Our Data Science team functions like the rest of our company: fast, accurate, and on demand. The clients we work for are often under pressure. When they ask us a question, they need an answer as quickly as possible. Despite the need for speed, we are careful to back Veracity’s answers with quality products containing rigorous analysis and detailed visualizations.
Understanding the Question
We begin by validating the client’s question. Sometimes the way a question is posed does not actually get to the heart of what the client is trying to understand. As Data Scientists, we understand how best to approach a data analysis problem and help the customer frame a question correctly. In other words, we help the customer ask the right questions to arrive at the most useful answers.
Flexible Forecasting Models
Veracity designs forecasting models to be flexible. Businesses and organizations often change their practices and hence their business rules. Our clients must have the freedom to adjust their rules in the model to see the outcomes of various decisions. For example, in one of our inventory forecasting models we use statistically determined rates. When the relevant business rules change, we are prepared to update these rates based on changes to real-world business rules.
Example Model 1: Unscheduled Maintenance
Part of inventory forecasting is understanding how the duration of maintenance affects the inventory. In this example, our model estimates the duration of maintenance events performed at two different repair sites, each with different completion rates.
Our Data Science team chose to model event completion based on a set of assumptions that related the rates to parameters of a hyper-exponential function found by fitting to the observed cumulative distribution function of event lengths (the probability an event will last less than a given time). As an example, we fit notional data with this function in the graph below. The figure shows the observed cumulative distribution function for event duration (stepped line) and model distribution (smooth line) for six different types of inventory (T1-T6). The blue and green colors represent different repair sites.
This simple model provides us with an intuitive understanding of modeling event durations. Thus, when a client decides to change the number of repair sites or the type of events, we can easily adjust the model and demonstrate to the client how the change will affect their inventory.
Example Model 2: Inventory Attrition
To forecast how many items a client can expect to have in future inventory, we first need to estimate how many will be removed due to failure. We can then estimate how many items must be kept in reserve over time to ensure sufficient inventory for future demands. Our team created a mathematical model describing the expected failure behavior using Weibull distribution, which accounts for a probability that changes as inventory is used. The graphic below represents a Weibull and exponential (a special case of the Weibull model) approach to modeling an example data set.
At Veracity use content-rich visualization to support our analysis. The graphics we generate capture key ideas, relationships, and changes, bringing a fresh and often eye-opening perspective to the problem being solved. We accommodate your preferences for using visualization, whether through Tableau, html, or state-of-the art modeling tools.
Models and visualization combine to answer the client’s initial question, but if we’ve done our job right, that first answer is merely the stepping-off point to successful, ongoing decision support. We continue to advise our clients with Decision Support, answering new questions as they arise. These are questions the client could not have even thought to ask before seeing the initial analysis, and it is through these ongoing, interesting queries that we enable our clients to become ever more adept at high-stakes decision making.