Machine learning and the road towards data driven decision making

• Data in Marketing
• Data driven decision making
• Machine learning
• Extracting knowledge
• Summary

Challenges

• Gut feelings
• Personal bias
• Fear of failure
• Accountability
• Silos

Half the money I spend on advertising is wasted; the trouble is, I don’t know which half. John Wanamaker

• Buzz word
• Data was always big
• Heavy investments
• Increased revenue
• Learnings?

For the majority of businesses it's really only a question of getting started with looking at your existing data.

Most businesses track their online traffic today using Google Analytics or something equivalent.

Therefor, the easiest data to extract is the online sales/traffic/behavior.

Let's look at a case study using Google Analytics data!

Non-paid

• Organic search
• Social
• Direct
• Referral

Paid

• Branded search
• Generic search
• Social
• Affiliate

Branded paid search

Organic search

Branded paid search

Generic paid search

Branded paid search

Display advertizing

Social advertizing

Display advertizing

Direct visits

Referral visits

• Your campaign may have a different impact on your customers depending on when they interact with it
• At the beginning of the purchase path, each channel helps customers gain awareness of your product or service
• In the middle, it creates desire and boosts interest
• And at the end, it helps to seal the deal

Though the previous section outlines the behaviors of buyers it fails to answer the key question:

How do I increase the convertion rate?

Basically it tells us all we would like to know about how the “king” is behaving but nothing about how we make him “king”

Enter machine learning

When a customer enters my store, forget me. He is king. John Wanamaker

Illustration

The monty hall problem

• Named after the game show host Monty who starred in Let's make a deal in the 1970's
• Three doors (1 car, two goats)
• Player selects a door
• The host opens a door revealing a goat
• Do you switch door?

\( P(C2|X1, H3)=\frac{P(H3|X1, C2)P(C2|H1)}{\sum_{i=1}^3 P(H3|X1, Ci)P(Ci|H1)} \)

• The future of all marketing
• Honest
• Unbiased
• Data driven
• Concise learnings
• Fast
• Robust

• Cross-sell / upsell
• Campaign management
• Customer acquisition
• Budget / forecast
• Churn / retention
• Fraud detection
• Promotion / pricing
• Content management
• Supply chain

• Needs data to learn from
• Gains experience as new data arrives
• Similar approach to learning as humans
• Discovers patterns hidden to the human eye
• Basically a statistically sane way to map a set of known variables to measured KPI's

• A new cookie is shows up on the Ad Exchange
• Should we bid on this individual?
• If so, how much?
• Machine learning can easily solve this for us given that we have access to the data
• The whole RTB and Re-targeting will experience an increased usage of more sophisticated algorithms

• There are a number of steps involved in modeling
• Define your KPI
• Select model type
• Split data into training and test
• Fit the model to training data
• Evaluate performance on test
• In this example we have a small sized neural network

The parameters adapt to the complexity of the data and improves over time

• In the beginning the model is stupid
• The predicted value is wrong
• As data arrives the model improves
• The predicted vs the observed KPI for a complex model

Take home message

• Machine learning is the bridge between data and decisions
• It is becoming more and more used in all aspects of marketing
• It can be used to leverage your domain knowledge
• In the end it allows you to make faster, better and more robust decisions with confidence

Use it for

• Content management
• E-mail frequency and personalized messages
• Media mix allocation
• Profit maximization
• Channel optimization
• Conversion uplift
• Churn minimization
• Path to conversion