How and why you need to tame predictive analysis
In recent weeks we’ve seen incredible action in the intelligent assistant market. Google announcing the Google Assistant and associated devices to take on Amazon’s Alexa, Microsoft at Ignite 2016 touting a new and improved Cortana, Salesforce launching Einstein, and Viv — a start-up by the developers of Siri — bought by Samsung. These AI-driven enhancements are becoming ubiquitous — from customer service to marketing, from the home to the car, and from the factory to the community.
They all have one thing in common — they use predictions to deliver results which help you. Predictions are the result of predictive analysis, which, like data science, is red hot in the minds of executives and CMO‘s. So hot in fact that they are well overhyped. Here’s why.
Siri 1 Allo 0
Google’s Assistant in its iPhone incarnation is known as Allo. I told Allo my daughter’s name — which it acknowledged — and then asked “who is my daughter?”. Allo had no idea. I then told Siri, and stated to Siri “My daughter is Misha”. To which Siri replied “OK… but I already know that”.
This example illustrates that AI is intriguing, but nevertheless, pretty spotty right now. Salesforce has built Einstein into the core of its platform. The thought of Einstein being spotty sends shivers up the spine of CMOs. According to very recent research from Accenture and CSO Insights, 59% of global sales executives say they have access to too many sales tools and are bombarded by too much disaggregated customer data to be effective. Another 55% say their sales tools are an obstacle to selling.
A half-intelligent assistant is their nightmare come true. But there is good news. Unlike the rush to hire 100 data scientists and deploy predictive analysis “somewhere” in the business, the likes of Einstein and Siri have a tame environment. Siri can bring together a lot of information from an iPhone which Allo cannot yet access.
The last mile — predictive analysis in context
Think about Einstein in context. A customer support person will be notified about a failing customer device via the internet of things, Einstein will have a prediction for the best solution based on CRM and other operational data about the device and the model’s history, and the person will approve the action. Done.
Success requires the context to be limited and the outcome to fit patterns of use cases, whether they be support or marketing or advertising or customer engagement.
The hardest part isn’t finding data scientists to employ, it is making prediction which supports everyday use cases and which stick because employees trust those predictions. This is the “last mile”. You might say start with the end in mind.
With that in mind, here are the 3 big questions of predictive analysis
1. What question do we want answered? What is the use case and how does it integrate with current systems and align with business outcomes?
2. What can we do when we have that answer? If we knew the answer do we even have the resources, skills, culture, capacity, will, knowledge and system to be able to act upon it?
3. How do we influence the actions in the way we desire even when we know that we can act. We’re willing and able, but can we actually produce a result?
Answering those questions before rushing out to hire 10,000 data scientists will bring your predictive analysis efforts in from the wild, and potentially tame the beast.Back