Dynamic Demand Forecasting in Alteryx with InstantML

The imperative for a new supply chain model

Balancing the supply and demand is top of mind for every company having to deal with a supply chain.

The current market conditions have created an immediate market opportunity, specifically within the supply chain function to deliver low-code, automated, integrated solutions that cannot be delivered by existing supply chain software players.

Structural changes such as Covid-19, complexity of the supply chain, new products and promotions make forecasting a difficult task.

The top notch Alteryx APA platform and the Tangent Works InstantML augmented, hyper automated machine learning solution combined with the Demand Data business knowledge deliver a solution for Dynamic Demand Forecasting that changes the paradigm of forecasting completely.

We called it “Dynamic” because of the use of Tangent Works InstantML times series data ML technology, that makes dealing with structurally changing data easy.

This is a new and easy way to deal with demand forecasting for companies dealing with a supply chain in retail, CPG or manufacturing.

The 7 Pain Points and Supply Chain Maturity

1. Too many processes have no automation

Research by RIS news tech investment shows there is lots of room for improvement in areas such as data interchange data, analysis, forecasting, back office processes, visibility and inventory management. Many retail manufacturing and logistics professionals ranked the degree of their supply chain automation low.

Leading companies are doing much more than just collecting and storing data in large quantities. They carefully architect their competitive strategies around data driven insights to drive business excellence. And they rely on sophisticated quantitative and statistical analysis and predictive modelling provided by powerful information technology.

2. Lack of Software integration

Although spreadsheet technologies were introduced in the early 70s, they are still the predominant tool for business users do their daily work. However, they give the business user the false sense of automation. Often issues such as software integration, downtime, frequent manual interventions are mentioned as preventing smart decision making.

The lack of traditional analytics and predictive analytics to make better decisions and extract maximum value from your business process is often the result of applications (or the lack off) that do not cover the scope of what you need.

3. Failing to deal with the current state demand forecasting and stock replenishment

The rise of multi-channel retailing requires inventory positions in more locations then before. Dynamic demand forecasting over many SKUs and locations and also per customer requires a different way of dealing with the process. To keep up with these demands, improving the planners productivity is essential. This requires automation of the forecasting process. With the pressure on margins, automation is vital. To make this a successful unified analytics data science and process automation is a necessity. These technologies are often insufficiently leveraged.

4. Lack of maturity of predictive analytics

The industry agrees: data is the new gold. Most companies still fail do use it as a potent competitive tool. The use of basic “after the facts” analytics is still well below what you would expect given the state of technology. For predictive analytics the maturity is disappointing. Companies are failing to get value out of predictive analytics.

5. Difficulty to move beyond the experimenting phase

Organizations struggle to integrate AI solutions, fuelling demand forecasting, with existing production applications, wasting time and money on data science projects that are never put in production. Projects stay in an experimenting phase not delivering business value.

Projects are often led by data scientists and IT specialists who do not speak the business language; while business user requires tools that are robust, agile, automated and easy to use.

6. No resilience against structural change

COVID-19 has had a significant effect on a wide range of consumer and business behaviours, resulting in data drift. This data drift means that many companies are struggling with the new normal. This also impacts machine learning (ML) models. They are decaying more quickly than before, even to the degree that they may have to be switched off and shelved for later use. Many Machine Learning technologies do not deal with the structural changes such as COVID-19 in an adequate way. In these situations Machine Learning technologies do not live up the promise of providing the much needed prediction for the business.

7. Far too complicated

Predictive analytics is fuelled by AI/ML technologies. This requires Data to be prepared (DATA), models to be constructed, features to be engineered (ML), models to be developed in the IT Landscape (DEV) and finally the operational processes to keep it running (OPS). This is often called DATA-ML-DEV-OPS.

Although this resonates with specialists it is far away from the world of the business users. The business user wants Low code, hyper automation, augmentation and human readable models also known as explainable AI.

Why this is important for you?

According to the RIS News Tech Investment Survey (Post Covid) a staggering 29% of the IT Budget goes to support of supply chain capabilities, a total of $8.8B in 2020.

More than 15% this will go to business analytics.

Top investment go in:

Advanced Analytics
AI/ machine Learning
Supplier portal
Replenishment
Transportation/Logistic Management
Sourcing optimization

A flexible solution for the problem: Alteryx DDF, Dynamic Demand Forecasting solution

Alteryx, Tangent Works and Demand Data deliver an End-To-End solution for Dynamic Demand Forecasting.

With Alteryx Analytic Process Automation (APA) you can achieve Supply Chain Automation across eco-systems through real time cross data sharing and automated analytics without forcing your organization to adhere to specific standards and technologies.
Tangent Works InstantML seamlessly integrates in the Alteryx environment and provides hyper automated, augmented machine learning for time series data that is extremely fast easy to use accurate and delivers explainable models. The speed, automation, ease of use and accuracy make it ideally suitable for Dynamic Demand Forecasting.
Demand Data is a super specialized consultancy for the supply chain markets it provides consultancy, out-of-the-box reports and data streams to make the prediction process easy.

The dynamic demand forecasting solution his supported by a worldwide ecosystem of Alteryx and tangent works partners.

Tangent Works TIM InstantML on Alteryx

Alteryx + Tangent Works bring a unique combination of technologies that enable those who work with time series data to easily benefit from machine learning for predictive and prescriptive scenarios.

The Tangent Works TIM Solution is an easy to use, fast and accurate solution for Time Series model generation, prediction/forecasting and anomaly detection in an Explainable AI way.

The Alteryx Designer Solution fully integrates Tangent Works TIM. You benefit from the world famous Alteryx functionalities that allow business users, Citizen Data Scientist and Data Scientist to do magic with data.

Tangent Works vision is to make machine learning easy to use and available to any users without the complexity of traditional AUTOML model selection, complex, times consuming data science or tedious manual feature engineering.

Essential to Citizen Data Science is augmented machine learning/analytics with automatic feature engineering, model generation and deployment whilst offering understandable models. Time series data are everywhere in many different industries. The Alteryx Vision is to create solutions that allow users to deliver on the “thrill of solving”. This is why we are so excited to make TIM available within Alteryx an working with this amazing Alteryx Community.

Dealing with time series data machine learning is hard, but has enormous potential to gain deeper insights and deliver faster decisions. TIM allows you delivery of business value rather than being stuck on the machine learning technology.

Dynamics Demand Forecasting in Alteryx with InstantML

 TIM and Alteryx Designer and Gallery

TIM Forecasting and TIM Anomaly Detection tools are available for Alteryx Designer. The TIM integration is compliant to the Alteryx rules for integration. TIM can be accessed from the Alteryx gallery. This allow users streamlines the process by delivering a repeatable workflow for self-service data analytics, leading to deeper insights in hours, not weeks using designer. Alteryx Designer empowers data analysts by combining data preparation, data blending, and analytics but now also advanced time series machine learning capabilities.

Advanced InstantML, RTInstantML, The Tangent Works Secret Sauce

Many companies offer AutoML capabilities to support the model generation and feature engineering process, allowing businesses to build, tune, compare, as well as test multiple models and deploy the selected model as an application programming interface (API).

Although AutoML offers a step forward, it still remains time-consuming and expertise-intensive. Tangent Works surpasses AutoML with its InstantML technology offered through Tangent Information Modeler (TIM), which generates models in no time delivering real Augmented Machine Learning.

TIM creates models in seconds based on target and predictor series with just one pass through the data. These models are made available through a single API, thus saving organisations the hassle of creating separate APIs for each of the models. Tangent Works has further enhanced TIM’s capabilities with real-time InstantML (RTInstantML) technology that provides organisations with optimally trained models for the data available at any time with immediate forecasting or anomaly detection.

RTInstantML not only simplifies the model selection and automates feature engineering but also tremendously shortens the lengthy process of developing the right model at the right moment, which is crucial for the time series problems. The dynamics in these problems change with every new data collected, and the new data availability can also change from one minute to another. TIM’s ability to generate a new model on the spot without any specific settings enables immediate forecasting and anomaly detection.

Current strategies vs TIM

Years of research have led to the creation of TIM engine, based on a unique blend of various advanced technologies including information criteria/ geometry—an interdisciplinary field that uses differential geometry techniques to study probability theory and statistics.

Automatic model generation for time-series goes hand in hand with anomaly detection. TIM generates high quality models over the in-sample training data and learns a normal behaviour. TIM then checks for any significant departures from this normality to find samples that are anomalous with respect to historical behavior. This process is intuitive as TIM models are transparent, and thus, describe the key drivers of the normal behavior. The degree of normality is provided as a quantity called TIM anomaly indicator. This approach is useful for both univariate and multivariate problems.

With the “consume” strategy, Tangent Works want to make it easy for TIM to be used by a variety of tools and platforms. TIM Clients are available not only for Alteryx but also for QLIK, Tableau, Denodo, Cloudera, and Microsoft or Microsoft Azure products, including Azure Data Lake, Azure Databricks, Azure SQL Data Warehouse, Microsoft SQL Server, Microsoft Excel, and Microsoft Power BI.

Summary – How does this solve the 7 Pain Points?

1. Too many processes have no automation

The Alteryx APA platform is a state of the art low code unified analytics data science and process automation that benefits from the Tangent Works InstantML Time Series Machine learning technology. This brings hyper automation and augmented machine learning to the next level.

2. Lack of Software integration

The Alteryx platform is a designed as a data and software integration environment. Alteryx lives up its vision “The Thrill of Solving” and makes it easy to integrate any data and application.

3. Failing to deal with the current state demand forecasting and stock replenishment

Dynamic Demand Forecasting is set up by experts in the supply chain market. Demand Data and Alteryx deliver solid market knowledge that allows for a fast time to market with Dynamic Demand Forecasting.

4. Lack of maturity of predictive analytics

InstantML delivers advanced machine learning that outperforms the standard Alteryx embedded ARIMA models that are really extremely easy to use from within Alteryx. You beef up your maturity level immediately without having to worry about complex task such as feature engineering, model building, model tuning, model selection, Model exposure, etc.

InstantML uses advanced the advanced mathematics of information geometry and criteria to make this a fully augmented, hyper automated process.

5. Difficulty to move beyond the experimenting phase

Alteryx APA and InstantML gives the power to the Business user. This low code, hyper automate also you to solve business problems fast. There is no need to get stuck in experimenting. This is the spirit of the Alteryx Grand Prix competition where people compete to get real business problems solved in minutes.

6. No resilience against structural change

InstantML shines in environments with many models and lots of structural change. InstantML gets out of the data what is in it. It can rebuild the features and models in seconds whilst maintaining the explain-ability of the model. This is revolutionary and is why we call it ML 3.0.

7. Far too complicated

The Dynamic Demand Forecasting has been developed with ease of use in mind. You can deploy it from the Microsoft Azure Market place in seconds as SaaS solution. It can also be deployed on your own on-premise or cloud environment (Aws, Azure) and comes with a adoption pack that allow a fast track deployment. This is AI/ML the easy way.