So, you are a data scientist? But I thought you worked in marketing!
Every. Single. Time. This was the reaction that I got when I told people I was a Data Scientist in the marketing department of a travel giant. Recently, I have come across a lot of articles on the internet mentioning “use of AI and ML driving optimal investments and sales on a performance basis”. But not enough that would demystify the exact applications of data science in this space. AI and ML are very broad terms to explain all the ways data science is helping businesses optimise their marketing spends and get a competitive edge in the industry.
So, let’s talk about it. With a real life use-case — understanding the business problem in performance marketing from a data science point of view and then to see how the tools of “AI and ML” can help solve it. Hopefully that will help demystify the role of a data scientist in digital performance marketing space.
Quick Pit Stop for Glossary
- Click — click to the platform: website/app/web-app
- Conversion — incoming user making a desirable action on the platform like click, purchase, scroll — whatever brings value to the business
- Click-in — when a user click on something to reach your website/app
- Click-out — the last click a user makes that take them to another website/app
- Search Engines: Anything that you use to search the web. But here we talk about mostly Google, Bing, Yahoo, Yandex and such.
Search Engine Marketing(SEM) is one of the performance marketing channels that different businesses use to drive more traffic to their products. SEM ads are the ads that you would see before the organic search results after you hit enter on your search engine like Google, Bing, Yahoo etc.
Most SEM ads work on an auction system, where you have to place a bid on each of the search term relevant to your business and if you win a position in the auction(there are usually 2–3 SEM ads per search, or more based on your region and the search engine you are using), then you would pay the cost equal to your bid or lower depending on which auction system the search engine follows.
Now, with a big business, you might be dealing with millions of search terms you would want to bid on — but you also want to bid to optimise both the profits and traffic volume(visits to your website/app/web-app)from each of these search terms(keyword). One way to do it is to set and update the bids for each keyword manually. Other way is to look at the “AI and ML” solutions to help you optimise your investments based on the value each keyword is bringing to your business, adding in some strategic business decisions to focus on volume of traffic vs the profits you make on your investments. So, let’s get started then!
“AI and ML” Solution
Let’s see this business problem from data science lenses (I can feel the data scientists glowing with excitement here) 😆
Just add AI to the problem and the world is sunshine and rainbows again! 🌈
You have all the cost and click data on each of these keywords, you have a clear business strategy for performance marketing and all you have to do is to use an out-of-the-box shiny algorithm to predict the value each keyword brings to the business and set your bids accordingly. For millions of keywords.
Easy peasy, right? Almost. 😉
Let me highlight that it’s a fairly challenging problem due to multiple reasons:
- The sheer size of the data you have to mine push you to go “Big Data” and need whole infrastructure set up for you to even start modelling the keyword values
- The sparsity in the data itself adds more noise than signal for your models to learn from — you have to get really creative in how you would slice, dice and aggregate the data to get some reasonable dataset to work with
- The sparsity is two folds here: due to the long tail nature of the search terms(some search terms are more popular than some other obscure ones) and due to the conversion ratio itself — not every click from these SEM ads converts to a purchase which is the eventual indicator of the value for all parties involved: user and the business itself
- The evaluation metrics is very tricky to pick due to the nature of the data — too sparse and too volatile — that you have to be very careful in designing the metric to help you make correct modelling decisions: Do we reward predicting 0 for keywords with no recent clicks? Do we still predict a value in order not to kill the traffic from the keyword completely? If so, what should be the error metrics which makes sense in this 0 inflated data-set?
- With a workable dataset, a sensible evaluation metrics and certain modelling decisions, then you go ahead and try various modelling techniques that suit the use case and compare them offline before narrowing down to a couple to test them live in test and control set-up 😄
And once you have satisfactorily(*laughs*) got to a model to predict the value per keyword for your business, you still have to think about:
- Modelling effect of competitors bid changes in the auction? Remember the second-price auction systems?
- Modelling seasonality? In different regions. Different for summer or winter travel keywords. In Corona times vs Usual times. You got the point.
- How do you know if you have reached an optimal level of investment?
- How do you know you are not missing out on opportunity to bring more users to the platform?
and so on.
But by harnessing the power of “AI and ML” 😏 — these problems can be tackled one by one by making data based decisions and getting closer to the business goals.
Impact of the solution on the business
And it’s not just sunshine and rainbows, but this has clear impact on the business — by leveraging the big data infrastructure, data science capabilities, statistical testing and analytical monitoring — we help to continuously optimise the marketing investments on a scale of hundreds of millions of euros per year and bring more value to the users, and the business.
Nothing gets the geeky data scientists out of their bed quicker than a boggling data challenge which has such a huge impact on business and millions of users all over the world!
The use-case in this post is just one example of how data science or “AI and ML” solutions are applied to optimise marketing efforts. There are many more such exciting applications to it: Customer Lifetime Value(CLV) modelling, specially for SaaS businesses, Attribution models for different marketing channels, Sales/Traffic Forecasting etc that all comes with their own set of exciting challenges! 😉
So next time you hear Data Science and Marketing together, you know at least some dots that connect! 😎
Originally published at https://lalwanivarsha.medium.com on August 21, 2020.