Over the years, we’ve helped many organizations bring predictive modeling in-house and have learned a lot along the way. Below is a “best of” list of ten tips that our customers helped us put together to make the modeling process – from idea through execution – go a bit more smoothly.

OLYMPUS DIGITAL CAMERA

1. Establish buy-in before your first modeling project

Establishing leadership support before your project gets off the ground will ensure that your modeling efforts have a strong foundation and will not flounder as you encounter challenges along the way. Communication during the modeling process is key and this is the first step. By openly communicating about your predictive modeling project from the start, you can be clear about expectations, outcomes, and the process before you begin.

2. Pick the best person for the project

The person in charge of predictive modeling will have to collect, analyze, and work with the results from your data. This person should be a creative problem-solver who is willing to learn. Ideally, this person is someone who is already working with the data. A person who understands your data – how it is stored, labeled, and used – can learn any statistical techniques they might need to know, but it is harder to learn the data while doing so.

3. Data preparation is 80% of the modeling process

You may have heard of the term GIGO – garbage in, garbage out – which is a creative way of saying that a model is only as good as the data you use to create it. That said, data preparation is crucial as you’re gearing up to build a model. This means creating new variables, checking for missing values, and making sure your dataset is “clean”. If possible, create a data preparation process that is repeatable, which will ensure that your time is well-spent. It’s worth the investment.

4. Avoid common mistakes

Talk to anyone who’s worked with models to see what has worked for them, what hasn’t and what advice they have. Accurate models mean better results, which is incentive to avoid any of the “easy” mistakes. There are lots of reputable sources online to check out as well – I’d start here and here.

Five Data Preparation Mistakes and How to Avoid Them

Top Five Predictive Modeling Mistakes

5. Incorporate as much data as possible

Gather as much data about your historical clients as possible. In collecting data, strive for the most complete picture of your constituents that you can create from your data. Once you know what information you have consistently been collecting about prospects, you can evaluate whether there are gaps that could be filled by collecting more or different data to impact your analyses going forward.

6. Get creative

There are plenty of ways to use the outputs of a predictive model. For example, in Higher Ed, by summing the probabilities, you can get an idea of your incoming class size. By weighting financial aid with enrollment probabilities, you can get an idea of your expected financial aid outlay. B In Fundraising or Direct Marketing, you could determine who is likely to give a first gift, or make their first purchase, or who is likely not to renew a membership or not be a return shopper.

7. Communicate with stakeholders often

Staying in touch with stakeholders works both ways – making sure that they are informed of your progress keeps everyone’s expectations on the same page, and checking in with them keeps you engaged with the bigger picture and how modeling will support your organization’s objectives. When framing communication, keep your end audience in mind – in lots of cases, a dashboard or visualization will convey what you’ve learned better than a series of raw numbers.

8. Choose the right vendor

When you are using the results of a predictive model to make key decisions within your institution, it’s important that you can own the decisions and the knowledge that went into the model. Additionally, make sure that you have the ability to change and update your models as needed without having to pay heavy fees. If you aren’t confident in your predictive modeling skillset, look for a knowledge partner who can teach you best practices and techniques. At the end of the day, you’ll need to be confident that you can explain the model scores to your team and how you got from raw data to the decisions you’ve mode.

Get a Free Trial of Rapid Insight Predictive Analytics and Data Blending Software

9. Measure your progress

As you’re implementing your model results, keep your KPI’s and the institution-wide goals in mind. Track the impact that your model is having. For example, if you’re using the model to decide which prospects to mail, keep track of the response and response rates, and compare them to previous mailings. By reducing the number of people you mail based on your models, you can calculate an ROI from the amount of postage saved.

10. Reduce, reuse, recycle

The process of building a model shouldn’t be a single-use effort. Reuse the data cleansing and extracting processes for other analyses. Reduce stress by automating reports whenever possible to keep everyone on the same page. Finally, recycle the knowledge you gain through your organization – the value of a model isn’t limited to the scores. Throughout the process, you’ll gain an understanding of your data, historical trends, and an idea of what to expect going forward – all of which is valuable and not necessarily contained in a single model score.