fbpx
Featured

Cracking the code, by Wunderman Thompson’s Adil Khan

Wunderman Thompson’s analytics lead, Adil Khan describes the merits of opening up to open-source marketing science.

Fifteen years ago, Avinash Kaushik from Google recommended a 10/90 rule for analytics budgets. For every $10 that you put into analytics tools, you should spend $90 on good analysts to digest the data and provide insights.

This idea always stuck with me. Any analyst would prefer to remain agnostic when it comes to platforms, instead of focusing on trying to frame the problem first before selecting the right platform for the task. Platform selection might be as important as the problem itself, since understanding the pros and cons of each platform (no platform is perfect) can help the brand decide if its needs will be met or not.

Fast-forward to 2021, and marketers are experiencing a proliferation of data from website behaviour, ad platforms, CRMs and e-commerce, to name a few sources.

The need of the hour is not more data, but rather making sense of it all to understand marketing effectiveness, predict behaviour and do deep analytics to bring out insights that can truly help drive business decisions.

Here is where open-source (read: free) platforms such as R and Python are making it easier to solve marketing-science challenges. A few examples:

Want to predict the expected lift in e-commerce performance by showing specific product recommendations based on items frequently purchased together?

Multi-channel attribution modelling is a key question that can help you understand channels that are driving conversions vs the ones that are upper-funnel and help influence the process to a certain degree. Instead of using pure historical attribution models, what if you could use a probability-based model to recreate all the conversion paths taken by users, temporarily remove one channel at a time from the mix and then understand the decrease in probability of getting conversions, thus knowing the incremental value of using the channel?

Let’s say you received a wall of text as consumer feedback and wanted to run natural language processing at a sentence level and address the negative sentiment behind specific issues in your next campaign. You could even standardise the feedback by running translations at scale and then submit just under a million requests per day to Google Cloud’s Natural Language API.

The answer to all the above problems? Yes, there is a package for that (code that does the heavy lifting in the background).

Are these free tools 100 per cent accurate and a perfect solution to all questions? That depends. Is your entire data 100 per cent accurate? Most probably not. Accounting for identity resolution challenges and cross-device behaviour, client-side tracking vs server-side, use of statistical modelling in Facebook Ads to even count conversions from iOS 14 users, among other challenges, the question now changes: Is your data usable (at a minimum)? Is it customised enough to be able to build business-specific trends? More importantly, can it help with improving marketing effectiveness in a scientific manner?

On the Lex Fridman podcast (he’s an MIT researcher on AI), a former US Navy pilot said an 80 per cent solution is typically good enough – because if you overthink it you’re behind.

I believe that we are in the same position. Marketers need to make quick yet smart, data-driven decisions. Solving for the 80 per cent solution always leaves room to apply adjustments in the next iteration and drive incremental improvements in performance.

There are incredibly smart individuals within the R/Python community who continue to write code that can help marketers in a variety of situations.

So, where should you start? Start with a specific business question and the code will follow. Trying to solve everything at the same time will almost certainly overwhelm the best among us. Data customisation is a fundamental piece of the puzzle that should not be overlooked. Here is where business requirements need to be mapped against a digital data collection approach to be able to answer the question, why are we collecting this data and how does it help the business? With enough questioning, robust yet rich data can be made available to your marketing science teams. Most importantly, invest in people and growth partners who are as keen as you to understand customers and anticipate their needs. Teams that can deliver data insights and solutions that enable you to create more memorable, relevant and effective experiences that drive business results.