- In order to measure the success of an advertisement and take effective data-driven business decisions, we use key performance indicators (KPI).
- Exploratory data analysis was used to find two primary learning objectives.
- Optimal mix of creatives per funnel + publisher
- Impact of number of creatives in market on performance
- Random forests and light gradient boosting machine were utilized in predicting average funnel-specific click through rate (CTR), site visit conversion rate (CVRSV), buy flow entry conversion rate (CVRBF) and sales conversion rate (CVRS) for a given creative/advertisement prior to market launch.
- Evaluation metrics on hold-out data
- CTR: R-squared (0.59), MAE (0.051)
- CVR(SV): R-squared (0.65), MAE (0.12)
- CVR(BF): R-squared (0.53), MAE (1.15)
- CVR(S): R-squared (0.37), MAE (0.30)
Packages: pandas, numpy, matplotlib, seaborn, statsmodels, sklearn, lightgbm, tkinter, holidays
The data set contains over 13,000 creatives/ads, each with selected features and KPI results.
| Description | |
|---|---|
| Creative ID | Unique Identifier for each creative |
| Channel | Type of Media: Social ( eg. FB, Twitter) or Display (eg Amazon, Google) |
| Publisher | Where the creative is served |
| Funnel | UF, MF, LF |
| LOB | Line of business |
| Product | Item advertised in the creative |
| Theme | High-level concept for a series of ads - a theme has numerous versions |
| Creative Version | A theme can have various creative versions |
| KPI Audience | Audience targeted by creative |
| Price | Advertised value of the item |
| Price Placement | Position of the price on the creative |
| Discount | Advertised value of the incentive |
| Offer Placement | Position of the discount on the creative |
| Offer Group | Advertised extra incentives |
| Length | Ad length measured in seconds; no length for display |
| Asset Type | Video or display/static creative |
| Video Type | Characteristics of video creatives |
| Ad Size | Physical dimensions of creative |