12 common mistakes with customer analytical models
This is a wonderful set of tools to ensure you are modeling your data properly. At Linx we continually adapt our models as we learn new demographic and behavioral data sets to test the new models against actual performance metrics. Use these all wisely!
Optimize your customer analytics by getting the models right
Customer analytical models can deliver huge value for companies that invest in them to improve their sales and marketing activities. But even well-known big brands can get it wrong when designing, implementing and operating these models. From Barclays to John Lewis, Cineworld to Pizza Express, businesses across all sectors are benefiting from the use of customer analytics. These days, it is unusual to find a company that does not analyze customer data, even in its simplest form. Customer analytics may fall under business intelligence, marketing operations, finance or even customer support, but wherever it lies it will have the potential to improve the optimization of sales and marketing functions.
Companies often want to know which products or services specific customers are most likely to purchase, which customers need a nudge to help them to complete a sale and which customers are most likely to leave them. When used intelligently the results that customer analytical models generate have a direct and quantifiable impact on the revenue and profitability of a company. Given this, one would expect that the development and operation of customer analytics would be second-nature to businesses, and a well-established methodology. At Intilery however we often find that companies have little or no use of analytical models within their sales and marketing functions. In addition, where models have been implemented, common mistakes are evident.
Areas for focus
Typically, there are three areas which need the most attention (not accounting for having no model at all).
The planning, design and definition of the models.
The deployment and operation of the models.
The refinement and lifecycle of the models.
We’ll look at the mistakes I see across these three areas:
1. Setting the wrong customer value
The second most utilized customer analytics model is the one most often designed incorrectly. The most common design flaw in the definition of the retention model is the valuation of the customer (sometimes referred to as CLV – Customer Lifetime Value). Consultants often encounter CLV definitions that do not represent true lifetime value, rather a current or recent customer valuation.
Utilizing the value attributed to a customer, a retention model may assign a deep discount or a margin-heavy offer to a customer who has only recently become a high-value customer and is likely to churn (either churn completely by leaving or churn to a lower value-level).
Conversely the retention model attributed via the customer value may offer too shallow an offer or incentive to customers likely to churn who were previously high-value customers.
“The retention model should look at the value the customer has previously generated for the company and provide appropriate offers/incentives to the customer to either bring them back to their historically high level of value or stop them churning.” The problem is that by looking at
The problem is that by looking at recent activity or overly averaged values, the wrong or inappropriate offers and incentives may be applied to the customer. Instead, the retention model should look at the value the customer has previously generated for the company and provide appropriate offers or incentives to the customer, either to bring them back to their historically high level of value or prevent them from churning.
2. Not utilising customer value propensity
The second issue with valuing a customer for retention is not taking into account the customers’ propensity to increase/decrease in value, by only looking at current value and/or past value. If this is the case, you limit your model to events that have happened and not events that could happen. 3. Ignoring the social value of a customer The third issue with retention modeling is not taking into account the social network value of the customer. If a customer leaves, or does not receive the service or incentives they expect (yes some customers do expect incentives), then the customer may broadcast this on their social network. Taking into account a customer with a very well connected (virtual or physical) network, you may wish to increase the value of a customer and incentivise accordingly for retention.
3. Not retaining customers all of the time
Another issue with retention modeling is that they are often only run at the beginning of the month to identify customers likely to churn within that month (or any other given period).
To be effective the retention model needs to operate in real-time identifying customers and visitors that are likely to churn and applying the required action to prevent it. Churn triggers are used to identify customers that are most likely to churn. These could include; a product being out of stock, a late delivery, a slow loading page, very few (or no) search results or simply gesturing to leave the page.
For a retention model to be effective, the model must include periodic (daily, weekly or monthly) analysis of churn detection along with real-time churn-triggers. Using these together it is possible to apply offers/incentives or other actions to retain customers and more importantly do so in a cost-effective and margin protecting way.
4. Ignoring seasonal variations
Another common error is developing a retention model that doesn’t take into account seasonal variations, or using a single retention model all year. Customers behave differently according to the season and according to seasonal habits. An obvious example is the increase in browse or spend at Christmas, but what about other calendar events? All variations should be accounted for. Also, products and services may have seasonal variations – such as buying cycles or budgets – and companies may have unforeseen variations due to unpredictable trends.
Whatever the seasonal variation, retention models should be careful to incorporate these into the design.
5. Lack of granularity
Segmentation models are the most implemented model across all companies and industries, and yet often the least used. Typically, this model will only place customers into high-level segments and built around value, product, basic behavioral data and sometimes geo-demographic segmentation. The issue with this approach, whilst valuable, is that it does not empower companies to deliver ongoing actions. Instead, it drives businesses to shape themselves around the customer segments and shape their offerings for them. For example; a company discovers it has a noticeable percentage of older customers and therefore develops specific services/products for that customer segment.
The issue with high-level segmentation is that it doesn’t take into account the detailed segments, where segments may contain a small number of customers or even a segment of one customer (known as one-to-one marketing). Detailed segmentation allows a company to take action (often in an automated way) on the various individual behaviors of its customers. The results of of this type of segmentation are much more useful and clear. For example, a detailed segment in the travel industry could show how each specific customer behaves; the number of searches before booking, the prominent day of the week for engagement, the seasonal variation in browsing behavior, the likelihood that a customer “surfs for vouchers” before completing or a change in the type of service/product the customer views/purchases.
Examining customers across detailed segments enables companies to take specific actions to change or influence behaviors, such as, targeting customers with offers when they are most likely to be receptive or delivering an upsell incentive for a customer who has dropped down to a lower than usual price-point. The key learning is that by deploying detailed segmentation it is possible to target specific actions rather than shape your company around a few high-level customer segments.
Channel Migration Models
6. Migrating customers for the wrong Reason
Companies often devise methods to migrate customers from one channel to another, for both marketing/sales communications and for customer services (sales and service). The mistake here is to do this for the wrong reason or at the wrong time. Companies will try and lower the cost of managing a customer by migrating them from a higher-cost channel to a lower-cost channel, the operational cost for managing customers may be reduced, but this could greatly reduce the lifetime value of the customer if the migrated channel is not sufficiently sales focused. Also a company may migrate customers from social to email, or from branch to online-chat. Again the operational cost of communications may be reduced but not taking into account the sales effectiveness of the new channel could result in lost sales.
Activity Optimisation Models
7. Selling instead of helping
Companies can model the lifecycle of a customer at an engagement level to ensure that each customer has their particular needs met, though often a company will only look at the sales category of actions to try and sell the next product or service to them. While this model is useful and can be used to forecast revenue, it fails to capture the bigger picture.
The next action a customer needs is dependent on many individual and personal factors. Companies should design a number of actions that can be applied (with personalization) to satisfy the customer and promote customer loyalty. More worryingly, only looking at the sales channel and bombarding customers with constant messaging about products or services they can buy could actually disengage customers completely. Intilery recently worked with a client that had successfully increased sales in the short-term but had not been able to see the long-term damage it was doing to customer loyalty and retention.
“A well designed model will take into account all possible needs of a customer and communicate solutions to the customer at the right time”.
- Informing a customer about the company’s app and its benefits
- Tell the customer about different ways they can get in touch
- Provide post-purchase advice and information, e.g. destination guides or product info such as warranties
- Provide details of other channel services, e.g. physical store location or opening hours
- Collect further personal details or preferences (but clearly explain why)
- Ask the customer for a referral (perhaps with incentives)
- Show the customer how to share their purchase/booking to help validate purchasing decisions Know when NOT to communicate with a customer (for a period of time)
8. Working in silos
Another common mistake is not gaining adequate internal support or buy-in from across the business. Stakeholders from all areas of the business should be involved with the design and operation of a customer analytics model. We recommend that when planning this type of model that you use a RACI matrix for identifying and involving stakeholders.
Why involve other areas of the business?
- Can the business operationally support outputs and actions of the model?
- Will there be operational costs that need to be budgeted for?
- Will customer support need to implement new policies and practices?
- Does marketing need to work with new key messages?
- Will sales need to adjust revenue targets?
- Do new KPIs need to be setup to ensure the model has buy-in longevity?
9. Treating new customers like everyone else
New customers must be treated very carefully, they top-up the customer base and directly impact churn levels. A new customer is the most receptive to offers and cross-sell, but is also the most likely to churn. Reasons for failure are often categorized as too much or alternatively incorrect engagement. Getting this right is key to a long-lasting and profitable relationship with customers. One approach is to simply not contact new customers for a period of time as the behavioral profiles of new customers does not usually reflect their long-term behavior. While this will improve the churn rate of new customers it will affect the bottom line. Instead, a better approach is to design a welcome program that utilizes next-action analysis, detailed segmentation and of course seasonal variation.
The actions that are taken for new customers should be unique and personalized to every individual customer, this may mean that for certain new customers no contact is made at all, for others a full suite of engagement activity, website personalization, offers and cross-sells are applied.
10. Not refreshing models
As part of ongoing operations, a typical mistake is not refreshing the customer analytical models. Few companies revisit and analyze the ongoing effectiveness of their customer models. Failure to do so depletes the effectiveness and efficiency of the models and can lead to less profitability and increased churn.
As time passes, the ability to gauge the effectiveness of the analytical models increases, therefore companies should regularly assess their models. A variety of assumptions will have been made following the design and deployment of models, when it was simply not possible to analyze or predict the outcomes or understand the customer environment. Taking the time to periodically review your models will allow you to test your assumptions against new real-world data.
Also important is the changing landscape of the business, as company strategies, operations and marketing activities change, the structure, desired outcomes and operations of the analytical models may also require updating. Taking the time to update these to match the direction of the company will ensure strategic alignment.
11. Testing the wrong way
The final common mistake concerns testing. Testing of analytical models should be based on clear and rigorous statistical analysis, but also “business common sense”. While a test result may show a clear and statistically sound increase/decrease in a measurable variable, looking at this from a different perspective may prove the opposite.
This is commonly caused by uncontrollable environmental factors or over reaching on causation/correlation. The most effective way to test customer analytical models is to test them over a number of business periods, whilst applying statistical methods, and finally simply asking the question “does this make sense”.
Source: SmartInsights February 25, 2016