USE CASES

In an initial state, the possibility of using AI or machine learning algorithms to predict outcomes is not always clear and even if it is, securing the outcome of the project means determining the business case and the actual need that should be fulfilled.

This can be done in a pre-study where the needs are investigated, the data is checked for quality and availability and this leads to the definition of KPIs and the return of investment (ROI) the project should accomplish. Determining these factors is crucial in steering the project and making it certain that the initial needs are still addressed.

Outcome from this phase includes a set of hypotheses of what needs or issues the model should address, a first report of the examination of the data and a business case that concludes on the bigger picture of KPIs and the return on investment.

Hour-driving tasks that are taking away costly hours from persons in an organization can be automated by machines. In our case by virtual machines that can process data and do robotic work pieces in order to automate mundane tasks or to do preparations before human beings take hand and add value. From taking care of answers on forms to organising data input from multiple sources, A Robotic Process Automation instance can do the job based on rules or on rules combined with machine learning algorithms.

In order to create order out of chaos that is data on the market or what customers are using what product, segmentation is a natural way of limiting the multitude of the data so that it can be interpretable. Clustering is one method and is used in recommendation algorithms, decisions on what products to develop for the market and much more.

In the case of when recurring customers choose to leave you for a competitor or because of perceived bad customer service – the chance is that you have collected data on the occurrence. Even better yet – the data could predict when or if this is about to happen again. “Churn” is the term covering the loosing of a customer because of his or her decisions and actions.

A churn prediction model is built on historical data about the customer concerning characteristics, behaviour, previously bought products or services, customer service contacts, web page interactions and so on, which is then combined with analysis of the time of actions as well as other external data to create a model that can predict churn given these variables on another customer. The prediction gives you time to update your offer to the customer or oversee your relative place on the market or find troubles in your offering that makes customers turn away.

The deployment of churn prediction models’ use can be multiplied in combination with segmentation analysis of products, markets and customers.

In order to draw insights from a large set of documents – Natural Language Processing is the set of algorithms that let machines understand written language in order to perform human tasks on a large scale. NLP can be used to automate processes and to structure data from documents into a searchable database as well as understand the context of the data and links to other entities. NLP algorithms can also understand underlying meanings, references to earlier mentions

Sentiment is the concept of emotional charge attached to a sentence or a word in an expression. In spoken or written text, our opinions are often coloured by positive or negative words which computers naturally have a hard time to interpret. Natural Language Processing is the set of algorithms that can decode the sentiment of conversations, reviews, news articles or other human expressions and collect that data on a massive scale. What are all the reviews on the Internet expressing about your product? How negative or positive is the tone in a client’s recent call to your customer service? An approach with NLP can solve these issues.

For finding entities in vast amounts of images or classifying them based on what is depictured – Image Recognition are the machine learning algorithms to use. This technology has given us cancer diagnosis systems for X-ray images, facial recognition, price prediction of crops from satellite images and much more. Image recognition is how we teach machines to understand what images contains and can thereafter be used on a large scale in almost any kind of application.

Many times, interesting insight in the data lies not in the volumes but in the signals that stand out from the mass. Anomalies can be outliers in the data that represent something different and be a sign of errors that need attention or be the pattern in the data that leads to a successful training of a machine learning algorithm, for example in Prediction Maintenance.

With many applications, predictive maintenance provides the chance to avoid unplanned downtime, maximize usage and to optimize resources in the production chain. Data from machines or devices is used to train a machine learning algorithm on what has foregone a breakdown or a need for service. Combined with anomaly detection, predictions can be made on when a machine is on the verge of breaking down and maintenance can be planned in advance.

Using machine learning to make predictions on maintenance allows for further integration of both purchasing (spare parts) and staff (conducting maintenance) functions. The relief can be measured for not only the production line but all dependant departments and lead the way in best practice on data collection and analysis.

Forecasts can be done in many different ways. Where AI and Machine Learning can bring immense value to this process is when the results of a forecast is reported back in order to update the forecast to learn and become better. This is what the feedback loop does, and it is the core of machine learning. Adding many different data sources and letting the system teach itself to a better forecast is just one of the many cases for sales and cost forecasting.

Combatting financial crime like fraud and money laundry takes place in a complex setting of transactions. Where a single money wire might not rise suspicion, machine learning can find criminal behaviour and single out unlawful occurrences in an amount of data that is incomprehensible. This case is a perfect setting for AI and Machine Learning when patterns in the data can make all the difference.

For loan decisions and credit scoring, an automized machine learning process is nearly a necessity today. For quick answers on loan requests and to avoid human bias, using rules in combination with machine learning makes for a fast and robust solution that can be cost efficient as well as safe and explainable.

Pricing can be a complicated process which in many cases is built solely upon assumptions and traditional rule of thumb. To optimize price, using the complete data of the organisation in combination with market data and competitor data is a solid base for use of machine learning algorithms. Not only to determine the price but also to find out what impacts the price elasticity and what strategies could be deployed if the market changes. The feedback from purchases should play a vital part in pricing and provides valuable data every time a purchase is made. Machine learning can tap into that source and improve the model’s robustness and strategic importance.

Chatbots are common apps that are reducing the workload from customer service personnel and provides quicker answers to simpler questions without having to wait in line. Even though chatbots can be made by rule-based algorithms, Conversational AI is the concept where machine learning is used to learn from conversations in combination with using algorithms that learns from the text itself; understanding entities, implications and references to earlier mentioned subjects. These models often use NLP for this. Using conversational AI can be powerful when gathering data from customer service functions is used together with other data inputs, as for example web scraping of reviews in order to draw insights on what customers really think about your products.

The landscape of where opinions are expressed in many different media across a multitude of platforms, it is not easy to keep up with trends or the overall attitude to a certain entity – such as your company, product or your competitors’ counterparts.

Employ machines to look for keywords of your interest and report back what attitudes or relations they are carrying in real time. Let them combine opinions with geographical data and you can build a powerful response to whatever is happening on social media. Enhance your market research and take informed decisions on data that is available but hard to sift through. Monitoring social media can combine different machine learning techniques such as NLP, Sentiment Analysis, Image Recognition and Anomaly Detection for example.

In optimization of inventory and the supply chain, machine learning plays a vital part of improving models of stock levels and how logistics should be planned. From prediction of purchasing to logistic routes and cost, the supply chain ecosystem contains many parts with relevant data that enables deeper analysis and predictions.

The goal of finding insights in data or specific important signals that can be used for prediction is not only made possible by the sheer amount of data that is collected. Big Data, as it is often referred to, is not a guarantee for success unless the data quality is up to standards or a validation is made that the data is in fact right for the hypothesis at hand. Data cleaning and validation are two important steps in order to get the most value out of the oncoming analysis and is a good step in order to reach better data governance in the organisation.

Data that is locally stored often incurs a longer lead time and more burdensome processes in order to acquire it when it is needed. To encourage small projects and testing out different models, performing the work in a cloud environment can be the change that enables businesses to act in time and in a cost-efficient way. Whereas or not long-time storage of data in the cloud is a viable option, collaborating on building and running models and making them available to the right counterparts in a scalable way is a golden opportunity for cloud. Contact us for advice on how to explore possibilities in the cloud in regard to data, AI and ML.