Skip to content

March 20, 2012


Next Best Offer Design: Solution Architecture

by Ravi Kalakota

Next best offer, next best action, interaction optimization, and experience optimization typically have similar architecture.  Machine learning and multivariate statistical analysis are at the heart of these cutting edge Behavioral Analytics strategies. Typically firms use statistical tools for segmentation models, behavioral propensity modeling, and market basket analysis.

The bleeding edge in next best offer is increasingly around:

  • Applying machine learning to find connections between product tastes and different affinity statements
  • Developing low-latency algorithms that help show the right product at the right time to a customer
  • Developing rich customer affinity profiles through a variety of feedback loops as well as third-party data source (e.g. Facebook user demos and taste graph)
In this blog posting we examine a more traditional next best offer solution architecture.

Targeted Offer Solutions

Good decisions require good information. The ability to target customers depends on segmenting customers by demographics, education, income, lifestyle, and other attributes. It also depends on the ability to assess the mix of products that the customer currently owns and the likelihood that they will find offers of products in adjacent categories to their liking.

An integrated approach to targeted offer management requires a broad business perspective – not just slamming in another software package.  Typically, the targeted offer initiative involves integration with the following infrastructure and tools:

  • A CRM system  or any other system that is considered the “system of record” for customer information
  • Predictive analysis, data mining, and statistical modeling tools.
  • A business rules and decision automation engine. Predictive models are integrated with an business rules engine which drives the workflow.
  • Visualization tool (e.g., Tableau)

Typically for next-best-offer modeling, firms can use SAS tools, like Enterprise Miner (EM). SAS can mine one year of customer historical data on products purchased from 10’s of product lines, as well as prior contact histories. SAS EM can then estimate purchase likelihood using logistic-regression models with 100s of variables. Firms can then layer their customized upsell and cross-sell models in SAS EM. EM scores and iterates these models frequently, recalculating the likelihood of each customer responding to or accepting various offers.

The predictive model calculates the relevant products to offer each customer on their next interaction, regardless of which channel they come through. Typical next-best-offer model that delivers contextual relevance consists of four modules:

  • Customer profile module. This calculates the likelihood of each customer buying one of the products lines a firm has to offer. The solution can test 100s of variables and uses products like SAS EM to perform these calculations. The final output of these modules is a matrix of several million rows (customers) by product lines columns (e.g., say 10 product lines) and includes the probability that each customer will purchase each product.
  • Economic profit relevance module. This module calculates the net present value of each customer purchasing each product line.
  • Eligibility rule module. This module consists of a set of hard rules that identify if a customer is eligible to buy a certain product line. Some examples include: “don’t offer a sweets if the customer always buys low-carb,” “don’t offer life insurance if the customer is over a certain age,” or “don’t offer credit lines if the customer has already defaulted on credit in the past.”
  • Contact history module. This takes each customer’s contact history into account when identifying the next best offer. So if the firm already offered a certain product to a customer recently and they refused it, the firm will exclude that product from the customer’s next best offer for a certain period.

Growing the customer relationship is the perpetual challenge of all companies.  As companies race to grow revenue faster than competition they are pursuing more sophisticated targeted offer strategies customized by channel. One goal of the targeted offer is to boost customer lifetime value (CLV).

In multichannel customer-facing business processes, world-class firms now have to continually optimize all offers and customer interactions through all channels, business processes, and touchpoints.

In the journey to world-class marketing, analytics is increasingly a core ingredient of a finely tuned marketing/sales strategy of retention, upselling, and customer experience optimization.

6 Comments Post a comment
  1. Kit Lloyd
    Mar 20 2012

    Ravi – interesting article, perhaps one area to expand upon is the inclusion of external (and unstructured internal) big data sets. Additional data can assist Bayesian Inference engines to improve prediction and can also assist the Markov Decision Process algorithms used for Next-Best-Offer. In multi-channel environments the inclusion of real-time data (eg a dropped call) can further increase making the right offer at the right time.


    • Ravi Kalakota
      Mar 20 2012

      Kit – you make a good point. While real-time data would be useful in getting more precision and also training the algols, most organizations I am working with are not ready for this. Hence the focus on a more simple architecture.

      Organizations are struggling to execute basic things. In the “crawl, walk, run” continuum, most firms are still in the crawl phase. IMHO when great analytical models collide with organizational fiefdoms, politics and lethargy…the latter wins every time. Hence the KISS approach.



Trackbacks & Pingbacks

  1. Big Data Analytics Use Cases | Business Analytics 3.0
  2. Apple IOS 6 Passbook: Enabling SoLoMoMe + Omni-channel Analytics | Business Analytics 3.0
  3. Big Data Analytics Use Cases | Twiller Moore
  4. MobileFirst Marketing Automation is HOT | Digital Transformation, Engagement and Experience

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

Note: HTML is allowed. Your email address will never be published.

Subscribe to comments

%d bloggers like this: