How long before we actually see some results here?
It’s a question the TMA team gets during a great many Q&A sessions. Everyone wants a time frame–some idea of the amount of time it takes to start getting actionable intelligence out of any given model.
But much depends upon the level of modeling that you are trying to engage in. There are three levels that you can choose from.
The first is the single model level. At this level, you’ll be building a single model for a single purpose. When starting from scratch, you can generally achieve this level within six weeks or so. However, in truth the single model level is not likely to produce much ROI. That’s not to say it’s useless–this sort of “level 1 model” makes an excellent pilot project, which can be used to demonstrate the benefits of predictive modeling to an otherwise recalcitrant leadership team.
The next is to build a repeatable process at the departmental level. This would be a model that’s run and updated on a regular basis. For example, your marketing department might regularly test the response to different advertisements in different zip codes to find out where it’s best to spend money and effort. They might re-run the model every six months to account for changes in external circumstances and demographics, and to adjust as the model prompts them to adjust. You can build a model like this in about six weeks, and you’ll get some solid use out of it.
The third and final level takes the most time–usually about six months. This is the “practice” level. That is, you’re going to create an analytics practice, a full department that serves the entire enterprise, a department capable of tackling a variety of business problems. As you can imagine, this is the analytics level that produces the greatest amount of ROI over the long run, even if you have to wait longer to see results.
One of the ways that TMA assists clients is by helping them build a true modeling practice, mentoring them so that the practice will ultimately be successful in the long term. It’s a great way to ensure that you maximize the actionable, profitable insights locked up inside of your data.
Remember, any grad student can use modern software to build a model in two weeks. Unfortunately, two weeks isn’t enough time to lay groundwork, provide context, or derive value.
It’s really only enough time to build a really fast rocket with no mission plan.