Guest blog by Jim Kenyon
Jim Kenyon is the Director of Operations for Optimization Group and the heart of the ROMI process. A published author, Jim is fascinated with the methodological application of technology to solve business and scientific problems.
“We use SPM because it lets us quickly and easily build predictive models that produce useful and usable results for our clients.” - Jim Kenyon
The Optimization Group is frequently engaged to help clients understand how their marketing efforts impact lead generation and sales. These requests come in many shapes and sizes but tend to coalesce around:
Which media are “moving the needle” and at what spending levels?
How do the different media work together?
How can I improve targeting for my direct marketing efforts?
The first two questions are typically answered through what we call a Return on Marketing Investment (ROMI) analysis. The second question focuses on understanding who your customers really are and finding ways to practically identify “Look-a-Likes”.
A “What-if” simulator is frequently one of the project deliverables for answering the first two questions. After we’ve done the heavy lifting of getting all the data into shape for model construction and building a useful model, clients want to know “what happens if we spend more on TV and less on print?” Of course, we could offer to take a set of values and process them with the modeling tool, itself. This could be good for OUR revenue if clients had the budget and time for processing each scenario of interest. Unfortunately, this is impractical since clients usually need to look at different scenarios iteratively, and do not have the time or budget for additional consulting hours.
Instead, we implement the resulting model in a free-standing “What-if” simulator, accessible on-demand. Using the Salford Predictive Modeler, we are able to offer multiple languages for exporting learned models, allowing us to take the generated code and drop it into our simulator framework with minimal programming effort. It is worth noting that we try to minimize delivering simulators in Excel as it is too easy and tempting to “fiddle” with the model, itself, to just “see what happens”. An unintended consequence of too much “fiddling” is that the model starts to produce predictions that are based more on what one wants the data to say, instead of what the data actually says. For this reason, we typically implement simulators online as “micro-sites” that let the user tune parameters at will, all the while “keeping the guards in place to avoid accidents.”
Models in an External Workflow
The third question, “How can I improve targeting for my direct marketing efforts?” typically requires the ability to insert the resulting model into an ongoing, automated workflow such as lead scoring. Consider a company, ABC Co., which has traditionally executed direct saturation mailing campaigns. ABC Co. knows that only 5% of the households in its trading area are actually useful leads. That means that 95% of the pieces in a saturation mailing will not produce results. Being able to more accurately predict which households are likely to transact can significantly reduce ongoing direct marketing costs.
Again, after doing the heavy lifting of getting data into shape for modeling and actually building a useful predictive model, the Salford Predictive Modeler facilitates our efforts by generating source code that drops into our lead scoring framework. We are able to hand off a free-standing application that integrates with our client’s lead-gen database and lets our clients automate lead scoring to improve the likelihood of response from direct marketing efforts.
Why is this important?
The Salford Predictive Modeler allows us to quickly and accurately translate learned models into simple, easy to use client-facing tools. The client benefits? Accurately reflect learned models, shorter project timelines, and lower costs to build the tools. That’s a Better, Faster, Less Expensive story we can all love.