CRM Predictive Analytics Examples to Grow Customer Acquisitions and Lifetime Value
- CRM predictive analytics forecast customer behaviors so companies can plan optimal responses that aid customer objectives, build customer relationships and grow company revenues.
- Forward looking analytics, which use data to forecast the likelihood of certain customer behaviors, is one of four sustainable competitive advantages.
- For these types of powerful analytics, data is the fuel, machine learning is the engine, and actionable insights are the destination.
How to Forecast Customer Behaviors and Grow Company Revenues
For many companies, Customer Relationship Management (CRM) software is a compliance tool where users enter historical information. Because users don't value manual entry or backward-looking information, their participation is often the bare minimum. They put little in because they get little out.
CRM predictive analytics convert static data to forward looking information such as next best offer, next best action or guided selling recommendation. Giving users something they don't know or delivering insights which aid their decision making shifts their behaviors and the CRM value proposition.
This type of decision support examines data to identify customer patterns, trends and correlations which can be used to forecast customer behaviors. This type of CRM business intelligence is no longer reserved for large companies with data scientists and large IT staff.
Pro forma analytics have crossed the adoption chasm and are essential to any company that expects to compete and win in an increasingly competitive marketplace.
Small and midsize businesses may have even more to gain as they are typically more nimble and able to act on predictions more quickly. Ignoring this technology and sitting the sidelines will seed customer share to those competitors who choose to forecast and better accommodate customers.
The Predictive Model Building Process
The process to create forward looking information is built around the particular use cases. Below is a sample predictable analytic, created on three building blocks, to increase customer share and company margins.
- Customer profile segmentation. We start by creating dynamic customer segments organized by a combination of firmographic, behavioral and RFM (Recency Frequency Monetary) purchase history. The purchase history is a strong factor which designates customers into tiers based on their value and identifies customers at risk of decline or churn. With the accumulation of more data the customer segments mature and refine into micro segments.
- Look-alike segmentation. We then apply a clustering algorithm to compare data patterns of the highest value customer segment to other customers who share characteristics but are not yet in the highest value segment. We know from experience these look-alike segments will convert relevant offers at a higher rate than other segments. These offers also tend to be for higher value, higher margin products.
- Recommendation Engine. For each account in the look-alike segment, a product promotion or next best offer is calculated based on a statistical model, regression or algorithm. We like to develop two scenarios and compare in an A/B test. Based on the individual customer profile, the first scenario is often a companion product or bundle recommendation for cross-sell or premium product suggestion for upsell.
Each recommendation places an emphasis on the customer's firmographics, demographics and online behaviors. The second scenario is a peer-based recommendation based on a tightly defined group of cohorts. It's like the Amazon promotion which makes a recommendation based on what other customers like you have purchased. Sometimes, we'll present both options to the customer. In addition to giving the customer a choice, the model learns and improves at an accelerated rate.
This type of predictive analytic is powerful because relevant and personalized cross-sell or upsell promotions to existing customers are two of the fastest and highest margin campaigns available to any company.
For new customer acquisitions we use a forecast model that calculates customer fit, lead score and propensity to purchase. With these insights salespeople can double down on qualified leads and not waste their time on leads they are not going to win.
Customer fit is calculated by comparing firmographics to your ideal customer profile (ICP.) Lead score considers firmographics and demographics but places much more emphasis on buyer behaviors derived from digital footprints. Buyer propensity to make a purchase compares buyer behavior to customer insights.
The lead score is the most complex calculation. We've learned over many years there is no single set of repeated behaviors that consistently apply to an accurate lead score, so we apply regression analysis or machine learning to surface the factors most common among historical successful and unsuccessful sale opportunities. We generally discover a few dozen highly correlated factors with around 6 to 9 of them delivering the bulk of the weight.
So for example, reviewing win-loss analysis history may surface the factors that most influence a sale result. We then validate each coefficient's influence by isolating it and then testing it by removing and manipulating the variable to measure its effect on the overall occurrence. Below is an illustration showing the definition and impact of two coefficients.
- Prospect firmographic data matches ICP (explicit data)
- Buyer behavioral data achieves MQL score (implicit data)
- BANT (Budget, Authority, Need, Timeline)
- Have customer relationship or access to power
- 2 or more information discoveries
- 4 or more engagements exceeding 30 minutes
- 3 or more engagements with decision maker
- Documented win plan with buyer account mapping
In the above table, for the first coefficient, the historical data shows that when a lead is successfully qualified pursuant to all 4 subcomponent criteria, and included with the other highly weighted win factors, the win rate is 68 percent. However, when the lead does not confirm to the 4 criteria the loss rate is 80 percent.
These types of calculations can be extended to forecast the salesperson's likelihood of success based on their actions. For example, when building sales win plans in the professional services industry the data has shown that when the salesperson has nine or more customer engagements (i.e., discoveries, calls, meetings) their sales win rate goes up 2.5X. Tracking this type of activity provides salespeople and their coaches insights that drive actions that produce results.
Forecasts can also be based on inaction.
When a customer's engagement, customer service frequency or purchase patterns decline it's often a sign of a perpetuated situation that precedes customer churn. A relatively simple statistical regression model can examine customers that have your left company in prior years and surface the factors and patterns that predict this event. This type of fall-off model in invaluable in applying customer retention levers and improving customer retention.
CRM Software Accelerators
Forecasts are built with data mining, machine learning and artificial intelligence (AI) tools that are now a part of popular CRM systems such as Salesforce and Microsoft Dynamics 365.
The CRM software publishers deliver simple but extensible forecast capabilities. Some of the most common use cases from Salesforce include algorithms for lead management, sales acceleration, pipeline velocity, client attrition and price optimization.
Use cases from Microsoft for Dynamics 365 Customer Engagement include sales lead and opportunity scores, relationship analytics, insights from LinkedIn InMail, talking points (recommended insights), who knows whom (recommended connections) and Notes analysis.
Microsoft Dynamics also delivers a few role-based use cases. For example, capabilities for sales managers include lead prioritization, pipeline analysis, sales performance scorecards, team performance analytics, sales conversion intelligence and sales data Q&A.
For both Salesforce and Dynamics, we have found regression analysis to be the simplest and most effective tool to create predictive formulas. This technique examines large data sets and identifies independent variables that statistically correlate to a result. We then perform further regression analysis on each variable to measure is correlation with the result.
This regression equation creates coefficients which show the measured impact each variable has on the outcome. You can then assemble and score the coefficients to create a highly accurate forecast model. The good news is that with the CRM software tools, this can now be done by business analysts, system administrators or power users.
CRM analytics create competitive advantage when the CRM app is transformed from a tool of administrative overhead to an insights engine. When the application delivers forward looking information that helps staff do their jobs more effectively and accomplishes their personal and professional goals, it gains user adoption and skyrockets technology ROI.