Sentiment Analysis

Sentiment analysis has many applications and just a few are discussed here. The application of sentiment analysis provides the ability to identify and categorize opinions or perspectives from within your data. For example, Is your customer happy? Is your student focused on the lecture? Will your customer buy this product?

Additionally, sentiment analysis is one of the most widely available AI solutions with several pre-trained solutions such as Azure Cognitive Services, SQL ML Services and many different custom approaches. As we demonstrated in our workshop, each approach can be implemented very quickly. The real questions when applying sentiment analysis is what sentiment data would drive business value and do you really want to know what your customers are saying about you?

Common use-cases:

• Call center (Automated complaint tracking, customer perspective, rep management)
• Company perspective (Social media, product reviews, complaints)
• Quality of care (Healthcare delivery)
• Deal closing likelihood (Customer perspective)
• Internal Communication Monitoring – Score internal solutions like Yammer or Teams to get the pulse of the workforce or use it as a “policing tool”

Quality Score

Quality scoring, i.e. providing an intelligent score in order to evaluate the ‘quality’ of that content. In the most common scenario, a quality score (or likelihood) can quickly be applied to your sales organizations deal data. Deal or no deal?!?! Helping your sales team identify quick win opportunities or where more effort could be applied to high value targets. Quality scoring can even be implemented with automated model training and low-code no-code solutions such as Azure ML Services and AutoML.

While quality scoring is basic predictive modeling, BEWARE as they can easily introduce bias depending on your feature input. You must understand the application of good vs bad bias.

Common use-cases:

• Opportunity Scoring (Sales, Brand Location, Recruiting Evaluation)
• Customer fit evaluation
• Deal fit evaluation (these 4 products and this customer… or … these 6 and this customer?)
• Employee performance
• Employee or Customer retention
• Care approach evaluations (input expected next step, receive quality of approach score)   

Forecasting

Forecasting is the ability to predict future outcomes, such as sales revenue or inventory quantity requirements, based on historical patterns. This is a practice found in every industry and organization, although typically driven by employee tendency not factual patterns and is a very time-consuming manual process.

Unfortunately, often AI forecasting is viewed as a risk, primarily because trusting an algorithm to make financial predictions in a chaotic economic climate can be unnerving. And with that, if it was perfect, we’d all be rich from making better investment decisions or at least would have put our money in cash before Covid-19.

For a quick and dirty approach to AI forecasting, leverage the output of AI as your input to human intervention. Allow the forecasting to guide your own forecasting approach and save time in the process. Like Quality scoring, AI forecasting can be implemented quickly with Automated model training and low-code no-code solutions.

Common use-cases:

• Sales Revenue
• Inventory Quantity Expectations
• Staffing Allocation and Alignment
• Student Drop Out Predictions

Recommender

Recommenders are another excellent quick win AI solution. Specifically, because they are very easy to build and are well known, not just in Data Science but in life.  What is a recommender? Think online shopping, adds on your phone, Netflix content recommendations, etc. Recommenders ‘recommend’ something based on community patterns found within your data.

Recommenders are tried and true solutions, well documented and with a bit of creativity, or not, can provide great business value. Like sentiment analysis solutions these could also be demonstrated in hours.

Common use-cases:

• Product recommendations
• Customer targeting recommendations
• Care approach recommendations (recommend the most likely next steps in the care process?)
• Staffing recommendations (which staff to apply for this case? How many employees do I need in this territory?)
• Predictive Maintenance (do I need to order parts next? Where’s the best place to look?)
• Developmental needs for employees
• Price for services, products, policies

Knowledge Mining

Knowledge mining is a new term for an old solution; document indexing. Traditionally, document indexing and the ability to scan common indexes and terms in a search engine has been a manual process of creating indexes that are linked to the original documents.

With new AI solutions knowledge mining is just the application of recognition (form, video, image, speech to text, etc.) integrated with the power of search. Thus, automating content storage of any kind with the ability to search and find relevant content that is generated from the content itself and not dependent on the user’s interpretation. New concepts are actively being introduced such as knowledge store projections that enable user validation, alerting, reporting of content and innovative approaches for additional business process automation.  

Like the previous examples, knowledge mining solutions can now be implemented quickly with developer platforms like Azure Cognitive Services, Azure Search, Azure Functions and low to no-code solutions like the Power Platform AI builder and Power Automate. Additionally, knowledge mining solutions are familiar, think organizational Google.

• Index PDF’s such as contracts, invoices and order summaries and trigger down stream business processes
• Index Call Center interactions and automate alerting
• Index Educational content
• Index, link, mine and automate Healthcare data enrichment

Recognition

Recognition is a rapidly growing application in AI. We see this with the facial recognition built into our phones, image recognition on social media or selecting the capture images for multi factor authentication.

Recognition might be considered one of the most advanced of the quick and dirty solutions, but like the others, they are well documented and more and more pre-trained predictive platforms are being introduced into the world. Recognition can be revolutionary in spaces like Healthcare, Insurance claims or even Education. Developer platforms like Azure Cognitive Services and low to no-code solutions like the Power Platform AI builder can really make this a quick and dirty solution.    

Common use-cases:

• Disease Recognition
• Document Content Extraction
• Person Recognition
• Defect/Quality Recognition (Predictive Maintenance, Quality Assurance)
• Sentiment Recognition (Student Evaluation against course material, patient deterioration, patient satisfaction)

Anomaly Detection

The most fun case of Anomaly Detection is clearly the application of finding all Waldo’s on the planet from every book within seconds. Waldo is in fact the anomaly within the image that we’re all searching for. However, how much value does finding Waldo provide your organization……. unless you’re Martin Handford evaluating the quality of the latest installment of “Where’s Waldo”. In that case, please contact Spyglass MTG, we can help with your quality review of “Where’s Waldo 2020”. i.e. Sitting at home watching Netflix and making TikToks or getting criticized on Facebook for being on a crowded beach without a mask.  

Anomaly detention is another tried and true application of AI. Fraud modeling has been around for more than a decade but has traditionally required custom algorithms and a team of statisticians (now called data scientists). With recent introductions in Cognitive Services and open source machine learning packages these use-cases can now readily be implemented as quick and dirty solutions.

Common use-cases:

• Fraud Detection
• Risk Analysis
• Forecasting exclusion – Don’t include in the forecast because it was an anomaly
• Predictive Maintenance – Vibrations from mechanical parts