- Data Science PyData
- Jose Luis Lopez Pino: Lessons learned from applying PyData to our marketing organization
Click on text below to jump to the corresponding location in the video (except iOS)
Our next speaker is Jose Pino. He works in marketing.
We have been trying to build marketing intelligence or marketing technology. We believe that a central part is the PyData stack which gives us the right tools to do it. We do traditional performance marketing, and the way we use the PyData stack is not magic.
First some braggingWe have grown 3x our marketing efforts.
We have reduced 90% of the time spent on creating a new ad.
Launched in 7 new markets in 6 different languages, without growing the team, and without expensive marketing software.
Two advices before startingYou won't go far without domain expertise
Marketing is a fast paced competition
What is GYG?Get Your Guide is a marketplace for tourist activities.
Marketing technologyCreate, control and target online ads
CreateAds relevant to the user
Within the limitations of the advertising service
Send them to the best landing page
ControlHow much should we bid?
TargetKeywords, keywords, keywords!
Reach relevant audiences
Three steps-Set up the infrastructure and tools
-Data retrieval and storage
-analyse, automate and train models
Infrastructure and tools-dedicated server?
-schedule requests, queue requests, multiple nodes, storage?-databases?structured?
-talk to other systems?
Data retrieval and storage-Data modeling?
-alembic (db migration tool) to make changes in your DB and keep track of them
-API migrations and sunsets: write better code
-rely on third party systems?
-optimize for speed?
-scrape without being banned?
-extract data from other systems of the organization
-users also input data
Analyse and automateWhen to use SQL? When to use pandas?-complex pivot tables with SQL?-load all my keyword space again and again in pandas?
Schedule queries to provide data for spreadsheets every day
Allow marketers to make changes that they can't do with any other interface
Some examples-What are the products that people show interest at this time of the year?
-Customer segmentation. What are the best attributes to segment them?
-How do I estimate the potential size of a market I don't know?
-What are the outliers in our accounts that need human attention?
-what are the most important keywords for a particular page?
-what are the products that I need in my marketplace?
Some ML examples-Are we going to sell out this product?
-Sentiment analysis on customer reviews
-regression model of our ROAS
-how to cluster our adgroups to make decisions on them?
Sentiment analysis reviewsUsed a library
Site analysisApproach was too complicated.
And we are still learningHiring and growing the team, but difficult to find the right people.
How to measure success?Are we really being successful?
Takeaways from this talk-Marketing is not the latest buzzword, but its fun to do for data driven people
-technology can have an enormous impact on the marketing results
-and the PyData stack provides tools to do it
QuestionsHow do you assess the value of any individual customer in a way which can be validated?Watch the revenue that customer is giving me, and also the cost of clicks for this group of customers. We try to make some predictions, but sometimes there is not enough data. Clustering helps. Video outline created using VideoJots. Click and drag lower right corner to resize video. On iOS devices you cannot jump to video location by clicking on the outline.