In all cases we look to measure and inform today's omni-channel marketing process, customer journey, and real-time CX decisions with analytical insights and support tools. Each project passes through three main stages of analytics:
An extension of data visualization and informed by interactive data workshops, this phase focuses on confirming the correct data and informing the scope of analysis given the opportunities or limitations in the data.
Preliminary analysis and results shared in an interactive workshop allows the team to challenge and ultimately gain comfort in the results. Recognizing that modeling is not always an exact science and that data is not always perfect, this phase of analytics is transparent and iterative.
Confirmation Analytics (Descriptive, Diagnostic or Predictive)
Depending on the specific project, the final analytics provides a descriptive, diagnostic or predictive solution. This final set of results is held up against the original learning agenda. In cases where learning’s have not been achieved or are not of significance, a test & learn agenda is set-up to pursue better data for the next cycle.
In the context of insights, we don’t believe in a one size fits all solution.
At this stage, we work openly with clients to tailor the model structure and approach to best fit the agreed soon needs. All of our Modeling platforms are designed to be flexible and enable the analyst or user the flexibility to customize as needed.
The trade-off however, is that these tools require expert, well trained users to operate. Unfortunately, fully automated platforms that promise to deliver reasonable results without the user having to know the technical details are simply too good to be true. Achieving the optimal balance of speed and depth of insights requires not just the best use of analytics technology but also an appropriate level of skill and expertise in the team using it.
Below are several challenges in any given analytics project that we find ourselves wrestling with in order to be confident that the final result presented is both accurate and insightful:
do we believe we have identified true causation and not over or under inflated specific results which may be wrongly ascribed from the true source
is the analysis as simple as possible whilst still generating a robust insight, have we ensured no over-fitting and minimized false positives
can we verify results through hold-out samples and forecast test
does the Modeling approach suitably control for all influences on the analyses consumer decision
does the modeling approach fully capture and represent the consumer’s decision being made
is the chose analytic approach appropriate given the type and structure of data available, does it generate a usable and interpretable result
When settling on the optimal model approach, one should think through how the results will be used and how the recommendations will be implemented. For example, it might be better to recommend validating the results through a test & learn program before full implementation if there is any lack of confidence (either statistical or managerial) in the results. Providing a fuller context including benchmarks and case work to support results is helpful in socializing and encouraging adoption.