Predictive Analytics for Customer Success

Predictive Analytics is a branch of advanced analytics which is used to make predictions about unknown future events. It uses many techniques such as statistics, data mining, machine learning, modeling, and artificial intelligence to connect the information technology, management, and modeling business process to make predictions about the future.

How Predictive Analytics benefits Customer Success teams

SaaS organizations have no lack of data about their customers. The data is in their billing system, CRM system, support ticketing system, and the usage data that they can collect from their product. The challenge is to make use of this data to find customers who are not recognizing the value they expect from a product, and also those who are ready to convert from a trial or buy more. Intensifying that challenge is the volume and velocity at which the data is created and the fact that it is spread across multiple systems.

However, technological advancements have made it possible to analyze the data efficiently and the use of Predictive Analytics in customer success platforms has been growing rapidly. Businesses are adopting various solutions that include Predictive Analytics to increase customer success rate by predicting customer behavior.

Machine Learning is one of the most important techniques used in Predictive Analytics. It attempts to predict the future by analyzing data from the past. Machine Learning unlocks the value hidden in the data acquired by a CSM (Customer Success Management) solution and configures the factors that drive various outcomes. It analyzes hundreds to thousands of factors and identifies which ones are the best predictors for each type of outcome.

CSM solutions that involve machine learning technology are continuously capturing data, and they update their predictive models based on this new data. This indicates that the models get more accurate over time, with no effort on the part of the SaaS vendor. The active nature of a machine learning model means that it automatically adapts to changes in overall customer behavior.

Customers may react differently over time as product offerings change. Having a more accurate predictor of customer behavior can significantly improve the effectiveness and productivity of the Customer Success team. More accurate predictions mean that there are less ‘false alarms’, which waste time that could rather be spent on customers in need. Not receiving alerts for situations that warrant the customer success teams can result in lost business (for example, no upsell alert) or lost customers (for example, no churn alert).

The key purpose of utilizing Predictive Analytics is that Customer Success Managers can spend more time with customers who need it, and where they can make an impact on growing the business.

How to implement Predictive Analytics for Customer Success

Machine Learning, as discussed above, influences the data being collected by SaaS vendors to predict which customers need the attention of the Customer Success team. It does this by evaluating historical data and learning how customers behaved in the past so as to predict how they might react in the future.

Here’s a step by step process of implementing machine learning models as well as what needs to be considered when analyzing CSM solutions with Predictive Analytics.

Predictive Analytics for Customer Success1. Define the Business Goals

When getting started with Machine Learning, it is important to define the business goals. For most SaaS Customer Success teams, the business goals revolve around predicting one or more of the following:

  • Which customers are likely to expand? (upsell/cross-sell)
  • Which customers are at risk of churning?
  • Which customers are likely to convert from a trial to a paying account?

It is important to ensure that the CSM solution addresses all the goals, including what is needed now and in the future. For instance, it is not recommended to settle for a system that only predicts churn. You may not have an upsell product for now, but you could in the future.

2. Identify the Data Sources

The next priority is to identify the data sources to be used in the machine learning models. This data should not be limited to behavioral data such as support tickets and product usage. Different divisions of customers may behave differently, hence it is important to indulge as much information as possible about the accounts (for example, revenue, number of employees, industry, etc.). Also, if the customers are classified into categories, that data should be included so that the models can assess each category for its own best signals.

Besides standard data feeds to a CSM system, such as support tickets, product usage and billing, it is necessary to consider any customer data that is specific to the business. For instance, if the business is an e-commerce site, the data may consist of the number of customers for a specific product or number of products a customer has bought on the site. In most cases, this type of customer account data can be very informative to a predictive model.

To make predictions on what customers will do in the future, it is vital to have good data on what they did in the past. With historical data, the CSM vendor can reduce the time it takes to develop working models.

Time is not the primary concern for collecting data before the models become useful. It may take weeks until the models have enough data to be useful, depending on what’s happening with the business. However, it matters whether there are sufficient examples of the outcomes that the business is looking for (for example, accounts which upgraded, churned accounts).

Consideration for CSM Solution – It is important to make sure that the CSM vendor is including as much data as possible in their models, and not limiting it to just product usage.

3. Feature Extraction

The next step involves preparing the data for modeling. This process starts with the basic set of raw data and constructs derived values, named ‘features’, that can better inform the model. Take an example of a customer who constantly uses a feature in the product, say, 50 times per week, and one week he uses it 60 times. In simple terms, this is a jump of 10 uses. Another user may have only been using it four times a week, and then one week he uses it eight times. Although this change of four uses is comparatively smaller for the second user, the rate of change is significantly higher (100% increase vs. 20% increase). In some cases, the rate of change can be more helpful as a predictor than the absolute change.

Data Scientists who create models look at many features from the raw data to see if they can achieve better accuracy from the model. By evaluating things like rate of change, ratios or momentum, they can extract more meaningful insights from the data.

Considerations for CSM Solution:

  • The CSM vendor should talk with the organization to better understand its business and seek help in extracting the most informative features.
  • CSM vendors who exclusively analyze ‘Customer Success’ outcomes, such as churn or upsell, across several organizations will have a better understanding of how to enhance a model as compared to general machine learning vendors who apply the technology for a variety of cases.

4. Model Training and Validation

Once the feature set has been designed, the next step is to train and validate the model. This process needs ‘labeled’ data, which means that for each account in the dataset, the outcome to be predicted is known. For example, if a company is trying to predict churn, it has to identify which of the customers in the dataset actually churned and which did not.

The CSM vendor generally divides the data into two sets. One of which is used to train the model using a ‘supervised training’ technique and the other set is used to test the model. This typically allows the vendor to measure the accuracy of the model, and make modifications as needed.

Consideration for CSM Solution:

  • A variety of machine learning algorithms and techniques can be applied to the features set. It is best when the CSM vendor tries several of those to see which one works best for the business. Usually, the vendor should provide a report on the accuracy of different models.
  • If the organization is looking to predict multiple outcomes such as upsell or churn, the CSM vendor has to create separate models for each prediction.
  • The CSM vendor has to create models that are unique to the business and not just a basic model that can be used for all of business’s SaaS accounts.

5. Model Maintenance and Refinement

Continuous model maintenance and refinement should be included in the CSM vendor’s standard offering. The vendor should have full responsibility for designing and maintaining the predictive models. This eliminates the requirement of having data scientists on staff.

Some CSM vendors deliver predictive analytics as a separate product or service. They offer to sell professional services where they’ll analyze the data, and suggest better rules to be used for alerts. The business pays extra for that service, but if the service is used only on a quarterly or semi-annual basis, alerts would not be as accurate as they should be. Machine learning models should be updated on a weekly or bi-weekly basis with the data being collected each day to generate new daily predictions.

Considerations for CSM Solution:

  • It is important to ensure that the vendor’s predictive analysis is part of their standard CSM solution. The business should not have to pay extra for this procedure.
  • The CSM vendor should meet with the company frequently to discuss any changes to the business that could impact the models, as well as provide regular updates on model accuracy.

Conclusion

Predictive Analytics is a dynamic technology that makes learned guesses based on the existing data. The use of predictive analytics is growing rapidly. Customer Success software that incorporate machine learning allows to functionalize the technology by establishing it into the business process.

By allowing the Customer Success team to proactively focus on accounts that are likely to churn, expand or convert, CSM solutions with Predictive Analytics can certainly improve the efficiency and productivity of the CSM team. Predictive Analytics helps Customer Success Managers prioritize their efforts on accounts that need their attention the most, and by doing so they are equipped to boost engagement, customer retention, and lifetime revenue.

 

Ridhima Rao Donthineni

About Ridhima Rao Donthineni