Video and Content Streaming Marketing Analytics and Video Streaming Marketing effectiveness
‘Remember, the most important part in Prescriptive Analytics is to ask even smarter questions, planning the next step, challenge the results and think further – always ask: “And Then What?”‘Antal Sofalvy – CDO, Certified AI Coach, ML Trainer
OK, then let us pick three elements for the basics of streaming analytics first – and we’ll discuss the rest in person
- Automated Streaming Marketing Insights and Analytics
- Advanced Content Analytics – Content engagement
- Subscription Analytics: Subscriber journey, features, metrics and KPIs
Track, model and guide Subscriber journey, increase Subscriber Lifetime Value – an example
Just to demonstrate how brilliant an Advanced Analytical solution can be…
Streaming Business Questions:
- How subscription revenue can be distributed among content buckets (Show ROI)
- What are the related content(s)? What shows consumed together?
- Who is going to leave our service? (Churn prediction)
- Aggregate who watched which show for how long
- Enlist neighbouring shows: “Association Rules” [If A and B then C]
- Figure out who will leave: historical data analysis
- What shows bring Subscribers – are there any “Exit” shows? (good for Customer journey model)
- Property clustering accordingly (vs Time): First viewed / Last viewed / Middleware
- What about free planners (same rule should be applied)?
- Use watch time in model – should it be normalised then?
It looks simple however gather and cleanse your data usually challenging. Do not forget the Agile BI method and the Disposable Analytics – these help us to draw your success blueprint – fast! Of course there are Video Streaming Analytical related decisions: during the Advanced Analytical process you’ll need to introduce sensitive / situational / vogue rules, like:
- What kind of watch-time do you want to consider in revenue distribution? How do you define / select “fully watched” content? Do you want to adjust this by show length?
- Do we want to honour / merit “Disguise” (disliked shows that were not watched fully), Skimmed content (watched just for a fraction of runtime)
- Would you use the same weight for Heavy users / Not so heavy users (100 show vs 2 content pieces per month)?
- What time-frame shall you use? It can be 30 days, but which dates (sync with Subs Period)?
Related content, neighbouring content: shows consumed together / after one another
- What level / granularity?
- Does it change over time?
- How to represent the results?
- User (ID) / User clusters / Unique users
- Content / Episode / Series / Show level
- Watched? (fully / partial / sampled): introduce normalised metric: % of full show runtime watched
- Skipped / Skimmed: Started not to watch certain content
Streaming Content Analytics – An Agile Prescriptive model
Goal: extend total subscription time; foster to renew subscriptions, increase Lifetime Value
Secondary: identify content marketing patterns (recommend additional content + timing of this)
- Track user’s content journey and intervene when potential success patterns break
- Identify standard and non-standard usage patterns (multiple non finished video then radio silence for example)
- Guide your Subscribers through your optimized product journey
Why subscribers leave?
- Consumed all interesting content (content journey comes to an end)
- Content became irrelevant (aging)
- Cancelling shows
- Technical difficulties
- Financial decision
- Paused subscription (chance to renew)
- End of Free Trial
In real life all mixed up, usually no dedicated flag to identify which happened… On the other hand we can generate some clues. Pretty good ones, actually… Let us concentrate to the first two content journey related stuff now.
Example Video Streaming Analytics Hypothesis
Subscriber leaves after a certain chain of content consumption and consumption behaviour can predict churn
Here comes some Data Science related questions (features to introduce):
Who is in scope? (which users / rows will be used for mining)?
- Clustering: frequency of consumption
- Clustering: number of incomplete video views
- Above Cluster changes in time (significant or not?)
- Full video views vs non complete ones – how they define completed view? What is the length of a property vs full view (5-8sec skipping vs 24:35 out of 25 minutes)
- Will an above typical number of incomplete views predict churn (?) – raises the importance of Normalisation (75 out of 200 vs 150 of 5000)
- Association Rule vs Sequence Mining (order is important or not?)
- Reduce Association rule granularity: introduce / create meaningful sub groups of content (great clustering case or classification works) even on Content level (but determine age group demography / gender & co)
Features to track:
- Typical successful video usage patterns (video views / content journey of churners / non-churners)
- Create such clusters and generate typical KPIs
- Track current measures against these
- Consider Streaming Video quality features?
- Does the account used by one person? (according to research: 60+% shares account; 40% uses other’s credentials – 25% family + 15% outside home, etc)
- Everything changes in time – so you need to analyse time chunks (optimized analytical windows)
- Consistent content consumption patterns
- Consistent churn flagging (lot of parameters can be considered)
- How to handle different Streaming Metrics? (Video Views, Video Starts, Video Completed, Watch Time, Total Time, etc)?
- Country/Region/Device (any pattern differences?)
- Connected views (started day1, completed day 3) – are these marginal?
- Behavioural glitches (holiday, sickness, etc)
- New content
- … and another 100 features D&A can help you to identify, optimise and utilise
Video Streaming Analytics Content journey(s)
Parameters (enrichment, feature engineering):
- What bought / consumed first, second, third, etc
- Time between actions / products; time-frame of actions
- Content features: Type, Genre, Season, Episode, Length, etc
- Content type: meta categories (as determined above)
- Consumption and usage patterns (repeat, new content); longitudinal features (lot over weekend, etc), periodicity, etc
- Content library changes
- How do you define “consumption”? Video views for how long? How do you optimise / find “right” length?
- [add yours]
Data Science: Association Rule Mining on the featured Video Streaming Dataset
Identify and map recurring patterns (Clustering + Association Rule Mining)
Product / Content Bucket A -> Bucket B -> Bucket D, meanwhile in another constellation / usage pattern: Bucket B -> Bucket C –> Bucket E
This approach can identify which video content watched in common, which order, what are / can be the related contents … and so on. All in all pretty talkative approach. [Hint: if any of the identified trajectories missed, journey can be triggered back with proper marketing action]
Moreover this will even identify Exit points (last played shows and potentially the behaviour pattern in your next analytics stage).
Video Streaming Prescriptive Analytics: If „chain” breaks, action can be taken immediately
- Reminder of potential content fit – the “what” (Product Clustering), model even says the “when” [Semi-Prescriptive]
- Validate with Churn Prediction (Time series analytics) and create synergy with direct marketing
- Recommend relevant (= similar users’ consumption) content – offline recommendation engine
Of course it is just the tip of the Content Streaming Marketing Analytics Iceberg
There are a lot of question we will help to answer:
- How to track marketing results effectively?
- How to measure the impact of Video Streaming Analytics above?
- As soon as Streaming Marketing actions taken based on Churn Predictions the analytical window parameters change (you disturb the data space – false positive rate will decline…). Accordingly model partially needs to be updated, actions, churn flags need tracking – yes, you are right it is a never ending story…
The good news is that this is not new for us, we know the data challenges and the best practices to deliver impactful insights fast. Usually all Streaming data processing, data wrangling and Content Streaming Advanced Analytics solutions can be scaled and automated for supporting effective and timely prescriptive marketing actions and create trusted intelligence.
At D&A’s we love to deliver such innovative solutions in Streaming Service Marketing Analytics.
Which Competitor of yours are doing all the above right? Want to be the First?
Have appetite to win your streaming market? Wanna innovate your business – now?
Data & Analytics Predictive Content Streaming Marketing Analysis Team 2016-2017.