push messages.
Last but not least, Pyze boasts a solid visual interface for reporting engagement data in a variety of ways in just a few drag-an-drop moves.
App Installs by Day and by Manufacturer via Pyze Data Query Builder.
Social AI for automated B2B and B2C lead qualification, activation and growth
Lead qualification, activation and growth on Instagram
Followadder is an interesting social AI web app for Instagram. It can help you achieve several aspects of social engagement and growth on top of the things that only humans will be able to do. Even in the future, it may help with Instagram Live engagement including split-screen dual broadcasts.
I found it helpful in discovering product-market fit in the early stages of mobile app experiences where the must-have persona has a strong millennial component.
For example, if you go after a niche market, you can target the followers of the instagram influencers in that niche.
You can DM a picture automatically to your followers.
The image below demonstrates how I set it up for our app.
You can also test hundreds of messages to quickly iterate and focus on those that trigger the most responsiveness from your core market.
The image below is an example of a portion of a test messaging matrix’s that I used some time back to help simplify engagement comments and surface qualified leads faster.
As you combine human social interaction with relevant AI automation, you get to spur social engagement to levels that are stronger than without AI collaboration.
This form of Social AI is symbolic. You input your segmentation, targeting and positioning data for your brand. AI then goes about finding the corresponding “symbols” within the Instagram social graph and automates some of the work that many still do manually.
Yet, there is a lot of room for improvement.
With a tool like that, you can target relevant keywords included in the Instagram usernames, or relevant hashtags, or followers of influencers in your niche and more.
However, you cannot target hashtags. At least I haven’t found a tool that does that yet. Being a developer and having worked on the Instagram API years ago, it wouldn’t be that complicated to create one.
You could really zero in very precisely on must-have personas to validate product-market fit on your first-time user experience.
Here is a quick example with simple figures.
Imagine you only want to target florists who sell orchids.
#florist has 6.5 Million posts on insta.
#orchid has 2.7 Million posts.
If you can only target #florist and #orchid, the odds are that you target people posting about orchids without having nothing to do with being a florist are high. And so are the odds of targeting florists who do not sell orchids.
Now if you target “orchidflorist” as part of the insta username, you get super targeted leads, but you only skim the surface of a much larger addressable pool of orchid florist leads on insta.
See, you only get a few dozens of orchid florists when you search for “orchidflorist” as People.
Even if there are limitations with this type of symbolic social AI, it can clearly move your growth needle by far. We can also understand the power of targeting social graphs and anticipate what will happen as those graphs keep growing.
This will lead to unprecedented levels of targeting possibilities paired with very sharp personalization that will make everybody blink to determine if a human or an artificial entity is actually communicating.
This leads us to personalization API where you can dynamically change your content based on detailed target customer information.
Personalization API to generate B2B leads
Many social APIs are not exposed. Meaning, the social network who hosts the data doesn’t make it available to the developer community to build apps on top of their social graph. Or if they do, they only do it partially.
Nonetheless, there already exists some interesting AI that can help you leverage the “personalization API” angle without actually having to know any coding skills, nor having to use an existing social API.
A very human artificial B2B growth agent
LinkedHelper is a social AI solution for B2B growth.
I find it interesting as it actually mimics human behaviors and automates them. You can pretty much automate anything you would do using a B2B social network such as LinkedIn. This includes writing and sending extremely personalized messages to anyone in your 1st, 2nd or 3rd-degree network.
Here is a tangible example of how you can leverage this AI agent to generate highly targeted B2B leads at a fraction of traditional MQL and SQL costs in B2B settings.
We are going to target the “Real Estate broker” persona in our 2nd-degree network, in a very personal way. We leverage the name(s) of our common connection(s) automatically and put forward our core value proposition to make them a 1st-degree connection first. Then, we can start an automated nurturing strategy to run a demo of our app in their office.
Step 1 – Search for “Real Estate Broker” and activate deep filters
I am using the “All Filters” functionality of LinkedIn Search to make sure to target 2nd-degree connections from my “Real Estate broker” search. I also want to make sure that their current title includes “Real Estate Broker”. That way, you refine the precision of your core persona targeting.
And here we go, we just surfaced 17,926 prospective leads in the exact target persona we are looking to market.
Now we need to assign them to a collection list that we could re-use in the future.
Step 2 – Collect “Real Estate Broker” leads
We simply hit collect:
As the AI starts collecting prospective leads among our 17,926 hyper-targeted prospects, we can start setting up a personalized invite message.
Step 3 – Write personalization invitations
Now it’s time to craft a personalized invite with relevant data from your LinkedIn social graph to boost engagement rate and get a maximum number of qualified leads to connect with you. Time to reel them into your growth funnel!
Click in the body of the message for the pop-up message editor.
Now we start typing a simple invite and include a few personalized data points such as the target connections first name.
To make the invite more personal, relevant and likely to spur a higher response rate, we are going to add the name of a mutual connection. We do this by leveraging the conditional test with a very simple “IF-THEN-ELSE” statement made available by the software.
Just click the conditional statement itself, it will create a new conditional sequence inside the body of your message in the editor. Here, I am displaying the text “including {full name of common connection}”. This should work all the time since I am targeting 2nd-degree connections which by definition have at least one 1st-degree connection in common with me. Yet, just in case a common connections full name is not returned, I populate the ELSE part of the conditional statement with a dot “.” to properly punctuate the sentence and end it in this case.
The image below is what the conditional statement in the message editor looks like.
As you save your invite, you get to preview it if you want with staged data.
It’s almost funny actually!
It’s now time to send our personalized invites to our LinkedIn prospects.
Step 4 – Send your personalized invitations
As we were setting up the personalized invitation in the message editor, the collection AI worked magic. We have already collected 840 highly targeted prospective “Real Estate Broker” leads.
This is more than enough to get started and to get a feel for the response rate on that target persona. It’s best not to send more than 150 invites per day anyway, so we have a few days of prospective leads to contact ahead of us.
It’s time to stop collecting and simply press on the invite button to start sending out our personalized LinkedIn invitations.
The AI starts typing the first invitation at natural human typewriting speed and we can verify that the elements of message personalization are properly populated.
We are now making sure the conditional statements we created, works properly.
And it does!
And the invitation goes out on its own.
Done!
Machine learning in conversational intelligence to increase sales
When I started my marketing career, I marketed the French and Italian subscription of HDTV channels for a leading satellite HDTV network in the U.S.