Predictive Analytics – What you MUST know!

The customer stories published about implementing Predictive Analytics into your own “happily ever after”?  I’m going to share a few things you can do to prepare for this next chapter of your Predictive Analytics narrative.

PEOPLE: Identify your Predictive Analytics cast of characters 

ProducerWho is funding this? How are they wooed?
ProtagonistsWho is leading this charge? Who are the champions?
AntagonistsWho may be opposed to or afraid of this development? How can you get them on your side?
Dynamic CharactersWho will be impacted by this development? How can you gain their support and make them feel comfortable and excited about these changes?
Static CharactersWho are the team members that you’ll need help from, but won’t see a big day-to-day impact from this change? IT? DBA’s? How can you entice them to be on your team?
Fairy GodmotherThe biggest hurdle in “Going Predictive” is what you don’t know; trustworthy guidance from an experienced Predictive expert is crucial to your success.

PROCESS: Create your Storyboard (with the Producer in mind)
Define the story

  1. What line of business is asking for Predictive Analytics capabilities?
  2. What kind of outcomes are they looking to impact? (Customer churn? Supply chain? Marketing? Employee Productivity? Etc.)
  3. What is the cost of ____? (a new customer (customer churn model), diminishing system down time, a fraudulent insurance claim, a patient re-admittance, etc.) This will help with your ROI calculations later.

Interact with your cast

  1. Educate your cast on how adding Predictive Analytics will improve their life and day-to-day activities. Be an evangelist for “Data Backed Decision Making” and spread the culture and excitement!
  2. Brainstorming Session: Use the Lodestar Post-It Note Exercise to gather all ideas and input from your cast. Elicit areas worthy of concern as well. Consolidate and prioritize ideas with the group, and keep a list of flagged issues to address (data access, training, implementation, deployment, budget, etc.).
  3. Keep regular meetings with your cast and build a strong team environment.

Create a plan

  1. Account for lengthy data prep (80% of building a model is obtaining and cleansing data), especially if your analyst is new to Data Mining.
  2. Consider time and cost for coaching, training, and staffing needs.
  3. Draft a diagram or flow chart for how you think your process will work, push it to your cast to refine. Look at where data is coming from and going to – label data, processes, and checkpoints with cast members associated with each. (Bubbles as data, Squares as Processes, and Triangles as Checkpoints/Approvals for example)
  4. Build in REGULAR Testing & Monitoring:
    • BUSINESS Success Metrics (is the model achieving the correct result?)
    • STATISTICAL Success Metrics (is the model accurate? Is it useful/significant?),
    • SUPPLEMENTARY Success Metrics (note other changes based on employing the predictive model i.e. system downtime is waning, employee satisfaction is up, overtime is down, etc.)

Selling the script

  1. Use the Nucleus Research ROI Tool to help build your case and craft your presentation from the CFO’s point of view! How does this make or save the company money? Overlay your diagram or flowchart with areas that will cost something as a starting point, compare this to the “cost to ____” (from “Define the Story” Part C) and determine how many you need in order to see a return on the investment. You can also calculate (or conservatively estimate) how long that would take, often times a matter of weeks or months.
  2. Now it’s time to meet with your Producer (probably the CFO), and gain funding! Bring your Fairy Godmother and Protagonists with you!

TOOLS: Setting the stage with the right props

Data is your biggest prop – It's what creates and fuels your Predictive Model

  1. Quality: Data which are “noisy” (errors and missing data) cannot be compensated for by any Data Mining Technique in the book. A good set of predictors can fail because of an error which masks its effect.
  2. Quantity & Suitability: Be sure you have enough of the right data, it should be large enough to represent the target population and cover all possible outcomes.

Data Mining Tool – You'll need a software solution with these key features

  1. Will work with your data and existing data housing, BI, and other systems.
  2. Is user friendly (unless you have a Statistician/Coder on staff that’s familiar with a more complex tool)
  3. Is easy to bring in-house (to cut down on consulting costs)

CRISP-DM – “Cross Industry Standard Process for Data Mining”

  1. Developed as an internationally recognized definitive process for mining data.
  2. Your Protagonist Analysts should be well versed
  3. Should be a main theme in your storyboard

As a part-time Predictive Fairy Godmother, I think this checklist uncovers a mindset for a successful Predictive Analytics adventure. Feel free to click your heels…err…call or e-mail me to help you get started!

Leave a Comment

Send this to a friend