Late last year, I helped Australia’s leading creative agency The Works build, present and win a big analytics project to a major client. Below are three ideas that I believe were key to driving this successful outcome:
#1 - The analytical solution proposed was simple to understand
It’s natural for someone working with data science to try to solve problems using the latest and - sometimes - the most flexible (and potentially) more accurate models. It so happens though that in solving social sciences, particularly marketing problems, analytical models that can be easily understood have a bigger chance of being adopted. For this project, the client wanted to solve a marketing mix and attribution problem. The idea was to provide the client with a solution that allowed them to understand which marketing channels/activities were working or not and to what extent. While it was tempting to suggest a few more flexible models, we decided that an extension of the linear regression model that accommodated non-linearity would work nicely as a starting point. It provided an estimate of specific and important parameters the client cared about without making too complicated for them to interpret the results. Not surprisingly, it was really well accepted.
"Simplicity is the ultimate sophistication"-Leonardo Da Vinci
#2 - Possible known constraints were identified and communicated upfront
Solving really complex marketing problems with data and analytics always involves the idea that there will never be enough data available to capture everything needed to explain why certain things happen. For example, nobody would agree that looking at only TV ad spend is enough to capture enough reasons to explain a spike in sales of ice-cream in a particular month. There are many other factors that could affect sales. For this project, the client wanted to - among other things - understand what marketing levers they could pull to affect specific outcomes. For the non-initiated, that could suggest that any given solution should encompass all necessary data to explain all possible effects. Obviously, that is not even remotely possible and the solution proposed addressed this issue from the beginning. Given our years of collective experience providing solutions to marketers, the whole team pinpointed a range of things the client should be aware of when it comes to solving marketing mix and attribution problems. Highlighting potential limitations using existing data provided, listing a number of potential variables that did not exist but that could help explain effect and, ultimately, constraints and future expectations on applying statistical models to real-world, social applications.
"Honesty is the first chapter in the book of wisdom"-Thomas Jefferson
#3 - Relevant parties were frequently engaged throughout the process
It was very refreshing to work on a project where the client was heavily involved from the beginning and wanted to know what, who, how and why certain things were being suggested. It is very common for data science and analytics projects to be conceptualised, built and delivered in isolation to key decision makers and business/marketing people. This was not the case. The stakeholders on client side were so interested and so invested in making sure this worked that it made the whole process a smooth sail. In addition to that, our years of experience working with marketers and business people providing analytics solutions helped. We were able to align overarching business goals to the specifics of the project and also translate technical things without including (too many) ‘equations’ on a presentation deck which, among many other things, helped keep everyone engaged in the process.
"Coming together is a beginning; keeping together is progress; working together is success"-Henry Ford
In summary, for those working on an analytical project of similar profile or engaging with an analytics solution provider to solve a similar problem, below are a few points I’d suggest incorporating into your project:
1 - Favouring model simplicity can be an important winning strategy;
2 - Clarifying what can go wrong is as valuable as clarifying what will work;
3 - Keeping the communication constant and the language simple will favour a positive outcome;