Thanks to everyone who has already completed our latest analytics poll, on real world usage.
If you haven’t yet recorded your scores, there’s still time (you can complete it here). But, with over 35 votes already in, I thought it worth sharing some interim results with you.
During the same week, I was reading a report from IBM as to the ‘real world use of big data in financial services‘. Although a number of sources in this report were from older surveys, circa 2012, their work with Said Business School in Oxford looks robust & relevant.
So, as an interesting backdrop to the results of our survey, here is a graph from IBM (comparing % of firms in Financial Markets using different analytics capabilities with a Global cross-industry selection):
A number of key differences are striking in this graph. I’d call out:
- Higher use of Data Mining within FS (and over 75%)
- Higher use of Data Visualisation within FS (and over 70%)
- Higher use of Natural Language Text Analytics within FS (and over 50%)
So, compared to other sectors, whose use of analytics appears still more focussed on simpler query & reporting – will we find sophisticated application of analytics in our survey?
The answer is not quite. As I normally discover when talking honestly with clients, their routine use of analytics is nowhere near as sophisticated as you would think when reading material from IT vendors. Now, I’m not suggesting any of the kinds of reports shown above are disingenuous. But we shouldn’t ignore the fact that there is a vested interest for analytics suppliers to be able to play on our ‘social norm’ bias (or ‘keeping up with the Joneses’ for business).
Without further ado then, what are early results from our latest analytics poll showing?
Our first question focussed on segmentation. We asked, ‘Which consumer segmentation approaches do you use?‘ Here are your scores:
A modern trend is shown here. Since the heydays of attitudinal, demographic & value-based segmentations, more awareness as to the limitations of each, appears to be driving hybrid solutions. Popular combinations include value-based with behaviour, geo-demographic with attitude & a combination of value, behaviour and attitude. However, it should be noted that often one of those approaches has been dominant in defining the clusters, with others more used to refine & enrich within that framework.
Given many organisations look to deploy their segmentations through CRM or Marketing Automation systems, it’s also not surprising to see next most popular option is a value-based segmentation. Such an approach is often easier to explain to frontline colleagues & appears more commercially focussed. The downside can be only a superficial understanding of the customer & poor evidence for FCA et al that you are really focussed on understanding those customers’ needs.
So, on to our next question regarding descriptive analytics (inc. profiling). We asked, ‘Which descriptive analysis approaches do you use?‘ You voted for:
It’s not surprising to see votes for Business Intelligence basics, like summary tables & graphs. Given the portion of our readership leading research teams, I’d also have been surprised in inferential methods weren’t scored highly. Good to see descriptive data analysis included in votes but perhaps surprising that univariate analysis is lower. The votes for dynamic data visualisations suggest some readers are operating with more advanced software than others. But the clarity of data visualisation design can be more important than interactive vs. static.
For our third question, we focus on customer behaviour analytics. Asked, ‘Which customer behaviour analysis approaches do you use?‘, you confirmed these:
The most popular votes are perhaps not a surprise again, with analysis of buying behaviour, media response & customer retention being expected by most marketing teams. Those research leaders in our sample would also be expected to identify research stage behaviour by customers, touchpoint interactions, advocacy & brand tracking.
What surprised me in these votes was the lack of any votes for channel switching behaviour. Perhaps data barriers make this harder for some insight teams, but in our omni-channel world, key insights can be found by the channel switching exhibited by different customers.
Two other types of behavioural analysis that I expected to poll lower in Financial Services are product purchase order (more commonly used by retailers) & relationship networks (often analysed by telco teams). But both can be of real benefit if FS insight teams do complete them. Understanding the reality of foundational products for your brand & the order of further product recommendations which make most sense to your customers can help refine Next Best Action approaches. Being able to identify ‘mavens’ or ‘hubs’ within your customer base, whose networks of influence are considerable, can help achieve targeted recommender marketing & opportunities for co-creation.
So, to the interim results from our question on predictive analytics. We asked, ‘Which types of predictive analytics approaches do you use?‘. Your scores are:
Given the almost ubiquitous equating of ‘propensity models’ with linear or logistic regression, the top two scores are not surprising (plus Occam’s Razor would have to encourage linear regression when prediction can be simplified to linear relationship). It is at least encouraging to see that some insight teams are also making use of decision trees & time series forecasting when those are appropriate. I’ve seen plenty of commercial applications where the transparency of decision trees is more useful than slightly higher predictive power of a logistic regression model.
The zero votes for machine learning algorithms reminds us again to not assume all the noise about Data Science is yet being translated into many firms day-to-day work. I’ve also seen forecasting more generally being neglected in many analytics teams in comparison to just fitting models to current or past behaviour. Perhaps that’s a missed opportunity for your analysts?
One technique I would encourage analytics teams to look at again is Survival Models. Coming from the world of Pharma & statistics for drugs trials, it provides statistical methods for estimating the likely survival rates of members of a population given available variables. It offers the potential to improve cancellation/retention analysis, beyond just predicting those more likely to exit or highest propensity to exit at a key milestone (like first renewal), to start forecasting tenure for cohorts. If someone can crack this & get it working for either better targeted comms or keener pricing with this input to LTV models, it surely offers real profit potential.
Anyway, that’s how the score are looking so far. If that doesn’t reflect the reality for your analytics team, please vote.
I’ll keep monitoring analytics poll scores over coming months & will publish an update if they change markedly. Hope that helps.