Scenario - bxp and best next action
From All n One's bxp software Wixi
Contents
1 Overview
Next-best-action marketing (also known as best next action or next best activity), as a special case of next-best-action decision-making, is a customer-centric marketing paradigm that considers the different actions that can be taken for a specific customer and decides on the ‘best’ one. The Next Best Action (an offer, proposition, service, etc.) is determined by the customer’s interests and needs on the one hand, and the marketing organization’s business objectives, policies on the other. This is in sharp contrast to traditional marketing approaches that first create a proposition for a product or service and then attempt to find interested and eligible prospects for that proposition. This practice, direct marketing, typically automated in the form of a campaign management tool, is often product-centric, and usually always marketing-centric. https://en.wikipedia.org/wiki/Next-best-action_marketing
BNA (Best Next Action) is an area of marketing that is making the most of Big Data. Large volumes of information facilitate the creation of models for what should happen next. bxp is uniquely positioned to help in this area to address the obvious and not so obvious challenges of this area.
2 Examples of clients using / have used this solution
bxp has performed elements of this activity in part and in whole already for a number of clients.
|
|
|
|
|
3 Background
There is mountains of material on this area.
- Google Search
- IBM - 5 Things to know
- XMPRO
- Next best action with decision management
- Next best action marketing
So where to start?
Imagine a customer rings in to a sales person. What do they know about the customer? Nothing. A dialogue occurs and the sales person attempts to work out what is the best solution for them using a number of techniques. Good sales people usually have a patter which they can use for scenarios but these are built up over time.
Image a customer rings in to a customer care agent. What do they know about the customer? They should know an awful lot, but trying to collate all that information between the customer phoning in and the person answering the phone is where technology comes in.
Essentially BNA is using statistics / models to guess what is most appropriate for that customer to provide an upsell or cross sell opportunity. The more information you have the more tailored a response you can provide.
4 The challenges
The dark arts of marketing and predicting what should and should not be done is not easy. It is possible to make informed decisions with large data sets but the word to note is "informed". Every case is unique and all of this work should be considered "is more what you call guidelines, than actual rules !"
4.1 Challenge 1 - The Data
Access to data is the first challenge. If you want to build models you need good, up to date data. Garbage in, garbage out. Collating all of that data is going to be the first challenge.
bxp can massively help by having flexible and easily configured data structures for storing all manner of data.
Secondly, as there is usually a LOT of data to consider, speed is going to be important. bxp uses Oracle MySQL MyISAM as a data storage structure. This is a faster engine than most products on the market specifically for large read processes.
So at this point, we have a lot of data and we can access it quickly.
4.2 Challenge 2 - Creating models / segments
The next phase is to create models within the data. So what does this mean? Well, image a very simple scenario. We have a date of birth field, which allows us to work out an age. We have an address field, which allows us to group localities. We have a firstname, which (in general) allows us to work out gender. If they are an existing customer we know which products they have bought.
So
- Age
- Area
- Gender
- Product
From this simple set of data, we can use a number of different approaches.
4.2.1 The manual approach
This approach uses knowledge from sales people and experienced marketers who create "segments". Groupings of customers who are to be targets. There is a pro and a con to this approach
Pro
- Quick and easy to generate
- Specifically focused
- Can be used to identify groups where we have no customer information yet
- Can map to existing business processes far more easily
Con
- No supporting evidence
Using this I could decide that I'm targeting teenagers (13 to 18). I will look in opulent areas, as they will have money to spend. Male only. We will sell them our "Hip and Trendy" package.
Using the same approach I could decide I'm targeting 30s (30 to 39). I will look in poorer areas, as they are hunting for bargains. Male and Female. We will sell them our "Super Saver" package.
4.2.2 The Data Mining approach
What is data mining?
- “…the process of discovering meaningful new correlations, patterns, and trends by sifting through large amounts of data…” (Gartner Group)
- “…the analysis of observational data sets to find unsuspected relationships and to summarize data in novel ways…” (Hand et al.)
- “…is an interdisciplinary field bringing together techniques from machine learning, pattern recognition, statistics, databases, and visualization…” (Cabana et al.)
Data mining has been fuelled by several factors:
- Explosive growth in data collection
- The storage of enterprise-wide data in data warehouses
- Increased availability of Web click-stream data
- The tremendous growth in computing power and storage capacity
- Development of off-the-shelf commercial data mining software products
It is possible to elicit Knowledge in Data.
“A particular customer transaction at the Ballymount branch on Friday the 14 August at 9:15am, listed a box of tea, a litre of milk with a total purchase price of €3.10 which was paid in cash”
- Such a specific collection of facts is known as a case
- Facts are essential raw material for the elicitation of knowledge from data
- Knowledge can be identified as patterns or regularities in the data
- Patterns (regularities) in data that can be expressed by a statement of the form (IF x THEN y) called a Rule e.g.
- IF a customer purchases a box of tea THEN they will also purchase a litre of milk
This approach uses existing data and mathematical models to statistically generate models.
For each rule generated there is Support (how frequently these facts pops up in the data) and Confidence (for all the times they show up, how often are they seen together)
There are a number of data mining algorithms available, each with different focus for model generation.
- Prediction
- Classification
- Regression
- Description
- Association rules discovery
- Clustering
At the end of this process I will have rules. These rules are my segments
As with any method
Pro
- System driven so doesn't require creativity
- Based on facts
- The larger the data, the more accurate the rules
Con
- Can only base rules on facts in existence, so "creative" thinking not possible
4.2.3 The artificial intelligence route
With modern computer systems capable of performing thousands of rules in real time, just think computer games, it is possible to build models that can adapt in real time. They can be exceptionally expensive to build and are often very resource hungry. Also letting a computer make a decision on your behalf without some level of control can be a dangerous business. it is important to note that it is still possible.
The way it adapts is to use the above data mining techniques in real time to see if the rules have changed. Using statistical tolerances it can change rule in real time if a better statistical result can be found.
Pro
- Real time adapting to situations
- Based on facts and not humans
Con
- Very resource hungry (CPU, disk speed, RAM available, network throughput)
- Very complicated to build (as the scenarios and tolerances need to be established)
- Requires trust and hand holding till it has proven itself as a model.
4.2.4 The usual outcome
The usual outcome is a combination of both. Start with a creative / manual process. Collect data. Refine the model.
It is also possible to use surveys and market research to add more facts.
4.3 Challenge 3 - Measuring results
Ok, so we have some segments we're going to target. It always costs money to do marketing, so the most important thing to do is to create a Return on Investment model.
- What is our base line sales figure? (i.e. what do we make without any extra effort)
- What do we spend targeting this segment?
- When do we put the plan into action?
- What are the results after the plan was executed? (The difference between now and the previous base line, is the ROI)
However being able to accurately associate marketing spend with outcomes requires a lot of systems, steps, checks and integration. bxp can deliver start to finish results.
5 The Elements
5.1 Overview
| Marketing drives communication into a contact point, say for example a customer care department.
|
File:bxp datamining 001.png
Media sources image from How to launch |
5.2 BNA Engine
The engine has a number of choices for best form of operation, but essentially we have all the key elements. We just need to decide what is going to work best. Combinations of the following can also be used.
5.2.1 Scenario 1: Homework already done
Out of hours, a number of processes run. These processes are the application of the segmentation and decisions based on the classifications above. Essentially we predetermine based on the data we have what the best scenario to present it.
bxp implements these rules through the MetaData module which can be schedule to run out of hours. When the contact arrives during the day it is a simple lookup exercise to see what BNA has been applied to the rules for this record.
Pros
- Out of hours allows really complicated rules to be built.
- Reduces operational overhead of real time calculation
- Useful where systems do not have an API. Batch processes are used.
Cons
- Not easily adapted with live information captured from the customer
- Requires the data to be accurate or the wrong rules will be applied
5.2.2 Scenario 2: Doing your homework on the way to class
As the call arrives, you look up the key data and you run that data through a set of fast acting rules. As the record is presented to the agent, the BNA has been calculated in real time.
bxp can pick up data from an IVR and add it to the data. bxp can make API calls as the data arrives to retrieve further information from available sources. Using a JavaScript engine, the rules are applied as the record is displayed giving instant feedback to the agent as to the BNA.
Pros
- Rule sets are easily manipulated and changed in real time
- Can have dynamically captured data added to the records
- Can read from multiple sources in real time.
Cons
- Not easily adapted with live information captured from the customer
- Increases operational overhead of real time calculation
- Requires the data to be accurate or the wrong rules will be applied
- Lookups / copies of data must be available for real time interrogation
5.2.3 Scenario 3: Doing your homework in class
The most dynamic solution of all is that the BNA is calculated as the agent is able to ask the customer questions to compensate for lack of data. The data added is in real time run through the rule sets and a BNA calculated live.
bxp using logic management within its forms to dynamically alter the script. As options are selected they can be easily interpreted by a set of JavaScript rules which provide the BNA when enough data is available.
Pros
- Rule sets are easily manipulated and changed in real time
- Dynamically captured data added to the records
- Can read from multiple sources in real time.
- Live data can make sure BNA is best available with live information
- Adapts for bad / out of date data sets.
Cons
- Not many!
- Increases operational overhead of real time calculation
6 End to End build
Now that we have discussed the various elements required, here is how bxp would suggest a build to provide the solution.
Base line measurement
- Capture base line figures for one weeks activity (Data Profiling or KeyStats)
Existing Marketing first
- For every piece of marketing ensure it has a Media Code.
- Publish the media codes in the marketing
- Load all the media codes into the bxp Media Management module. (Media Manager)
Segmentation
- Using the various data mining models construct the appropriate segments (MetaData)
- Using surveys and market research identify new segments (Survey Management / Form Management)
- Create a rule set to support known facts (MetaData & Form Management JavaScript)
- Create a rule set to model targeted new segments (MetaData)
Homework
- Apply the rules to the existing data. (MetaData scheduled task)
- Identify outcomes that demonstrate Sale, Retention, Upsell and Cross outcomes. (Form Management)
Operation Preparation
- Identify which forms are to make use of BNA
- Implement JavaScript on the forms to mirror the rules (Form Management)
- Ensure that "Media" field is enabled on the form
- Ensure that the "Media Schedule" is enabled on the form
- Force logging of Media Code for every contact
- Ensure sale value is captured as part of the form
Operation
- Capture the data for a given period e.g. one week (Inbound Contact, Outbound Contact, Case Management)
Analysis
- Capture base line figures for one weeks activity after implementation (Data Profiling or KeyStats)
- Calculate the return on investment
- Calculate the return by media on the investment
ReEvaluate
- Reperform the operation from Segmentation until the models reflect targets
7 Conclusion
bxp can provide a flexible end to end solution which will enable BNA and facilitate enhanced marketing information without disturbing existing solutions. It has proven invaluable and operationally practical in the field.
bxp is the ideal choice for all data mining and Best Next Action solutions
