A Simple Customer Relationship Model

Being able to reach customers in an efficient way has become a pre-requisite for any companies nowadays. The amount of communication received by customers increasing every day, and the number of channels a company can use to do so as well, it has become a competitive advantage to find the best time and channel to deliver a message to a customer in order for them to see it and interact with it. In the following lines, I will show how companies can do so by using a simple model embedding the customer behaviors to define the best communication setup.

CRM, a quest for the right channel for each customer

In the past years, both CRM and Customer Experience have evolved a lot, allowing to define more and more granular customer cycles dedicated to each interaction customers have with companies’ product or service. At the same time, the diversification of communication channels is larger than ever and companies cannot afford to be absent of one of them, risking that window of communication to be taken over by competition.

While both developments offer tremendous opportunities to personalize the communication sent to each customer, improving customer experience, they also yielded a high level of complexity. Companies need to choose wisely what topic to communicate to customer, and where, in order not to overwhelm them by too many communications on too many channels. Moreover, the recent changes in legislation indirectly push companies to make their communication as relevant as possible for the customer.

Finding the right channel, for the right topic, for each customer, is not an easy task. More and more tools start to support such type of recommendation, making use of more and more data. However most of the companies still lack the capability to organize efficiently their communication:

• Young companies cannot afford the more complex tools that could automatize this omni-channel communication strategy, but still need to try to achieve such goal to grow and stay on top of the competition
• Larger companies, even if they make use of such complex tools, still need to define boundaries and basic rules in order to guide the tools in their decision.

Indeed, imagine a tool that choosing SMS as a channel to send a promotion based on engagement data. If a customer that receives this promotional SMS makes the desired action, i.e. a purchase, the program will identify the channel SMS as a good channel to send the next message, which could be the next day, or in a couple of seconds in the case of an order confirmation for example. We see that this can be problematic as SMS is an intrusive channel for many users and receiving 2 within a short period of time can damage your brand and de-engage you customers. However, by the time the program realizes it, it is often too late. This is why Managers still need to define rules, boundaries, to restrain the frequency of communication on some channels, and also to decide which message to deliver.

Defining these rules can be tedious and complex without a good model.

In the following I suggest a simple model to help managers define on which channel to communicate which message, whatever tool they use. In a second part, I will expose how the model can be easily complexify to make use of the different data a company can collect on their customer behavior.

A Simple Communication Model

It is important to note that the simpler and most efficient way to determine a communication strategy is to analyze the communication set under the Customer point of view, not the Company point of view.

Leaving aside the notion of timing (when to send a message on a channel), a communication strategy can be summarized as a set of Messages and Channels and finding how to couple them is the purpose of the model.

From the point of view of customer experience, we only need one characteristic to define a message: Relevance. In other words, is the message useful for the customer? Am I communicating to her/him a expected or interesting piece of information? For the rest of this article, we define R_m as the relevance factor for a message “m”.

Similarly, one need only two parameters to define a channel in terms of customer experience: Intrusion and Usage.

Intrusion represents the sensitivity of a user to a specific channel, and his/her tolerance to the number of message he/she is willing to receive on that channel over a given period of time. We thus define I_c as the Intrusion factor of a channel “c”.

Usage represent the number of time a channel can be potentially used within the whole company to target to a user over a specific period of time. We define U_c its usage factor.

Finally, as stated above, a communication strategy aims at coupling messages and channels in the most efficient way. However not all channels can be used for all messages (e.g. one rarely sends a weekly newsletter by SMS). So, we define delta_mc, a factor equal to 1 if a channel c can be used for delivering the message m, and 0 overwise.

Based on this, we can define for each couple message / channel (m,c) a recommendation factor kmc:

The higher k_mc is, the more recommended it is to use the channel c for the message m. Its value is affected as follows:

• The more relevant the communication is to the customer, the higher k is, compensating Intrusion and Usage
• The more intrusive a channel is, the less we can use it
• The more a channel is used, the lower k gets, reflecting spamming behavior

Below we will give a concrete example showing how a Manager can use this simple model to build her/his communication strategy within a couple of hours.

E-commerce example

In the following we assume the point of view of a Manager working in an e-commerce company. Neglecting for simplicity the transactional communication such as “rest Password” or “account change confirmation”, let’s suppose the list of topic the Manager wants to send are:

• Offers
• Cart abandonment
• Order confirmation
• Shipment confirmation
• Shipment notification and tracking info
• NPS survey

Let’s also suppose the Manager has at his disposal the following channels:

• Email
• SMS
• Push notification

The first step is to define the d_mc factors i.e. the channels we allow ourselves to consider using for each topic. Let’s assume our Manager comes back with the following:

Defining which channels to use for each topic — d_mc

Then, one has to define the relevant time period under which we will estimate the Usage factor and Intrusion Factor. Let’s assume here that period to be 2 weeks, which seems a reasonable period of time between an offer received by a customer and the potential reception of an item upon purchase. Assuming all the message above will reach the customer within that period of time, we get

Defining the Usage Rate of each channels — U_c

Here U is just defined as the sum of dmc for each channel, as we assumed only one instance of each message would be sent to the customer over that period of time. If we had assumed that 2 offers would reach the customer upon the defined period of time, U would be 7 for Email and 4 for Push notification.

Now our Manager needs to define the intrusion factor. One can use gut feelings to give a number from 1 to 10. Here we will use the number of sendouts one think is acceptable for a customer. Believing that he/she can only contact the customer 5 times over the period, and defining that it is ok for a customer to receive 5 emails in that period, but only 3 push notifications and only 1 SMS, one would find the following I factors

Defining the Intrusion factor — I_c

For example I is 5/3 = 1.6 for a Push Notification.

Finally, we are left with defining the Relevance factor. Once again, one can use gut feeling numbers ranging from 1 to 10. We in turn will rank the messages by order of importance, 1 being less important, and consider that number to be the number of other messaging I am willing not to send to make sure a customer receives this one. In practice

Defining the Relevance factor — R_m

Here we say that, for that Manager, NPS is the most important message. In order to get customer feedback to be able to improve Customer satisfaction, the manager is willing to not send the 5 other messages if it means this one message is reaching the customer. Then comes order confirmation, which is important to avoid customers reaching Customer Care enquiring about their order. Shipment Notification comes after as the customer wants to know when his/her package arrives. After that comes only Commercial email. Note that if one takes the company point of view these messages might rank higher, but remember that the customer point of view prevails. Finally, shipment confirmation email is the less relevant one as this information might be found in the order confirmation and the account area.

Using the formula defining k, we obtain in the end

Resulting k_mc factors

From that table, one can then proceed to assign messages to channel. Order and shipment confirmation are automatically assigned to Email as there is no other option. Then, SMS being allowed only on 2 messages, and since it can be used only once over the period considered, we start with defining which message to assign to it. The obvious choice is to use SMS for Shipment notification as NPS/email combination has the highest k factor.

Using k_mc to assign channels and messages together

Shipment notification is recommended to be assigned to Push Notification which also makes sense as one wants the customer to open the tracking link on his/her mobile phone.

Note that sending NPS via SMS provides a high Answer Rate usually and the Manger could make that choice. Additionally, one could have used the Push notification for Shipment notification and choose in the end not to use SMS at all. This shows why automation of such strategy is not as trivial as it seems.

Then we are left with the commercial messages to send either on Push Notification or email. Score Clearly indicates that Push Notification should be preferred. This is also coherent as these messages require high reactivity of the customer.

The picture above follows blindly the highest score logic, and we see SMS has been factored out. However, a Manager might decide to actually use SMS for Shipment Notification and reduce Push Notification pressure as in the solution above we maxed out the number of Push Notification allowed (3). Likewise, one could use Email for Offers instead of Push notifications, or, use SMS for NPS as we know answer rate is higher on that channel.

As one notice, the recommendation factors are guiding the Manager decision, but he/she still has freedom to tune the assignments based on some external factors not embodied in the parameters we use.

Automation of the model

Has seen in the example above, while in its simplest form the model relies on 3 factors which value are assigned by gut feeling, it can be complexified to integrate data the Managers have at hand, making it more and more accurate/reliable.

In the following, we detail how each factor could use today’s data and tools to be automatically computed, on a real-time basis, hence making it a powerful recommendation engine for all CRM Managers.

d factor

The factors identifying if a channel can potentially be used for a certain topic can be personalized for each person thanks to communication preference settings and optin data a company can collect from its customers. Imagining a customer said they do not want to be surveyed, the d factors for all channels regarding surveys will be 0. Equally, a customer opting out of Push Notification will turn all d factors related to push notifications, irrespective of the topic, to 0.

Usage factor

Usage factor automation is a direct expansion of the automation of d factors. One can use the optin and communication preference data to define U. A company can also ask its customers how many times a week, or a month someone wants to be contacted on a channel and use this data to compute U factors.

Relevance Factor

The relevance of a communication for a customer can be identified within the engagement data for each customer. The more important a communication is to a customer, the more he/she will engage with the communication material. Measuring this engagement, and defining threshold of engagement (manually, or thanks to Machine learning algorithm) is tantamount as defining R factors. The relevant data is:

• Open Rate / Impression Rate — The more a customer opens an email or sees a push notification, the more it means he/she is interested by the topic communicated. She/he wants to know more about it.
• Click-Through Rate — The more a customer interacts with your communication, the more he/she is attracted by what you are telling him/her. Whether it is a redirection to your blog, a survey link, or an offer, if a customer clicks, she/he is interested and wants to perform the action you highlighted to them
• Optin/Optout — If a customer optouts after receiving a specific communication, there is a high chance that the topic you communicated to them is not interesting to them. Vice-versa, if a customer optin to receive certain topics form you, there engagement at that time is probably very high and the R factor should be too
• Time spent on a communication — How long a customer spent reading an email, how long have he/she spent on the landing page lying behind the link of your SMS? The linger the better engaged a customer is
• Other KPIs — For each communication sent to the customer, except purely informative ones, the company expect a certain action to be performed as a consequence (buy an item, complete a survey, read a blog post etc). The Rate of completion of these actions add another level to the engagement of a customer (Indeed, someone can click on an offer but in the end not finalize the purchase)

Combining all these elements via an algorithm, and evaluating them in real-time, using data from all channels and platforms, will allow to define very accurate R factors.

Intrusion factors

The Intrusion factors are also relying on engagement data. In the example we have shown how one can define them defining “acceptable frequency”. But this can be automatized using:

• Optin / optout data
• Open Rates / Impression Rates
• Communication preferences settings

Conclusion

We have established a simple model allowing Managers to allocate their communication to the right channel based on few engagement considerations, ensuring efficiency and relevance for the customer.

The model has the interesting property of being open to complexification using part or all the data a company obtain from their communication tool and/or internal systems.

How to combine all this data together into a model that is relevant for your company is out of the scope of this article, but we want to highlight that the model will need input from the Managers to help the algorithm decide what is best for the customer. For example average open rate of emails differ from industry to industry and from topic to topic. Defining which threshold corresponds to engagement or disengagement need to be appreciated by the communication Manager in order to initialize the model. Equally, one can ask: How much a relevance or intrusion factor should be degraded after someone does not open an email? Do we actually degrade the factor after the first non-opening event? The second?

All these questions could be answered by the algorithm automatically but one would probably need to implement toggles to allow managers to adapt the model’s sensitivity to their industry and specificities.

In the quest of building a model to improve personalization of the communication going to the customers, one should not forget to also allow the model itself to be customized.