Frequently Asked Questions
Frequently Asked Questions
Below we answer some of your commonly-asked questions.
Below we answer some of your commonly-asked questions.
Compellon20|20 uses proprietary techniques to select the drivers and generate a custom predictive model. However the predictive model itself can be downloaded, examined, and evaluated against test splits of data, and incorporated in applications. More importantly, the resulting model output clearly explains why the predictions are being made. This heavily contrasts with black box models like Random Forest that spit out a probability without a clear explanation of the reasons they make a certain prediction.
Compellon20|20 generates custom model for each predictive model. There is no notion of fitting data to a preconceived hypothesis space represented by an existing algorithm. There are many benefits from this approach; a few include eliminating the need for analytical assumptions on the part of the user, incredible simplicity and speed in developing and iterating models, mixing different types of categorical and numeric data in any predictive model, and the ability to uniquely generate prescriptive advice and outputs.
Many predictive analytics tools target business analysts by wrapping graphical user interfaces around packaged algorithms attempting to simplify the experience, particularly versus programming methods in R or Python. However, wrapping a GUI around the complex, difficult predictive modeling workflow does little to reduce the fundamental challenges of inadequate expertise, lengthy and complex efforts, and the need for many hypotheses and assumptions around model algorithms and variables. Compellon20|20 uses the assumption-free methodology to deliver a completely different approach to obviate all of this complexity and deliver powerful, unique prescriptive insights.
Compellon is built around predictive and prescriptive analytics capabilities, as contrasted with Business Intelligence tools. The industry positions Business Intelligence as descriptive in nature, looking at what happened, through many different lenses and dimensions. Predictive analytics, on the other hand, utilizes complex mathematical models to find hidden relationships at more granular levels – down to individual customers and behaviors for example, which can be used to predict future outcomes.
Compellon’s core intellectual property is a unique approach to machine learning and machine discovery that delivers our “assumption-free” approach. The contrast is to the typical predictive modeling workflow demanding extensive effort, judgement, bias, expertise, and tooling effort across numerous tasks like feature engineering, feature selection, visual analysis, model selection, tuning, and maintenance. A sample of the associated assumptions include: which variables matter, often by visual inspection; how variables interact; which algorithm to select; how to transform data to get results that “make sense”; and even what data changes require re-training.
Compellon20|20 utilizes a dramatically different approach. Proprietary algorithms automatically learn the data’s hidden patterns and relationships, discover which matter, and how they combine to amplify weak individual predictive signals. There is no notion in Compellon20|20 of fitting to pre-constructed model types. Instead, the Compellon20|20 platform generates custom models for each data set, across non-transformed categorical, numeric, and ordinal data. Users benefit through the unique combination of simplicity and speed with powerful outputs that clearly explain why predictions are as they are and provide powerful prescriptive advice.
The “target” is the dependent variable or condition that is the goal of the predictive modeling. It can be nearly any logical expression for the categorical or numeric data in a column, or a combination of logical conditions across multiple columns.
Compellon has many customers equipped with other sophisticated analytical tools such as SAS and SPSS. Compellon20|20 complements their capabilities with unique prescriptive insights, automated driver discovery, quicker time to results, and extending the population of users in an organization who can apply advanced analytics. Data scientists can speed the feature selection process using Compellon’s machine discovery capabilities to sifting through thousands of columns and uncover drivers and subtle, non-obvious interactions. Others will choose to get results more quickly using the by the Compellon20|20 custom generated predictive model. The Compellon platform delivers evaluation metrics (confusion matrix and related derived statistics) and a validation function so data scientists can utilize typical techniques of splitting data into training and test sets and evaluating the results versus known models. Finally, many data scientists have more requests from their internal customers than they can practically accommodate, and Compellon20|20 helps their users become more self-sufficient on some projects or progress on exploring scenarios prior to engaging the scarce data science resources.
Compellon20|20 is fast, delivering results in seconds or minutes instead of days or weeks of effort as with traditional approaches. Compellon20|20 is easy, enabling advanced analytics for use by the line of business or analyst without specialized data science skills. Predictive models can be produced in as few as 3 clicks. And Compellon20|20 delivers targeted prescriptive insights that go far beyond predicting outcomes, to uncovering why and recommending the optimal actions to influence outcomes.
Compellon20|20 pricing is based on an annual subscription to the Software-as-a-Service. Service tiers are based on the amount of usage and data, the sophistication of the data analysis, and the number of users. Contact Compellon for a detailed quote.
Yes! Compellon20|20 Customer Analytics is simple to use for non-analytics experts. As long as you have data, and can frame the business goal with data, you can use Compellon20|20 to get actionable insight. Examples of framing the business goal include having a target in your data. For predicting customer loyalty, this framing equates to “What will drive respondents to rate me 9 or 10 on my Likelihood to Recommend question”? Or for Likelihood to Purchase, it could be “tell me what is driving people who purchased more than $200 per year”? As long as you have a data set where part of the population meets those indicators (such as a column with the Likelihood to Recommend question, or a column with annual purchase in the previous examples), the Compellon20|20 engine can use your data to build a predictive model.
Yes! Compellon has many customers that have analytics experts and are equipped with other sophisticated analytical tools such as SAS and SPSS. Compellon20|20 complements their capabilities with unique prescriptive insights, delivering results quickly, and extends the population of users within their organization equipped to apply advanced analytics. For example, some data scientists utilize the Compellon20|20 platform primarily to uncover the drivers and hidden relationships in the data even if they prefer to hand tool their predictive analytics algorithm instead of using the models generated by the Compellon20|20 platform. Because Compellon20|20 is unique versus all other offerings, we encourage you to experience a demo to explore how it might benefit your team.
Yes, Compellon has several customers using private instances. Contact Compellon for a solution that meets your needs.
Compellon20|20 allows research agencies to deliver more sophisticated insights much more quickly and with the same resources. Compellon20|20 can arm more researchers to utilize predictive analytics insights and go beyond dashboarding and correlations, instead of being limited to personnel with advanced data science training and expertise. This can help enables research agencies focused on qualitative research to expand the insights from their data with quantitative research.
Most other solutions use simple regression techniques that compare individual drivers to the loyalty target one at a time. Compellon20|20 detects the subtle interaction of drivers on the target, such that drivers that might ordinarily be ignored because they look weak individually can be found to be a strong predictor in conjunction with each other. Consider the example of likelihood to buy a luxury car. Regressing “income” or “interest” to uncover likelihood to buy may suggest either are weak candidates. Compellon20|20 automatically captures the combinations of the two that form a very strong combination prediction (high income AND high interest). Unlike this simplistic example, in most cases variables relate in non-obvious ways that would otherwise be missed. These combinations are even detected between data types such as numeric and categorical.
Compellon20|20 delivers unique and actionable predictive and prescriptive insights beyond the capabilities of any other customer feedback management platform, such as Compellon’s prescriptive advisor, predictive profiles, and individualized drivers when scoring whether new or existing customers will meet the desired target. Compellon is the next step in analytics for customer loyalty.
Absolutely! Today, Compellon20|20 can accept data in common formats (CSV or TSV) that can be exported from most survey platforms. In addition, a seamless data integration with the Qualtrics platform has been developed by directly connecting with their APIs (contact Compellon if you are interested in having an API connection to additional data platforms).
Compellon20|20 is built to run a scalable cluster architecture that can scale to thousands of columns and millions of rows. And one of the unique values of the Compellon20|20 architecture is the ability to sift through up to hundreds or thousands of columns and detect those few that are predictive of the target.
Yes! More and more companies are moving to combine data from across the customer journey, including survey data, usage data, and operational data. Compellon20|20 will find relationships across any type of data if they are predictive to the business goal. Data would be blended prior to uploading to Compellon20|20.
Compellon20|20 ingests data in “rectangular” format. Think of this like a spreadsheet of rows and columns. As long as one or many columns represents your dependent variable of interest (the “target”), Compellon20|20 will build a predictive model. Columns can be numeric, categorical, and ordinal (such as responses ranked from “Below Average” to “Above Average”), and can include nulls and missing values.
Compellon20|20 is for organizations striving to use their data to improve customer loyalty and optimize sales and marketing effort where moving beyond dashboards to actionable decisions is demanded. Common use cases include understanding the causes of likelihood to recommend, likelihood to purchase, overall satisfaction, perception of value, prioritizing sales leads, win/loss analysis, segmenting users and building personas, targeting personalized offers, and diagnosing churn.
Compellon’s simple and easy user experience uniquely enables those closest to the data and the business, such as the line-of-business expert, data analyst, insight professional, and product manager, in addition to those with deep skills in analytics, to explore and generate these insights.
Compellon20|20 software-as-a-service is a predictive analytics solution that provides an unprecedented user experience allowing Insights Professionals to quickly and easily achieve impact by explaining the drivers of customer loyalty, delivering specific recommendations on actions to take to most efficiently achieve the desired goal, and personalizing actions for individual customers or populations.
Compellon is based in Laguna Hills in Southern California.
23461 South Pointe Drive
Laguna Hills, CA 92653
Compellon is an innovator in knowledge discovery and predictive analytics solutions. We pioneer a unique approach to machine learning and discovery to revolutionize the speed and ease that organizations with or without data science skills can predict outcomes, uncover the reasons behind outcomes, and identify (automatically) the optimal actions to achieve their goals. The focus of the company is the Compellon20|20 platform which is incorporated into software-as-a-service solutions both from Compellon and from partners utilizing the platform.