Attribution Modeling
- 1. Exploring New Methodologies for Attribution in a data sparse world. Marketing Attribution Modeling Michael Reese
- 2. Traditional Attribution Approaches and Shortcomings Positional The customer purchase was the results of the (first, second, last) tactic used. Linear Everything we did had equal impact on the decision to purchase. Time Decay Our tactics accumulated to create a critical mass that triggered a purchase. - Fine for digital activities, but mutes the effects of other channels. - Digital fails if customer looks, decides but is not identified. - It’s arbitrary to assign 1st, 5th or 100th tactic as the deciding factor. - Data often lacking or does not validate the numbers. - Difficult to really control the path the customer takes in digital. - Defies credibility to believe all tactics and collateral have equal impact. - If change is mandated, where does the model identify opportunity? - Outdated approach, few in business believe all influences are equal. - Renders data you do collect as irrelevant. - If all touch points are equal, why didn’t they convert on earlier ones? - Better credibility, but still difficult to identify opportunity. - If change is mandated, where does the model identify opportunity? - Still assumes all customers follow the same path, difficult to effect change.
- 3. Problem: None of the approaches take into account the behavioral history of the customer.
- 4. Using 3rd party behavioral metrics to understand the mind of the customer. Response rate to digital Response rate to magazine Response rate to newspaper Response rate to radio Response rate to TVResponse rate to email
- 5. Digital Receptivity: Measured by metrics gathered on the amount of times a person takes an internet offer. Classification: Highly Unlikely
- 6. Digital Receptivity: Measured by metrics gathered on the amount of times a person takes an internet offer. Classification: Highly Likely
- 7. This creates a series of measurable outcomes: Extremely Likely = “X”% Highly Likely = “Y” % Likely = “Z”% Unlikely = “A”% Highly Unlikely = “B”% Extremely Unlikely = “C”%
- 8. There are multiple segments: -Digital Receptivity -Magazine Receptivity -Newspaper Receptivity -Radio Receptivity -Television Receptivity -Email Receptivity
- 9. We get an overall picture of the grower: -Digital Receptivity – Highly Likely -Magazine Receptivity – Highly Unlikely -Newspaper Receptivity – Extremely Unlikely -Radio Receptivity - Likely -Television Receptivity – Highly Likely -Email Receptivity – Extremely Likely
- 10. 2015 Order: $125,000 Total Customer Metrics Joe Grower - Influences: Digital = .25 Mag = .13 News = .12 Radio = .12 TV = .13 Email = .25 Analytical Attribution Modeling Digital Mrkt Magazines Information and Insight Cycle Newspapers Radio Television Email $31,250 $16,250 $15,000 $15,000 $16,250 $31,250 Customer Journey / Purchase Behavioral Attribution Modeling: Making sense of consumer behavior.
- 11. Turning data into insight: Visualization approaches. Dig TV Rad Email Mag News Dig TV Rad Email Mag News By comparing rolled up income by channel, you can compare against your media mix spend at various cuts (geography, customer segment, customer share) to gain insight Into optimizing your media spend dollars. Income Expense