Twitter Facebook LinkedIn
Email [email protected]
Phone +1 713.457.7474


Case Study: AWS Machine Learning


The Challenge:

A Fortune 500 company specializing in the trade of electronic components is utilizing a custom-built CRM system to manage its trade processes. With tens of thousands of new requirements they receive every week, the buyers have little information regarding which requirements that they should work on, causing opportunity losses and potential revenue drain.

  • Many requirements are forgotten due to volume of requests or shortages in the market.
  • A solution to prioritize the requirements existed, but it didn’t account for many important factors such as number of active vendors, cross-product sales, current product trends, and quantity requested, rendering it unrealistic, inaccurate, hence often being ignored.
  • The client’s data science team had tried to improve the scoring solution and built out on-prem sandboxes, but the servers were limited in their ability to scale with more memory-intensive algorithms and data scientists were confined to using a small set of data.

PREDICTif Solutions was brought in to leverage machine learning to help improve the requirement scoring system, increasing gross profit. The project was dubbed “Requirements Prioritization v2”.

The Solution:

Working alongside the client’s data science and IT teams, our solution architects executed a 3-month plan to develop an improved requirement scoring model and integrate it with CRM to provide a more precise prioritization for the buyers. We recommended using a serverless AWS machine learning stack that includes SageMaker as well as API Gateway, Lambda and SNS for notifications. The below picture illustrates a high-level architecture design.


SageMaker is an all-in-one machine learning environment. With it, we were able to provide a sandbox to code, train, and test various models utilizing Jupyter Notebooks and then deploy trained models as API endpoints that CRM integrated with. SageMaker was selected to simplify the process of productionizing ML workloads. It provides a dockerized environment and powerful API’s to deploy trained models as a microservices architecture which can then be accessed via other AWS services. To reduce time, we also leveraged the hundreds of AWS-provided algorithms to develop our new requirement scoring algorithm.

AWS offers a very rich set of API services, which have made it very easy to integrate with CRM. Every service in AWS exposes an API. SageMaker has an intuitive UI exposed through the AWS Console, but it was the management API that really excited us. Without having to incorporate any third-party tools, we were able to start/stop, schedule, and promote training jobs in a continuous integration fashion to automatically roll out newer models. 

Challenges Addressed:

  • Model Retraining – while the process of building a new model is relatively simple, model retraining for comparing performance, approving changes, and testing with live data poses several challenges. We incorporated a SageMaker feature called production variants to simplify the process. A variant allows the user to deploy multiple models to the same endpoint and declare what percentage of the traffic will go to each model. This way, a new model can be tested until confidence is high enough to remove the old one. 
  • Parameter Tuning – One of the most time-consuming phases of developing ML projects is hyper-parameter tuning: the art of tweaking the configuration parameters that control model training. SageMaker hyper-parameter tuning jobs helped by allowing us to choose a performance metric to maximize. After each job was finished, 20 independent training jobs had been run, each using the output of the last to enrich and further optimize our performance metric. The SageMaker Console provided an easy-to-use comparison tool, enabling us to quickly identify the right hyper-parameter combinations. 
  • Cost Management – After determining the target algorithm would be an AWS provided XGBoost algorithm, the DS team wanted to try using 1-hot encoding, wherein every possible product was converted to a feature. This made the final feature set very wide (~700,000) columns, which was much better for CPU performance but needed more memory for each training iteration and endpoint call. A Cloudwatch event was created on a schedule to start/stop notebooks and endpoints during non-work hours, thereby cutting development and testing costs by 60%. 
  • Security – Since a fully-deployed ML cloud solution was brand new at this client, there was skepticism and concern around utilizing cloud, particularly regarding security. AWS shared responsibility model simplifies the project teams task list o API Keys were used for communication between on-prem and API gateway 
    • S3 transactions are all SSL encrypted by default, and we enabled encryption at rest in the S3 bucket. 
    • All information transferred between services within an AWS account is secure and monitored. 

The Results: 

PREDICTif Solutions has been innovating exciting solutions for our clients for over a decade now, so it was satisfying in using some of most cutting-edge technology that AWS offers to breath new intelligence into an older, deterministic CRM system for this client, that has resulted in an increase of profit margin by over 30%, after just the first phase of this project. 

  • Productivity Increase – the productivity of the data science team has increased by over 200% with the AWS machine learning technology. This project would have taken at least 6 months to deliver if we had done it using on-prem sandboxes 
  • Development Cost Reduction – A TCO analysis revealed a solution of this size would have cost 3x to host on prem and taken many months to procure. By using AWS, the DS team began building models from day 1. A notebook instance was created in minutes, rather than weeks of waiting for servers to be racked, or clusters to be configured. With SageMaker’s 1-click deploy methodology, architects could integrate developed models much faster because we weren’t waiting on DS team to finalize the feature set 
  • Solution Portability – The ML model exists in a docker container, which can easily be ported to another cloud provider or even on prem if the client wanted 

The client’s executive leadership wanted to pick something with a low up-front cost, and minimal impact to business process until the value could be proved to in front of a larger audience. We were able to deliver on this mandate and provide an impactful solution to move the business at the speed that the market demands. 


PREDICTif successfully implements Hyperion Planning and Budgeting Cloud Service (PBCS)

PREDICTif successfully implements Hyperion Planning and Budgeting Cloud Service (PBCS) to replace an oil and gas company’s manual Excel based forecasting system.


The Customer was faced with multiple challenges by manual consolidation processes and data entry to prepare their Budget and Forecast financial plans. The characteristic of the old reporting system includes:

  • Manually linked financial statement spreadsheets
  • A system of off-line data repository with actuals from GL and planning data in a web of Excel worksheets
  • Restrictive data access to safeguard the repository – a.k.a. aforementioned worksheets
  • Link breakage between workbooks and worksheets which caused inaccurate forecasting

These issues made the existing solution severely limiting in terms of the ability to predict its revenue and costs for each of its business units. The Customer decided to replace its manual, error-prone, labor intensive financial reporting and planning solution with a more automated, streamlined and cloud-based one.


The project is to build a PBCS solution to better service the Financial Planning & Analysis (FP&A) team, providing an updated and fully supported financial system for FP&A use. The solution encompasses:

  • Replacing the current Excel solution with PBCS
  • Creating an automated data integration application (using FDMEE) to allow for loading of actuals data
  • Allowing for analysis of further capital expenditure ventures by overlaying with the existing plan
  • Create a Budget and Forecast application to allow for Budgeting and Forecasting at the desired level
  • Customized training program to increase technology adoption
  • Streamlining data submission by providing easy to use form templates to end users
  • Automatic upgrades and scheduled maintenance through PBCS cloud environment

The solution provides a scenario based Planning and Forecasting option while also allowing for a single reporting repository to allow for easily retrievable and reliable data, providing one version of the truth.


PREDICTif adopts a phased approach that minimizes the impact of change in an effort by breaking the implementation into two phases by its business units: (1) Oil Exploration and (2) Production and Gas Pipeline. This allows concurrent efforts of phase 1 user acceptance testing and phase 2 solution development, meeting Company’s tight deadline.

  • Builds and deploys the desired hierarchy in PBCS for initial use by the exploration and production team.
  • Uses the hierarchies built as the basis for building out business rules, calculations, and Hyperion webforms.
  • Implements a connection to the general ledger system using Financial Data Management Enterprise Edition (FDMEE) and an SQL query to produce a flat file to source data into the application.
  • Builds and deploys the second portion of the hierarchies for gas transportation.
  • Creates forms and calculations.

After user acceptance testing, both pieces were migrated into Production on-time and on-budget.

Additionally, PREDICTif provides customized SmartView and PBCS training as the team begins the preparations for User Acceptance Testing. As the go-live date approaches, enhanced training is given to the administrator level users to prepare them for maintaining the solution going forward.


PREDICTif successfully replaced the current outdated Excel based Planning and Budgeting model with the updated PBCS solution. The Customer has already received significant return on its investment:

  • New Hyperion PBCS based application suite with full support form Oracle
  • Retirement of the manual spread sheet planning process
  • Dynamic reporting solution allowing for bursting, batching, and grouping reports for easy use and run time
  • Reduced time for data load and hierarchy maintenance using Hyperion Planning Admin Extension for Smartview
  • Data form submission templates to organize and simplify data submission for end users
  • Training gives the user a broad knowledge of the Hyperion technologies as well as a focused, more practical knowledge that they can take back to their desk and use within days.
  • Segregating the trainings into a user group and an administrator group not only allows for the User Acceptance Testing to be performed in a more timely and efficient manner, but it also allows for users to effectively manage their time by not having to take trainings that will not apply to their skill set.

The success has laid a foundation for future phases that will further automate the existing financial planning and reporting process by leveraging other PBCS functionality such as EPM Automate and cloud to cloud data integration.

PREDICTif Solutions to Present at Oracle OpenWorld 2015




PREDICTif Solutions, a Houston-based technology consulting firm, today announces its upcoming presentation at Oracle OpenWorld 2015. The conference, held from October 25 to 29 at the Moscone Center in San Francisco, CA, will be a premier event for business and IT professionals to learn about Oracle technologies through the interactions with Oracle and its business partners. PREDICTif’s CEO, Jeff Huang, VP of Global Sales, Karl Harrocks and OEID Practice Lead, Jimmy Philip will represent PREDICTif at the conference, meeting customers and Oracle colleagues.“This year we are very excited to have two of our customers, Wilsonart and Cox Automotive, speak on our behalf and share with our customers and partners their successes with PREDICTif. The Wilsonart story is featured on the September issue of Oracle Profit magazine, describing the incredible journey that Wilsonart and PREDICTif travelled together to build a fully integrated EPM, BI and Big Data solution” said Huang.

“PREDICTif has built a lot of successes in our three core practices, EPM, BI and Information Discovery/Big Data last couple of years.” , said Harrocks, “We are excited to leverage this platform to share our extensive experience in Oracle cloud offerings and Big Data Discovery. Whether we’re sharing our knowledge or learning from others, it’s our continuous goal to grow as a thought leader in the age of Big Data and bring this advantage to our customers.”

High-Value Health Check to Enhance Existing BI Solutions


PREDICTif Delivered a High-Value Health Check to Enhance Existing BI Solutions and Develop a Master BI Implementation Plan


A leading Oil and Gas Equipment services company.


  • Remedied performance and instability issues existed with the BI applications
  • Delivered a master Implementation roadmap for an enterprise BI and FPM architecture


The customer has over 25 subsidiaries, many acquired through acquisition. Each subsidiary had built out its own business intelligence (BI) and finance performance management (FPM) solutions using a myriad of BI and FPM products. As one of the leading global equipment service and pipeline construction companies, the customer faced numerous challenges related to instability, inconsistency and slow performance of BI applications. In preparation for the opportunities and demands of becoming part of a publicly held company, the customer needed to expand its BI and FPM architecture beyond the business unit level.  Before further expansion could take place, the customer needed to consolidate their division level BI and FPM initiatives as well as map out a strategy and roadmap to establish an enterprise BI architecture. The existing BI systems presented numerous deficiencies in terms of people, process and infrastructure:

  • All its business units had their own BI and FPM technologies, support teams and project initiatives without an enterprise level governance and standard;
  • There was no single version of truth in terms of data on the enterprise level which rendered the enterprise level reporting and financial consolidation inaccurate and labor intensive;
  • The existing BI infrastructure contained a myriad of technologies and products making support and maintenance difficult and cost of ownership extremely high; and
  • Lack of a BI competency center rendered uneven quality of delivered BI applications that resulted in slow performance and instability.


Conducted a health check to identify areas of risks and opportunities, recommend remedial measures, and produce a master implementation plan and a roadmap to build an enterprise level BI architecture.

  • Examined existing BI infrastructure and organizations in terms of people, process and infrastructure to discover areas of risks and opportunities, and more importantly, produce a list of actionable recommendations to improve the overall BI strategy and execution;
  • Reviewed the existing architecture, configurations and codes;
  • Identified bottlenecks that were causing slow performance and instability;
  • Aligned business objectives and priorities with BI initiatives;
  • Created organizational philosophy and culture to support an enterprise BI strategy; and
  • Developed a master implementation plan and roadmap for an enterprise level BI architecture.


PREDICTif Solutions was engaged to provide a 2-week health check. PREDICTif’s senior architects reviewed the existing BI infrastructure, made recommendations for improvement and delivered a master implementation plan for rolling out BI to the entire enterprise.

Based on PREDICTif’s standard health check agenda and customer priorities, PREDICTif and the customer created a detailed agenda before the engagement. The agenda addressed both tactical and strategic aspects of the customers business. The tactical aspects were to analyze the existing BI solutions to identify areas of risk and opportunity as well as provide recommendations on both. The strategic aspect was to understand the customer’s business priorities and deliver an enterprise BI implementation plan.

The two week agenda was filled with workshops, interviews and document reviews during which, PREDICTif’s architects met customer’s personnel from numerous areas of the business to discuss business objectives, current BI architecture and business user concerns. Conversations focused on areas such as data quality, data governance, data modeling, report requirement gathering processes, architecture, support/maintenance and technology best practices. The architects also discussed future plans, business initiatives, short term and long term business goals as well as technological direction in the BI and FPM arena.

The PREDICTif resources combined our own set of BI and FPM best practices and standards with information gathered from the discussions in order to produce a  set of documents:

  • The tactical health check deliverable is in the form of a Microsoft Word document and a Microsoft PowerPoint presentation both of which outlined areas of opportunity, improvement and identified risks as well as the necessary steps to mitigate those risks.
    • In this case, the PREDICTif architects discovered that the BI architecture was not designed appropriately for the increased number of users and business cases which caused significant degradation of performance and stability;
    • The customer did not have a well defined data governance process that was accepted by all business units. It resulted in poor data and metadata modeling as well as an overall lack of data quality and transparency;
    • The architects made recommendations to improve the Bi architecture in terms of performance and stability; and
    • The architects applied PREDICTif developed and tailor-made data governance and data modeling best practices to the customer’s business and IT environment.
  • A master BI implementation plan and roadmap was delivered in order to consolidate the dispersed BI architecture as well as establish an enterprise level BI infrastructure.
    • The implementation plan entailed a multi-phased project plan that detailed the sequence of activities, dependencies and high-level estimate of effort as well as timelines;
    • It illustrated a 18-month BI architecture evolution through the phases of implementation;
    • The roadmap also considered the trend and best-in-class BI technologies; and
    • The implementation plan included the steps to establish an enterprise level BI competency center and governance process.


PREDICTif delivered considerable value to the customer’s return on the software and IT infrastructure investment. The health check’s findings and recommendations significantly improved both the performance and the stability of the customer’s existing BI architecture. The implementation plan provided a clear roadmap to establish an enterprise level, world class BI architecture.

  • The improvement decreased the rendering time of a complex report from an average of 6 min to less than 30 seconds, greatly increasing the productivity;
  • The upgrade and addition of hardware and software instance increased the overall uptime from 95% to that of 99% which was a requirement of the business;
  • The customer completed all three phases of PREDICTif’s recommended implementation plan and the overall BI initiatives continue to gain adoption and approval from the business use population.

Business Intelligence (BI)


Business intelligence (BI), once a competitive differentiator, is now a commodity. Most companies have implemented BI solutions that provide historical reporting, dashboarding, metrics and scorecarding for past events. Companies know what has happened but the ability to know what will happen will be the competitive advantage that companies need to excel in this volatile and ultra-competitive environment. Predictability is the next step in the evolutionary process of Business Intelligence.

Traditional Business Intelligence and data warehousing focus on strategic, long term decision support. While strategic Business Intelligence continues to be a requirement to support long range vision, Predictive Business Intelligence (PBI) takes business Intelligence beyond a process that has traditionally looked backwards and has been reactive in nature. PBI empowers the enterprise in realizing competitive advantages and provides the business with the necessary agility to meet the challenges of today’s rapidly changing business environment by mitigating risks and maximizing opportunities. PBI greatly improves both long term strategic decision making and near team operational decisions.

The ability to make Predictive strategic decisions will separate enabled companies from their competition enabling them to capitalize on opportunities and reduce exposure to risk. Statistical analysis on operational and transactional data will provide insightful information on business trends and enable the business to make strategic decisions quickly and more effectively. For example, when a retail chain is determining whether to establish a presence in an unfamiliar territory, it could utilize growth data from other locations and combine it with the local data as well as current projection data to provide support for the decision making. The data might be sketchy and sparse, but statistical analysis will offer a sound basis for decision making. Other examples can be found in oil & gas exploration or pharmaceutical development projects for instance. These projects often entail long development cycles and considerable up-front cost the outcome of which has significant impact on the overall performance of the business. Predictive decision making will enable those companies to analyze more data in order to gain a complete view of the business cases and leverage proven statistical models to aid those impactful decisions. The completeness and quality of the data analyzed are critical to determine the accuracy of the prediction. Although companies in these verticals go to great lengths to develop elaborate risk management models to address common concerns, predictive business intelligence In support of existing risk management processes, provides a richer set of data and more interactive analysis to ensure a better outcome.


email [email protected]
phone +1 713.457.7474
Twitter Facebook LinkedIn
© PREDICTif Solutions. All rights reserved.