Predictive Analytics with SAP (1)
What is predictive analytics and how to get started with a predictive use case?
I presume you are hearing more and more talking about predictive analytics as this has become another common IT buzzword.
What is predictive analytics?
Explaining predictive analytics is pretty straightforward: you basically want to learn from the past to predict future outcomes. We often hear, in private and professional lives, that we should learn more from our past don’t we? In IT project management, I was always told that lessons learned is key in order to continuously improve and achieve success. So why don’t we try to learn from our past corporate data (external data can be very interesting too) and predict the future?
The first question you might raise will probably be “what do we want to predict?”
I’ll try to wrap up a few ideas to make it clearer:
- Imagine a telco company (just a random example…) that is able to predict the probability that its customers will churn (unsubscribe their contracts and leave to the competition). Wouldn’t that be a great value for the marketing department in order to retain customers?
- Imagine an industrial company able to predict when its industrial machines will breakdown. Wouldn’t that be valuable information to allow proper machine maintenance planning and avoid blocking a whole production line?
- Imagine a bank able to understand who might be interested in new financial products. Wouldn’t that help the bank to advise its customers efficiently?
This was a short insight into the predictive world, but there are plenty of other use cases as you can see in the image below. Some are specific to industries or departments but with a little imagination, you’ll be able to find a bunch of other ones.
Got it? Now I’ll try to predict your next question: “how does predictive analytics” work?
You first want a use case and need to start describing it. If we take the Churn case for the Telco company then you will need to understand your customers’ behavior and try to predict the outcome of the case (the probability that a given customer will churn) based on his/her behavior.
The behavior of your customer might be predicted by the use of his phone contract overtime (number of calls made, of SMS sent, of calls received). Some segment of customers might also have a higher risk of churning (age group, number of products subscribed, geographical location) and therefore customer’s data will be important here.
This is basically how predictive analytics works: for a given use case, you need historical data to learn from in order to predict an outcome (the target variable). Finding the relationship in your data to predict the outcome is the essence of a predictive model which is based on statistical algorithms such as classification or regression models.
OK, so “do I need to be able to program statistical algorithms”?
No don’t worry, you don’t need a Ph.D. in mathematics or statistics to start with predictive analytics and can rely on new software that will simplify the game tremendously.
I like the analogy that Andreas Forster from SAP made during a presentation (I definitely recomment reading the very interesting SCN Posts from Andreas): SAP Predictive Analytics is like the “Auto” mode on a reflex camera. You can create great predictive models without needing much statistical knowledge. But if you want to start spending time to enhance your models, than you have all the options to do so.
Feeling more confident? Now I guess you wonder “how reliable can predictive analytics be?”
Well, probably somewhere between crystal ball gazing (where everything predicted is supposedly going to happen) and pure randomness (where the odds are 2 to 1). The game is simple, the more data you have and the more you’re able to integrate the new data into your predictive models, then the better the outcome of your predictions will be. I presume you start getting the link with Big Data here (I highly recommend Martin Gutmann’s recent post regarding Big Data by the way).
Again, it’s not about fortunetelling, sorry to disappoint those who were expecting Sunday’s lottery numbers, but a way to understand much better from historical data (customer’s behaviors, buying patterns) and therefore be way more reliable on what might occur in the future.
Is predictive analytics comparable to business intelligence?
Predictive analytics actually goes a step further than classical business intelligence (descriptive and diagnostic analytics) in terms of analytics‘ maturity.
Business Intelligence encompasses the tools, techniques and technologies to transform raw data into information (such as dashboards or reports) that can be analyzed by a business analyst or a Manager. A sales Dashboard showing aggreated and detailed data on how a company’s shops, employees and customers are performing is of a great value for a sales Manager to interprete results and make decisions.
Predictive analytics doesn’t rely on a human analysis but on predictive algorithms instead, which is from my point of the view the main difference, and pushes the automation one step further. If you’re able to add predictive outcomes on a sales Dashboard such as how your company’s shops and employees are likely to perform in the next months, then you’ve definitely added value to your BI Dashboard.
The decision still needs to be taken by a person though (like contacting the customers that are likely to churn in order to retain them or pushing a marketing campaign to the targeted audience). If you try to go one step further, which is automating the decision making, then you’re reaching for the Holy Grail and going prescriptive. But let’s concentrate on predictive analytics here.
Next week, we’ll introduce a use case that we’ve been working on at Swisscom to predict an outcome regarding dunning activities (payment reminders). It’s hard to hold my horses so I’ll just throw a few facts and I let you make the case up:
- Tens of thousands of customers are dunned on a monthly basis at Swisscom and gently reminded to pay their due phone, internet or TV bills (yep, we’re not all good customers paying our bills when we should…)
- Dunning customers is a costly process as data needs to be thoroughly analyzed first. After that, customers need to be reminded to pay their bills (by different means such as internet, web or letter reminders). Imagine the cost of sending tens of thousands of letters per month and the cost of analyzing a percentage of people contesting the fines received…
- Haven’t you ever received a dunning letter but actually validated the payment a few days before? The cost of sending the letter to you and time analyzing your complaint could eventually have been saved.
Do you see a way to use predictive analytics to optimize the dunning process here?
See you next week for the next Blog Ticket: Predictive Analytics with SAP (2) – How to optimize the dunning process with SAP Predictive Analytics