We did an article series on the 9 indicators of digital analytics transformation recently.
If you have read them, you might be wondering why we didn’t touch on data analytics at all.
That is because we felt that data analytics in itself is a powerful function and deserves a separate post and so, here we go.
The volume and the different types of data generated by small and medium-sized brands can be truly staggering.
This can typically include data from marketing websites, post-login websites, mobile apps, social media, search marketing campaigns, display campaigns, CRM data, survey data, POS, loyalty systems, etc.
This gives marketers a great opportunity to derive insights and optimize their marketing programs to maximize targeting efficiency.
Different marketers are at varying levels of maturity when it comes to data analytics.
In this article we will talk about some factors that indicate the need to transform the way you leverage analytics for marketing.
- The nature of data
- The type of data used for analysis
- The reducing effectiveness of existing models
- The applications of predictive analytics in marketing
Let’s get started.
#1. Your data is in siloes
The number of data sources that marketers use, is proliferating.
Source: Salesforce State of Marketing Research 6th Edition
The greater the number of sources, the higher the opportunity for analysis. However, you know that this isn’t so straightforward.
With the increase in the number of data sources, there is a corresponding increase in the complexity involved in aggregating them and making them analytically ready. Brands therefore end up letting data reside in siloes.
In fact, McKinsey reported about the impact of data siloes in 2012. But even today, data residing in siloes continues to be a huge challenge for marketers.
Many marketers consider the unification of customer data sources as a key challenge, which they didn’t acknowledge even a couple of years ago.
Source: Salesforce State of Marketing Research 6th Edition
When your data resides in siloes, you will face the following challenges.
- The model building time grows exponentially and impacts your time to market. According to Forbes, 80% of the work involved in model building goes into the acquisition and preparation of data.
- Pressed for time, you may end up compromising on the number of sources that you consider for the building of the model. This prevents you from getting insights that you may get otherwise and impacts model-efficiency.
- Data residing in siloes also causes duplication resulting in the need for an increase in storage space that can impact cost, especially when scale happens. Secondly, this also hurts the integrity of the data thereby negatively affecting the whole purpose of model-building.
Data in siloes creates a big obstacle for organizations who want to increase the value of their data assets and exploit emerging analytics technologies such as machine learning.
One of our clients, an aggressive Indian Private Bank just embarked on an aggressive analytical journey with an aim to get into the creamy league of Indian Banks. While they have procured all the core and surround systems on analytics and campaign management, they lacked a single version of truth for all these systems to consume. As part of the larger solution, we built an Operational Data Source by integrating data from 30 different systems that included the core banking system, cards system, CRM, CMS, channel sources, actuary system, mutual fund system, loan origination system alert services, etc. With all the data being available in a single location (ODS), our client was able to make informed data-driven decisions with respect to sales and marketing.
Data being in siloes is a chronic challenge that hampers both marketers and data scientists alike. Earlier, brands would invest in a data lake or warehouse to aggregate all the data, process it and keep analytically-ready data in data marts.
Today, with the arrival of Customer Data Platforms (CDP), marketers don’t need to invest on a data warehouse or data lake and forever be dependent on IT teams.
According to Salesforce 2020 State of Marketing Report, among marketers who say they use CDPs, 86% are increasing or maintaining their use of them. This indicates that CDP investments indeed deliver value and this is our recommendation.
Decide whether you need a CDP just to integrate data and create a unified view of customers or you if need capabilities to act on it, through the CDP itself. Some call the latter CDXP (Customer Data Experience Platform).
#2. You do not sufficiently leverage behavioural data for analytics
Demographic data is a great starting point for brands that are just getting started on analytics, especially when it comes to segmentation.
Demographic segmentation allows you to quickly slot one batch of customers into a pre-defined segment based on demographic attributes and replicate a marketing action with the expectation of getting a similar response.
But you cannot rely only on it.
On the other hand, you have behavioural data of your audience - data that is based on actions that are taken by your audience.
Unlike demographic data, behavioural data / variables can be in large numbers and can be different from one industry to another. The beauty of behavioural data is the relevance that it provides.
It’s ok to use demographic variables to segment audiences if there is no other data with you. But if you have their behavioural data, you should definitely use that, and then reach out to each segment with appropriate communication tailored particularly for them.
Why do you need to use behavioural data?
Because, segmentation done using behavioural data is of greater relevance and campaigns specifically tailored for such segments, comfortably outperform demographic segmentation-led ones.
Here is a comparison of the response rates of two campaigns that we ran for one customer, targeting segments based both on demographic and behavioural variables.
Google itself offers and prefers audience targeting based on behaviours over demographic factors.
Here’s how behaviour-based audience targeting compares vis-à-vis demographic targeting in Google ads, according to Google itself.
Despite the efficiency of behavioural variables being apparent, many marketers still use demographic segmentation in their predictive models.
Such a low perception of the impact of customer segmentation is because these companies still use traditional methods to group customers using demographic variables.
One of Europe’s largest loyalty management groups was on a mission to provide hyper-personalized communication to its subscribers and wanted to have 4 unique next best actions for each customer. Earlier only demographic variables were used to segment customers. But with the new requirement, we had to switch to behavioural segmentation. Thankfully there were 21 billion rows of data, spanning over two years, detailing the purchasing habits of shoppers. We used this to create over 4000 micro-segments based on transaction behavior. We built propensity models using the data and created 16,000 unique treatments for emailer campaigns by employing multi-variate testing and achieved the business goal in style.
There are tons of behavioural data that you already have access to. And this is what you need to do to take your analytics to the next level.
- List down behavioural variables, both direct and derived that might potentially be valid to create segments. It’s absolutely ok to start with heuristics
- Define hypotheses. For example, users that bought a mobile phone through a mobile app 30 days ago using “Buy Now Pay Later” mode are very likely to buy a laptop after repayment.
- Segment your audience and test your hypothesis on a sample set.
- Once successfully validated i.e.response on test group is statistically significant than control group, roll-out hypothesis on the entire segment
- Keep developing new hypotheses as well as adding more parameters on tested hypotheses to move from segmentation to micro-segmentation to maximize relevance.
#3. Your models’ efficiency is dwindling
The good news is that you are using predictive models.
As an organization, the level of monitoring that you provide on model accuracy keeps dropping over time.
Similarly, the performance of models in terms of accuracy and precision will also drop.
Unless it reaches a critical point, you are unfazed mostly. But when it does, the panic starts and you end up setting up a war room and deploy resources to fix things.
This process goes on as a vicious cycle.
Isn’t it apparent that when the model accuracy drops for a considerable amount of time, say for 3 months, it’s time to change the model?
Why would you want to allow it to hit a critical low in order for you to act?
Regular monitoring and updating of models is essential to ensuring that the intention behind the models are indeed met, especially when you built your models years ago.
Regardless of when you built your models, here are some hygiene items that you need to implement to preserve model accuracy.
- Ensure the inputs to the model is correct and fresh
- Ensure the outputs go to the right places, in the correct formats
- Ensure the code is organized
- Ensure correct documentation
Apart from the above, the single most important and strategic reason why your models lose efficiency is that you don’t take into account changing business environments.
For example, if you have a cross-sell model built 5 years ago, it is not right to assume that the hypothesis on which the model was built would be valid even today.
At your business level, you might have opened up a new channel for your customers or upgraded your MarTech and AdTech stacks.
Such things would have increased your ability to capture data in manifold ways. Unless you include those new variables in the model, the output is going to be sub-optimal.
Similarly, take for instance Covid or Demonetization. Those are large political / global changes that would have changed customer behaviour completely.
Source: McKinsey & Company
Unless you change the hypothesis itself in response to these factors, those models will be of no use, thanks to the unexpected changes in customer behaviour
Brands that are using advanced analytics across industries are already feeling the heat.
Source: McKinsey & Company
We know it is very difficult to get budget approval for any ongoing model response optimization as it requires extensive investment on hardware and resources. But, we certainly believe it cannot be the reason for denying marketers those benefits.
We have built and automated this process exclusively in our Acquigo Algorithmic Cloud. We statistically arrive at a response benchmark and determine featuresets for each campaign. We have automated the data preparation, model building and model testing. Only if the evaluation benchmarks are met, will the campaign be executed. Else, the feature set would be modified until the performance benchmark is met.
To ensure your models’ efficiency are sustained over time, we recommend you use the champion-challenger approach and quality and most recent data. Given below are our recommendations to implement the champion-challenger approach.
- Every time you add new variables, check if it will have an impact on model goals. If so, include those variables and refresh the entire model workflow with updated data
- At every quarter or at least once in every six months, formulate new hypotheses to challenge the current model and build models in line with it
- Test the new model (challenger) against the old model (Champion) and compare performance
- Deploy the better performer as new champion
- Rinse and repeat
Note: While it may consume additional cost and effort to keep challenging models on a regular basis, we strongly believe the benefits would certainly outweigh the cost and effort.
#4. You are unsure of where data analytics can help in your marketing
Are you a marketer who has heard of others successfully leveraging different types of data analytics techniques for their marketing, but don’t know where to start?
You are very likely to be one, if you are largely into digital marketing and sophisticatedly use Facebook, Google etc. to drive your acquisition needs.
While Google and Facebook themselves exhaustively use predictive analytics, machine learning and artificial intelligence the applications of data analytics in marketing function is still extensive.
In fact, you can use analytics across the entire customer lifecycle.
- With data analytics, you can gain understanding of existing customers. Use behavioural segmentation to create various clusters of your customers with similar attributes. You can then use these segments to reach the right prospects with look-alike acquisition models
- With data analytics, you can move away from spray and pray marketing. Use marketing mix models to optimize channels and offers, and use multi-variate testing to test communication and maximize response to your acquisition efforts.
- With data analytics, you can understand behavior of your customers. Build propensity models to project their likelihood to buy more products, upgrade to higher variant of products, renew a subscription etc. and plan your marketing efforts to increase their share of wallet, product holding ratio etc.
- With data analytics, you can uncover patterns of usage behavior. Use product and recommendation analysis to understand context behind usage / non-usage and drive adoption and usage of products and features with appropriate marketing communication.
- With data analytics, you can get smarter with customer interactions. Use engagement analysis to learn about the drivers of engagement and sentiment analysis to gauge their satisfaction levels and plan your communication proactively.
- With data analytics, you can drive retention and deepen loyalty with your customers. Build churn models to proactively identify customers with higher likelihood of churn and plan for retention campaigns with personalized offers. You may also build lifetime value models and loyalty models to identify customers with huge potential to be loyal ones and also deliver maximum lifetime value.
One of the fast growing Indian telecom operators wanted to leverage predictive analytics, but was unsure of where to begin. When we interacted with them, we realized that customer churn was the biggest pain point to our client and we decided to use analytics to predict customer churn and use it as a base to get senior management buy in for investing on analytics.
We took the data of churned subscribers from one circle comprising of over 100 different transaction variables. Binning, normalizing and categorizing techniques were employed and extrapolations and predictions were done. We used CHAID algorithm to understand the churn behaviour i.e. Longevity of more than 10 months, drop in ‘validity behaviour’ and ‘talk time behaviour’ and train the model. The model was scored on a new base and prediction was found to be 87% accurate. Thanks to the demonstration of value, the client’s senior management gave their buy-in for rolling out analytics across all circles.
Embarking on an analytical journey is not as hard as you may think. Here are some recommendations to get you started.
- Many analytical journeys fail because, brands aspire to build a 360-degree view of their customers and miss out on low hanging fruit. Best is to start with the data that is already available with you by making it analytically ready.
- Identify which of the above applications would be of relevance to your business and critical from your customer point of view. Start building capabilities to achieve those. This helps in avoiding less value / ad hoc analytical activities.
- Finalize a proof of concept to demonstrate the value of analytics with finite and measurable success metrics.
- Once you successfully complete the PoC, build an analytical project plan with prioritized analytical techniques and ensure the same is adhered to.
- Before getting into implementing the plan, integrate siloed data into a common location and use that as a single version of truth for analytics.
Data Analytics has been in existence for several decades. However, the adoption of data analytics by marketers is still not so high. There is a proliferation of martech and adtech platforms on one side where data analytics of highest quality are getting utilized and there are brands that still have data in siloes, leave aside their quality of predictive analytics.
If you take a deeper look at the gulf above, brands can actually end up leveraging the benefits of data analytics on their acquisition efforts by just subscribing to relevant adtech and martech platforms. It’s only with their first party data where their data analytics maturity is exposed.
Your audience is starting to expect personalized communication as a part of your service and its high time for marketers to take their data analytics to the next level.