Sigmoidal Success Story

20%

Reduction in model training time

12%

Rise in predictions quality

2x

Accelerated work of Data Scientist

What was the business problem?

Forecasts are about the future. It’s hard to overstate how important it is for a company to produce an accurate sales forecast. Privately-held companies gain confidence in their business when leaders are able to trust forecasts. For publicly-traded companies, accurate forecasts confer credibility in the market.

Sales forecasting adds value across an organization. Finance, for example, relies on forecasts to develop budgets for capacity plans and hiring. Production uses sales forecasts to plan their cycles. Forecasts help sales ops with territory and quota planning, supply chain with material purchases and production capacity, and sales strategy with channel and partner strategies.

Our client, a giant from the pharmaceutical industry, challenged us to create a forecasting and analysis system, to accurately predict their sales and financial performance over dozens of countries they operate in.

What savings and profits does the client achieve?

Sigmoidal facilitated the growth of this product from its inception to a market-ready status. We started with the event storming technique to visualize the business processes and adapted our infrastructure to maximize our team efforts further. To prioritize computational reproducibility, we reported analysis details of hyperparameter values, model architectures, and data-splitting procedures. For the project’s duration, we used sophisticated algorithms, massive data sources, and innovative tools. To show some efforts behind the curtain:

  • Sigmoidals’ team engineered a new code structure, allowing to shorten model training time by 20%
  • We utilized among other Arima, Prophet, and ETS family for Forecasting Time Series forecasting to create a robust system
  • We deployed the product on Amazon Sage Maker
  • Our Object-oriented programming to refine the organization of the software design around data.

How did we accomplish it?

Sigmoidal helped the client transform the project from the research stage onto the production cycle by rewriting it from R. into Python. Our proprietary Knowledge Sharing and Product-focus Method facilitated the rapid growth of the product and reached the market-ready phase much faster than anticipated.

Sigmoidals’ data scientists were able to increase the quality of the predictions by 10% and decrease the time needed to train the machine learning model. This allowed our client to roll out the product to the market much faster than anticipated.

Sigmoidals’ software improved clients’ processes and delivered many values, among them:

  • Reduction of sales pipeline and forecast risks
  • Better alignment of sales quotas and revenue expectations
  • Ability to focus a sales team on high-revenue, high-profit sales pipeline opportunities, resulting in improved win rates
  • Reduction of time spent planning territory coverage and setting quota assignments

In conclusion, our solution aided our client to closely monitor and react upon market shifts across many countries, which resulted in higher revenue and improved supply chains with material purchases.

How did we boost the project with Sigmoidal DNA?

Design Authority

Sigmoidal utylized architecture for reproducible research. We rapidly adopted machine-learning models to cope with the scale and complexity of data to prioritize computational reproducibility.

Product-focus Method

Since Sigmoidal was present from the inception of the project we facilitated Event Storming technique to visualize the business processes from the earliest phase. We set discovery sessions and concretis business domain information under which software is produced.

Deployment Strategy

To maximize the effectivness we employed versions management. This provided us with an extensive control over machine learning models, data sets and intermediate files, throught the project.

Knowledge-sharing Method

To maximize the effectivness we employed versions management. This provided us with an extensive control over machine learning models, data sets and intermediate files, throught the project.

What technical stack has been designed and implemented?