Top Tips for Outsourcing Modeling

Share this article:
A client recently mentioned that she had squeezed as much as she could from her current way of segmenting, using previous recency, frequency and dollar spending. "I want to graduate to the next step of modeling," she said.


She had been to several conferences and had been in contact with colleagues who were using modeling, but she still did not know where to start. She decided that she wanted to engage the services of an external consultant, but beyond that, she didn't know what to look for.


Many marketers face this problem.


When using a consultant to build your model, pay close attention to: the model proposal, interaction with the consultant during development, the presentation and the aftermarket.


The model proposal. A handy rule is, do not substitute length for strength. After the initial meeting, the consultant will prepare a proposal based on his or her view of your needs. This document should roughly cover:


* Background. This section will review your situation and articulate why the consultant was originally asked to submit a proposal.


* Business objective. What is your goal? Do you want to maximize response or perhaps optimize revenue?


* Project objective. What will the consultant do to help you reach your goals? Frequently, development of a statistical model is required.


* Approach. How will the project be handled from an analytic point of view? Is the explanation clear?


* Method. Will a logistic regression be employed, a neural network paradigm or is a tree analysis sufficient? Perhaps a combination of methods will be needed.


* Data. What categories of data will be explored? Will internal data be used or will external data have to be overlaid?


* Timing. How long will it take to complete the project?


* Costs. What will the fee include? The model code only? Will there be a presentation of the finished model, and will there be assistance after the model is completed?


Interaction. Keep in touch with the consultant. Be aware that some consultants pass the work on to someone else, so demand to speak to the person crunching the numbers. This will allow you to probe further than if you had only spoken to a middle man.


The presentation. This serves as documentation for the project. The original proposal may be attached, or a review of the critical components may be included.


A typical table of contents for the presentation may include: highlights, background and objectives, method, approach, model performance, how to use the model, key predictors, analysis of predictors by decal, implications of results, next steps and model code.


The highlights display an overview of model results. The background and objectives review the purpose of the exercise. The method describes how the results were achieved.


A summary of the approach should address the type of model developed and the number of models.


Model performance is a critical ingredient. Make sure you know what sample was used. Was it a hold-out sample that was used specifically to report on model performance, was it a totally separate sample or was it the same sample used to develop the model?


The principal model predictors should be outlined. Marketing implications and how to use the results should be clearly outlined. Any logical next steps for testing or future analysis should be included. Finally, the model algorithm or rules should be included.


Aftermarket. Your relationship with your consultant doesn't end when your model has been completed. Make sure your consultant will still be there to answer questions and provide further assistance.


Sam Koslowsky is vice president of strategic analytics of the marketing planning and analysis unit of Harte-Hanks Direct Marketing, New York.
Share this article:
You must be a registered member of Direct Marketing News to post a comment.
close

Next Article in Data/Analytics

Sign up to our newsletters

Follow us on Twitter @dmnews

Latest Jobs:

Featured Listings

More in Data/Analytics

Word to the Wise: 100% Viewability

Word to the Wise: 100% Viewability

100% viewability is quite the myth.

12 Big Data Facts for Marketers in 2014

12 Big Data Facts for Marketers in 2014

The idea of Big Data is nothing new, but its potential to solve today's problems and spark innovation is unprecedented.

Harvard Prof: Marketers Need to Step Up Their Predictive Abilities

Harvard Prof: Marketers Need to Step Up Their ...

Statistics expert Edo Airoldi says data must be paired with predictive analytics before marketers can truly forecast customer behavior.