Predictive analytics is fun. No really, it is. The predictive part of analytics in research has always been of interest to me. It's like placing a bet or forecasting the stock market. Statistical modeling and algorithms are nothing new to market research but with the recent big data push and more attention from big money businesses being paid to analytics, this structured form of ROI-like decision-making is growing in the industry. Much of this is driven by the sheer amount of data at our fingertips from customer databases, social media campaigns, Google AdWords, and other sources.
In its simplest terms, a predictive model is a step-by-step method of calculations using data to better inform reasoning and business decisions. Years ago, analysts were asked to transform themselves into consultants, and now consultants are asked to become data scientists. All good news for businesses like Drive Research.
"Roads? Where we're going (with my algorithm) we don't need roads".
Predictive models are very common in any type of feasibility work in market research
Clients commission a firm who have experts capable of creating algorithms to predict the traffic for a new store location, estimate the number of assisted living beds a market can sustain, or even predict how much ROI an advertising campaign brought to a business. This type of data-driven decision-making is vital nowadays. Many organizations still fall in the trap of blindly spending on advertising without knowing key metrics like impressions, clicks, actions, likes, etc. So they have no data to help guide future decision-making.
Keep the following 6 tips in mind for your next predictive analytics study:
- Tip 1: Start thinking about your process early. Before the project kicks off, know your end-goal and the major pieces of data it will take to arrive at those conclusions. This will ensure the critical information which needs to be tracked is indeed tracked or at least offers the ability to be synced to another database using a unique identifier.
- Tip 2: Determine what's available. Survey the landscape of data available to you. Can you involve multiple channels of data (Facebook insights, Google AdWords, website data, demographic data, customer data, etc.) to add more reliability to your model? Figure out what you have to work with.
- Tip 3: Lay the ground work. All good algorithms start with an outline. I typically start with the last step or the end-point. What is the main objective of your study such as a specific metric or number (cost per customer, cost per impression, etc.) and then work your way backwards. If you have a lot of sources of data which go into determining the end-goal it can prove overwhelming if you try to wrap your head around Step 1.
- Tip 4: Understand any limitations. What data cannot be built into your model? All forecasting models attempt to base estimations using a controlled number of variables. Unfortunately, a predictive model will not be able to anticipate major market fluctuations or other game-changing impacts. For instance, if your model calls for a major Facebook advertising campaign based on 2013 ROI data, it still won't be able to predict a potential privacy scare and hack of Facebook data which may occur 2 months from now (in theory) which scare users away from Facebook.
- Tip 5: Plug and play. Once you've constructed your model, plug-in the numbers and see what your final results looks like. Use an eye test to determine if adjustments need to be made and at what step of the model. Sometimes seeing some end results will help keep your focus and momentum so you can then go back and adjust rather than getting stuck in the mud.
- Tip 6: Always remember, you do not have a DeLorean, and you are not Marty McFly. Predictive models are just this, predictive. There is no right or wrong answer. You cannot predict the future. It is your job to take all of the data available to you and put together a sound and well-reasoned model. Have faith in it and retrace your steps. If all predictive models were accurate we'd still be watching the XFL on our couch drinking a New Coke while we are heating up our Colgate Kitchen Entree in our microwaves.
Drive Research is a feasibility study company located in Syracuse, NY. We use primary and secondary market research for predictive analytics studies to determine the viability of a product or service through a market analysis. Contact us at firstname.lastname@example.org or at 315-303-2040 for more information about our firm.