David Myers - Product Manager at Marketo talks explains:
What is Predictive Content
How it works
Recommendation Algorithms
Case studies and real-life examples
How Marketo uses Predictive Content
Algos
in dev
Classification algo’s: (trying to classify which audience will create a click and which won't for each asset that might be presented to it and at what probability)
Logistic regression - in dev. based on attribute vectors.
decisions trees - were using boosted trees - in dev. boosted
Bar
uses Markov, SFO and metadat
Markov-style algo (looks one step back): Item-based. based on the last seen asset. finds assets that have the strongest connections between them
Collaborative filtering (sfo) - looks at the asset clicks, grades them based on key attributes (country, state, industry, search term, visit #). then , when a visitor comes in we find the asset with the highest combined grade for the visitors attributes. highest for sum of grade for location, industry
rule based - user attributes vs asset attributes. this is the fallback
rich media
new (discovery date), trending (14 day slope) , popular (forever)
Now lets talk about how we’re drinking our own champagne at Marketo
We have the content recommendation bar on our blog right now
And with the release of the rich media recommendation functionality we wanted to work that into our new website launch
A lot of work was done for the rich media recommendations was done in conjunction with the launch of our new website
Objectives:
Primary goal is to improve site conversions – as always
We also want to improve engagement with our content
Our first step was to select and align our exiting content to our new solutions
This was the objective of our initial rollout – each solution page has solutions specific asset recommendations
To do this we first identified which assets aligned best to each solution
From there it was just a matter of configuring the recommendations
Now we still have rule based campaign that will be running, such as promoting local events, changing content for different business types, and targeting certain key accounts and visitors. But we also want to compliment those with predictive content recommendations that will run in parallel, require minimum set-up, and yet still be very targeted to individual visitors.
Now I’ll show you just went into that process
The content discovery feature by itself provides a ton of value and I highly recommend enabling this feature whether or not you utilize content recommendations
This will let you see the performance of your content (whitepaper, case studies, blog posts, etc), both from an engagement and lead generation perspective.
Your code may look a little different depending on the template you select and any style edits you’d like to make
There are many configuration options available to match the style of recommendations to your website’s look and feel
Presto! Like magic, your content recommendations are live!
From start to finish it took less than 3 hours to get 30+ asset recommendations live on our website
Granted we had all the images done already but it was very easy to get going
Both the core RTP code and the recommendations code were implemented through our tag management system
The container div I added myself through our CMS.
That was it!
We have several ways of tracking the impact and performance of our recommendations.
Right on the recommendations screen we can see the clicks, direct conversions (if there is a form present) and assisted leads.
It’s a good practice to have a mix of gated and un-gated assets to recommend so not every will require a form completion – but the great part is that you can still see if a click on that ungated asset led to a conversion later on.
This give you a true picture of the value of your content pieces.
You can also drill into individual assets for a more detailed look at performance.
This trend shows downloads for our Definitive Guide to Digital Marketing
Our new website improved the consumption, as was intended
But you can see another definite lift once the recommendations were deployed