Here are six steps to an effective revenue-generating strategy using predictive analytics.
Use It or Lose It
Unless organizations use information to drive action, Big Data—or any data—is more of a liability than an asset. The bottom line: Just get started. Here’s how:
- Define one outcome you want to achieve.
- Start with the data you have and can get with minimal work.
- Add third-party data to create models to identify patterns associated with your customers.
- Align your communications to the needs and behaviors of these “customer clusters.”
- Do A/B testing.
- Consider partnering with an experienced data insights provider.
Although the term Big Data was deemed 2013’s most overused corporate buzzword, it doesn’t seem to be losing momentum anytime soon. There is, however, one Big Data fallacy. Unless organizations use information to drive action, Big Data—or any data—is more of a liability than an asset. The Big Data market for predictive analytics will have officially crossed over from hype to a $3 billion dollar industry by 2017, up 50% from today, according to Survey Analytics. So why does it seem that more and more business executives are becoming increasingly frustrated by not seeing results from their data?
First, there seems to be a gap between the expectations and the actual use of analytics to achieve results from the insights. The disconnect may largely be due to the fact that business decision makers are unsure how to move from operational analytics (insight on financial and performance management) to operational predictive analytics (using models to affect real-time workflow behaviors or processes).
Translating the concept of using predictive analytics for dramatic revenue lift hasn’t yet come to fruition within most companies. While many corporations believe in the benefits of predictive analytics and are investing heavily in building out the capability, seeing the desired results is unlikely without the right process and application in place.
Here are six critical steps to consider before deploying an operational predictive model:
1. Define and understand your objective.
It is said that a problem well stated is a problem half solved. To simplify your strategy, it’s best to define one successful outcome that you want to achieve. This can vary from selling more products per customer, increasing sales of a specific product, increasing customer satisfaction scores or even increasing the number of customers that use electronic versus paper notifications. By clearly understanding and communicating your objective, your team is better positioned to develop a relevant model by focusing energy on what will achieve the greatest lift as it relates to your goal.
2. Collect available data.
One of the biggest misconceptions about applying analytics to business practices is that you need to have a lot of data to get started. This is simply not true. Start with the data that is currently available with minimal work and create a data set of information that shows the outcome, along with any other characteristics that are attributable to the customer database. Based on this past data, (historical results) models can be built to visualize the common attributes of the customers who either failed or succeeded in achieving the desired outcome. Be cautious not to spend more effort on data acquisition and management of the data when the focus is on using the data you have to extract insights that will predict various trigger points to create business value. Many times, executives attempt to bring in data from systems beyond their core platform or believe they need to have a robust CRM system to predict desired customer behavior. In many cases, additional data can indeed assist with model precision. Additional data feeds can also support the response measurement capabilities of a technology platform, but neither is required to begin a successful journey.
3. Design the predictive model.
Once you have insight and outcomes on your historical data, the next step is to append third-party data to your existing data. From the larger data set, the models can identify more relevant patterns associated within your customer base.
These patterns are then segmented into persona clusters based on similar attributes and outcomes. A persona cluster is a grouping of your customers or prospects that have similar characteristics such as life stage, location and demographics. It’s more likely that people with similar attributes and life stages will respond in a similar way. This predictive model is the framework for your strategy and will change the game by offering insight into what has happened and what will happen if communication happens in a specific way to a specific audience.
Once the models have identified patterns associated with the customer information, it becomes easier to target the specific message or offer to the right audience. For example, inactive customers can frequently be segmented into clusters that respond positively to certain product messages. The models compare low activity customers in one specific cluster to current higher activity customers in others (five products per customer). This provides insight into the right message to use to increase the number of customers who are active with more products in the identified cluster. Overtime, models can track the right messages and right cadence of message to trigger the optimal response rate of the target segment.
4. Personalize the message.
The same message and channel do not work for everyone, which is why your strategies need to speak to the individual, not the masses. Align your communications to the needs and behaviors of your customer clusters as determined by your predictive model. Plan persona-specific campaigns based on the habits, preferences and life stages of the individual to increase the probability that the target audience will be receptive. Personalizing persona-specific campaigns can be as simple as changing an image or as complex as a combination of image, content, product and promotion for each persona.
5. Measure and optimize.
After the models and strategies are complete, define the testing process, otherwise known as A/B testing. Control groups are usually 20% of the 100% that is identified. Keep testing different combinations to achieve the most successful message, channel and cadence of touches for each customer persona. These results should be continuously fed back into your predictive model to recalibrate, making the system smarter and more accurate over time.
6. Ask for help.
The collection, management and analysis of mass amounts of data can be very complicated and time consuming. A recent study by Accenture found that a higher than expected percentage of companies are outsourcing their end-to-end processes related to analytics, with 43% of companies surveyed outsourcing their analytics and decision support systems.
Don’t be discouraged by system or software expense. The myth is that in order to collect and apply data intelligently you need expensive hardware, data storage, data scientists and intelligent software. Operationalizing analytics can be done cost effectively with a high return if you apply the right practices and the right people. With the advancement of technology and cloud services, coupled with open source software, much of the expense of predictive analytics is dramatically lower than it was just a few years ago.
Consider partnering with an experienced data insights provider. Besides reducing costs and establishing a centralized base to govern analytics, knowing how to develop personas and setup correct data structures is a science that is heavily dependent on statistics, A/B testing and machine learning. Knowing the cadence of when to execute the next message through the appropriate channel and to even recognize when a person has reached a new life stage based on historical and third-party data as well as statistical assumptions, can all be self-optimized through highly intelligent technology. This technology, managed by data scientists, continuously optimizes the right message, imagery, call to action, etc., over time and will automatically update based on multiple layers of data.
With increasing pressure on insurance and financial institutions to close more sales at a lower cost per acquisition, the Dialog Direct analytics group built an intelligent routing system to route calls to customer service advocates based on historical performance with states and personality types. By appending historical CRM data, core transactional data and third-party data, proprietary scores were developed. These scores were then used in the call routing strategy to pair a customer with a call center agent that has the best ability to convey the benefits of the ideal product to the customer. The results to date show that the routing technique used in the pilot group had a 34% lift in revenue, while holding the cost per sale constant.