Customer analysis and insight can help companies in a variety of ways from simply understanding who are buying their products and services, to more advanced techniques such as calculating lifetime value and share of wallet. Furthermore, this information can then be used to find and acquire prospects who look like valuable loyal customers. Simply put, insight is business critical.
However, in many companies Analysis Managers have walked into roles “where it’s purely a support function doing management information and report after report.” While the intention is to provide proactive insight, because of changing businessdemands, the reality is that most analytic teams end up being more of a reactive/supportive function.
However, when analysis teams action repetitive requests, they are unable to spend time and provide the little nuggets that can be truly ground breaking, the little details that tell them not only which customers are leaving but why, when and what are all the different variables that lead to this event. Nuggets that might provide direction to the business by identifying new markets or potential threats.
To avoid having a team fall into this cycle, one recommendation is to structure the analysis team into two functions – one responsible for identifying and reporting on the key customer questions regularly asked by marketing teams, and a separate team responsible for proactively driving additional insight. Separating the two provides the bandwith and time for the insight team to instigate true insight.
Data analysis projects come with their challenges and the following tips provide some ways of getting around these:
1) Data availability
We have seen a number projects that have not come to fruition because data availability is only discussed after project sign off. Only then does it become apparent that the customer or market data required is not readily available from core systems and so timescales will have to be totally revised. Always check to see if the data sets that are to be the building blocks for your project are available. For example, in-bound call centre data is usually key for retention modelling and in most organisations is not linked to the marketing database.
2) Stakeholder management – know your audience
If the key stakeholders are from a business background, make sure the results are based on business relevent questions and not too technical. They might not be interested in code! Similarly if it’s a technical audience that has ultimate sign off make sure anything they want to know, such as the variables used and model strengths, are addressed. For example, for a marketing audience, presenting a profile of the top decile of a propensity model will be received better than a gini coefficient!
3) Regular feedback meetings
A common occurrence on these projects are instances where there is a huge gap in what companies percieve as their customer base to what it actually is. Regular meetings during a segmentation project and the sharing of findings as you go along ensures there are no surprises at the end of a project and concerns are addressed as you proceed. For example, a sophisticated clustering algorithm may identify a niche group of high value, older and low affluent customers. Can the business work with a such a group? Is it of sufficient size to tailor propostions to?
4) Rescoring and rebuilding
A short time after a segmentation or modeling exercise, comes the question of refreshes and updates. Be clear at the outset on terminology:
Rescoring or Refreshing: Using the current algorithm to update a customers score or segment.
Updating or Rebuilding: Revisiting the original project and refining or totally re-creating the algorithms.
Agree at the outset on the timescales and on how often these will takeplace. It may be relevant to re-score in real-time in some high customer activity organisations such as the telecoms sector, although quarterly re-scoring is the norm.
The typical lifetime (before update or rebuild) of a model or segmenation is 2-3 years, but it is prudent to check on a quarterly basis that the solution is still fit for purpose and the profiles and performance of the algorithms are still in line with what was seen when they were first built.
Analysis and insight are like a puzzle, putting the different pieces together to complete a picture. The key is having a structure in place that allows this to be a focus and managing information so when the final picture is revealed there are no surprises!