A Look Back at INBOUND 2018: Insights on Predictive Analytics
So it’s been over three weeks since INBOUND, but I’m still pumped over the new technology and innovative strategy and tips I learned while there.
My approach to session choice was quite different this year. 2018 was my fifth year to attend INBOUND and I decided to shake it up and ran with topics completely out of the norm for me.
In the past, my choices have run toward sales, or managing and growing an agency and though I still did a few sessions like that, I chose some with topics that honestly made my head feel a bit full. One of my favorite of these sessions was Predictive Analytics presented by Katy Robbert, CEO and Co-Founder of Brain Trust Insights.
The presentation started off with a brief outline of the hierarchy of analytics – how they are mapped and how they work together.
- Foundational Data is the quantitative data like Google Analytics, CRM data, etc. This data set is quantitative and tells you what has happened.
- Diagnostic Data is qualitative data that tells you “why” is happened. This could be customer support/call logs, consumer surveys, market research – anything that points to the behavior of your customers.
- And finally, we have Predictive Data that will (hopefully) tell you what will happen next, or give you a prescription of what you should do next.
A lot of companies are stuck at the descriptive analytics, looking backwards stage, and Katy compared it to driving a car going forward, but you’re looking over your shoulder behind you. You’re going to crash because it’s kind of like driving with your eyes closed. By using only historical data if something changes right in front of you – you won’t even know it.
We all use a type of predictive analytics every day but don’t realize it – the weather forecast!
There are two types of predictive analytics - Driver Analysis and Time Series Analysis and both of these types of analysis answer the specific question, “Why”.
A good example of this would be, “Why Tuesday do our tweets suddenly go viral on a Tuesday?” Driver analysis would be used to pick apart the different actions that led the tweets to go viral – and once you found what was the cause, you’d want to replicate those actions.
Then time series analysis would be used to determine what day, time, and month likely caused this to happen. Time series analysis is the analysis that marketers most commonly use. Historical data would be used and projected forward. An example would be to use 5 years of Google Trends data, feed it into a predictive algorithm (either proprietary or purchased) and receive a projection for the next 365 days. You could adjust the projection down to smaller pockets of time that would better fit your plan structure.
So how do you get predictive analytics started? Follow these 5 steps.
1. Determine the Project
This will allow you to set the strategy, understand goals and the desired outcome. The next step is to pull data. Data is necessary because your forecast will be based on mathematical principles. As mentioned before you can use Google Analytics, your CRM data, etc. If you have no proprietary data, you could use Google Trends which is a useful data set that is publicly available and often used by marketers for general information.
2. Prepare Your Data Set
Data can be exported as a text file, but you may need to transform it into a CSV file depending on how your algorithm is structured – just make sure all the data is structured correctly to enable all to run smoothly.
3. Pick Your Variable
You can use driver analysis to figure out which variable you want to predict. Using our Tuesday viral tweet example, if the goal is now that everything we retweet on Tuesday goes viral, we would use driver analysis make sure it aligns with our goals and outcome. Once we know that our variable is retweets on Tuesdays, run a time series forecast to determine what that looks like and what other elements would need to be done in order to get that desired outcome of viral tweets.
4. Run Your Predictive Algorithm
Use the data you’ve pulled and prepared, and the information you’ve determined you want to predict, and run a prediction. If you have multiple questions, best practice is not to try to answer more than one using the same algorithm.
5. Build Your Plan
Use all the information you have gathered from your algorithm and start to plan for what you want to accomplish, and how to execute it. The benefit will be that you can really understand what may be coming in the future and not just what has happened in the past.
Bottom line, predictive analytics is a supplemental data set to enable you to create better planning – forward-looking planning with the ability to try to determine what’s coming down the line.