November 1 by Kingman Tang
Growing up watching Sesame Street, I remember a wonderful song that was regularly played as a fun object lesson. Viewers would be shown a group of four objects as the song with the following lyrics would play:
One of these things is not like the others,
One of these things just doesn’t belong,
Can you tell which thing is not like the others
By the time I finish my song?
Did you guess which thing was not like the others?
Did you guess which thing just doesn’t belong?
If you guessed this one is not like the others,
Then you’re absolutely right!
Machine Learning Platforms
When I think about machine learning platforms like the recently announced Einstein from Salesforce, the Jeopardy winning IBM Watson, or Microsoft’s Azure machine learning, I see that they are really good at providing algorithm development and solution building tools. This allows the rest of us to construct machine learning use cases on top their platforms. Amongst the three machine learning as a service platforms mentioned, Microsoft Azure and IBM Watson are ahead of Salesforce’s Einstein in terms of maturity in the market place.
Azure Machine Learning seems to be the most atomic of the three in terms of basic building blocks, leverage and apps already built. Its ability to incorporate existing R (a very popular open source machine learning and statistics programming language) and Python code makes it suitable for many data scientists.
Watson, the high profiled AI platform, makes weather predications, fights cancer, and even wins in Jeopardy. Despite its highly publicized profile, Watson has not been a general purpose machine learning platform. Until recently, Watson appears as purpose built solutions with a blackbox approach along side some published APIs. Anecdotally, this has translated into Watson requiring a fair amount of professional services to stand up a specialized application. At the World of Watson last week, IBM introduced a new Watson platform that resembles a more general purpose machine learning platform that is accessible to everyone.
If Azure machine learning and Watson appear to be preschoolers, Einstein is simply Baby Einstein right now. Beyond the big marketing splash at Dreamforce last month, it is too early to know what is really under the hood aside from the amalgamation of AI companies (e.g. RelateIQ, MetaMind) that Salesforce acquired over the last two years. Make no mistake, Salesforce will mature in its platform offering, it will just a little time.
Applied Machine Learning
In contrast, applied machine learning takes a practitioner’s perspective versus a platform toolkit viewpoint – that is, it takes data science, algorithms, and domain expertise together and packages machine learning into usable, real-world business process insights that are easily digestible by a non-expert user. The end-user is not a data scientist, nor a data analyst or even a subject matter expert on the ins and outs of your business process. As a result, not only must the recommendation be digestible, but it also must be trusted by the user. The moment your insight is not trusted, is the moment your recommendations will sit on the shelf and gather cobwebs.
So how do you build trust? In Quote-to-Cash, the user is your sales rep trying to generate a quote, and the applied machine learning piece would be in the form of a product and or price recommendation with rationale explaining why the recommendation was made. The reasoning behind the recommendation could be that 10 other sales reps with a like-profiled customer (e.g. company size, industry vertical, geography) purchased these recommended products and closed at a specified price off of the list price. Presented with this rationale, the sales reps can easily digest the logic of the recommendation and can now choose to accept or reject the recommendation.
Applied Machine Learning Use Case
One specific example of Applied Machine Learning is in the area of Configure Price Quote (CPQ). CPQ technology, with machine learning, helps an organization’s sales force navigate toward the optimal deal by utilizing the data flowing through Quote-to-Cash applications, as well as data from other sources (including ERP systems, websites, social data, usage data, and sensor data). Here are three specific sample CPQ use cases with machine learning:
1. Quote scoring that predicts the probability of winning the deal and gives sales reps recommendations to improve quoting and increase win probability. By learning from historical quote data, quote scoring helps all sales reps sell like the top 20%.
2. Cross-sell and up-sell recommendations for products and services that customers are most likely to purchase based on analysis of past purchases and similar customers. Cross-sell and up-sell recommendations help sales reps maximize deal size and increase account penetration with machine learning insights.
3. Pricing intelligence produces optimal price or discount levels for each deal based on machine learning analysis of the deal characteristics and sales history. Pricing intelligence helps sales reps and sales operations deal desks through the quoting and negotiation process with guidance for initial, target, and walk-away price levels for every deal.
Apttus’ applied machine learning is not like the others, but we do leverage Microsoft’s Azure platform. With over 25 Quote-to-Cash machine learning use cases identified, and the future is bright for Applied Machine Learning technology and solutions.