Above all, it’s imperative that you understand the dynamics of your data before deciding between an off-the-shelf software or a customized, more transparent solution. Knowing which data you will use and how the data sets interact must be defined up front. People typically are better at this than computers, not only people who have a strong statistical background, but also experience in marketing and business. Data governance is also crucial to the outcome and longevity of a predictive analytics model. While collaboration with a vendor varies in every situation, it will be a one-time event for off-the-shelf suppliers and part of an ongoing process when working with a customized analytic solution provider. No matter which direction you end up going, predictive analytics is most effective when applied across all stages of the customer lifecycle.
As a starting point in choosing your approach, sit down with your team and mull over the following five questions:
How vested in the outcome are the sponsors of the modeling project? How will the results be socialized around the organization?
This is a critical question you should ask yourself prior to making any decision about what type of methodology to use when undertaking a predictive modeling project. It’s also important to determine how your model will be measured and the business case to support the minimal lift that the model should supply in order to drive profitable marketing strategies. Off-the-shelf software does have its place in an analyst’s toolkit especially when speed and efficiency trump other considerations. However, if the sponsors and stakeholders of the project have shown interest in not only the outcome but also the path as well as their ability to understand and interpret the output, then software may not be the best choice.
For example, when models have far-reaching implications for operations, sales force optimization or even new service launches, the stakeholder population grows exponentially. Another example is when a model is used for fiscal planning and projections based on what it is predicting (acquisition, growth, retention, reactivation, etc.). Accurate financial planning is key, and understanding the basis for those projections is critical. A custom, socialized modeling approach would better suit these situations as this gives you the power and information to showcase each decision and how and why it was made, resulting in a stronger adoption from your backers.
What are all of the model use cases? Will the individual components be used for operational, sales or marketing decisions?
Models do much more than simply help isolate good prospects for a product or service. The underlying data drivers of the model and how they interact can be invaluable when it comes to developing and executing on the customer communication strategy. Off-the-shelf solutions rarely provide insight beyond a score or who to target, but if that is all that’s needed, you should be in good shape leveraging those solutions. However, if you can provide your marketing organization with key insights into why an individual contact or business is a good prospect, then having a deeper understanding of the important elements in a model will serve you well. For example, if your customized model indicates that tenure is a major driver of sales, your team can leverage that information to prioritize their sales efforts toward contacts with the longest history with the company.
What is the volume and frequency of models being built and how often are they to be updated?
Off-the-shelf solutions are advantageous when the volume of data and model refresh schedule are frequent and there simply is not enough manpower or time to keep up with employing a customized solution. To determine whether this approach or a more custom, adopted methodology is best, take some time and build a 12-month roadmap. Determining which models are going to be needed and how often they will need to be refreshed can provide insights as to what type of solution will best fit your needs. This planning carries the added benefit of involving your stakeholders from the outset. It is also important to understand how the models will be monitored and evaluated over time. Are they degrading or do they need to be re-fit? Some, but not all, modeling software can help with this, so make sure you understand the capabilities of any software-based solution before using it.
How comprehensive is your available data?
Why is this important? If you’re working with limited data, it may not be necessary to employ the rigor that a customized model demands. In this case, an off-the-shelf solution could be a good choice. However, if you enjoy a robust database with detailed transactional and behavioral information, you may want to consider a more custom and transparent approach, especially when answering question #2 above. This solution begins with the development of analytic records, which creates the foundation for significant insights and discovery. The value of this approach goes above and beyond the mere development of the predictive algorithm. After all, the more data you have at your disposal, the more value you can provide to your stakeholders.
What type of data governance has been established?
This question basically comes down to the level of your data’s cleanliness and usability and, more importantly, its consistency across your sales force and business units. It’s common knowledge that the most important part of modeling is getting the data into a form that is workable. An off-the-shelf solution does not determine whether a variable has been coded correctly or if the meaning of a data element is not what the name claims to be. Only you and the analyst know this. The phrase “garbage in, garbage out” applies here. Don’t let a pre-packaged modeling solution make decisions for you. Unless your data is pristine (which is as rare as seeing Bigfoot riding a unicorn across a double rainbow), make sure you go through the process of ensuring your data is clean, current and correct. Otherwise your output may be misleading at best or simply wrong at worst.
Once you’ve answered these questions, you should be well on your way to determining the ideal predictive data analytics solution for your business. Ask When Choosing a Predictive Analytics Solution