How DecisionFacts makes it easier for businesses to consume outcomes of data science projects

Background

As business uncertainties are changing rapidly, organizations can no longer rely on historical data to make forward-looking decisions. For example, inflation, weather patterns, labor shortage in docks, supply chain disruptions, or social media sentiment have a direct impact on predictive models’ outputs for decisions. Typical analytical systems provide a good representation of past data through descriptive or pre-computed models in a dashboard. While these may be sufficient for steady-state operation decisions that are made only a few times a year, businesses are now challenged with high variability in data and organization’s state that has condensed the decision cycles from quarterly to monthly to weekly. Accordingly, Data scientists have to continuously rerun the models with different conditions from business users and manually track data, business logic, and outputs.

At the same time, business leaders, using data science algorithms for decision-making, often strive for the following:

• Scenario Analyzer: How to utilize a probabilistic algorithm for decision-making by evaluating a range of scenarios on different conditions that a decision-maker can independently define and control. Currently, business users are constrained by spreadsheets, the computing resources of their laptops, or their high dependence on technical experts to run complex models and analyze outcomes.
• Governance: How to build trust with algorithm outputs, including capabilities to track what input, data, or parameters led to an outcome for decisions. Today, teams use disconnected applications to deliver results and most of the systems don’t maintain the history and context that is essential to understanding what led to these results.
• Productivity: How to bring the domain experts’ knowledge and technical teams’ solution closer by maintaining a structure and avoiding a lot of back and forth among teams.

How DecisionFacts bridges the Data and Decisions Gap

DecisionFacts provides a rich set of features for business units to integrate data sources, develop models, run scenarios, and track the context behind decisions. The foundation of the product relies on the following 3 pillars:

• Exploration: Unlike most analytical systems that are ‘unidirectional’ with processed data presented to business users, DecisionFacts enables ‘bidirectional’ analysis, where business users run new scenarios on the cloud with a new set of data or parameters/conditions. The platform provides two layers of abstraction — one at the cloud infrastructure level and the second at the algorithm interface level. Self-service cloud computation allows users to evaluate new paths for data-driven decisions they didn’t consider before because of time or resource constraints.

Figure 1: DecisionFacts Project, abstracts cloud and algorithms for running scenarios

• Automation: DecisionFacts decouples parameters, algorithms, and data so that each can be independently changed to see the impact on the outcome. Non-technical, domain experts can change the parameters that drive the algorithms, enabling them to run multiple scenarios with different business logic without modifying any code. All the scenarios are automatically tracked with the full context of input data, parameters, and algorithms for corresponding outputs. This gives domain experts more confidence in the output, speeding up business validation of these model outputs to enable faster decision-making.

Figure 2: Scenario Outputs based on model predictions from different ‘Business levers’

• Collaboration: The domain experts and data experts collaborate at each step, improving the quality of the algorithm and shortening the analysis cycle for decision-making. Further, the system allows overriding outputs and maintaining versions of all human interventions.

Figure 3: Scenario tracking of all runs with contextual collaboration

Outcome for Business

The following are the benefits that users of DecisionFacts customers are highlighting:

• On-demand Real-Time compute — No more constraints on shared resources of server or local compute for complex business analysis. Business users run self-service analyses on the cloud without needing help from technical resources.
• Parallel Execution — Since the parameters and code are decoupled, business users run parallel simulations, which is not possible with Notebook. Further, as the system tracks the input/output of each scenario, teams have seen a productivity gain of over 100s of hours.
• Security — The integrated setup is designed with additional layers of security such as secured access to data sources via VPN, encryption at rest for data stored in Google Bucket, and more. Users avoid writing scripts, and configuration files, thereby expediting the acceptance of data initiatives by stakeholders.