Actionable, accessible insights

The World Health Organization estimates that 1 in 10 medical products circulating in low- and middle-income countries is either substandard or falsified. Veripad, a NYC-based startup, has developed a mobile app that scans and analyzes paper card chemical tests to rapidly verify 60 common medications.

Slalom partnered with Veripad on a proof of concept to use machine learning to improve the accuracy of its analysis. We used SageMaker to build a machine learning solution, using regression and clustering techniques. This helped us quickly set up the ML work environment, leveraging S3 for data storage.

In just five weeks, we helped Veripad significantly improve its fraudulent medication detection system to an accuracy above 90%, which is higher than human experts. This cost-effective solution is far faster than a typical lab test, which can take a week or more.

Amid major shifts in the energy industry, PG&E is driving continuous improvement and process efficiency through innovation. The company has developed cloud-first analytic products on its data and analytics platform, built on AWS. 

New AWS machine learning capabilities, including SageMaker, presented the opportunity to dramatically improve PG&E’s Corrective Action Program (CAP), which enables employees to identify and track equipment and safety issues, and provide suggestions on how to do things better. These geotagged reports include unstructured comments, details, and photos. Diagnosing, prioritizing, and addressing issues was a highly manual and time-consuming process—often requiring several rounds before successful resolution. 

PG&E, Slalom, and AWS partnered to put the latest AWS services to the test for this application. The team had an ultimate goal of creating a chatbot UI and enabling semantic search to reduce manual labor of visually looking through the unstructured incident report details, comments, and resolutions.

We developed the following process:

  •  Incident data stored in S3
  •  Parsed through Comprehend
  •  Ingested by SageMaker to create inference data 
  •  Resulting models stored in S3
  •  Indexed using Elasticsearch 
  •  Presented via search UI using Kibana

We also created a chatbot UI framework for future integration leveraging Lex and Lambda.

Instead of reading through thousands of tickets, CAP analysts can now reach deeply into years of data through a search interface and eventually a chatbot interface. They can use keywords, phrases, or synonyms to search for related tickets and get results ranked by relevance. The Q&A chatbot facilitates conversational style inquiries, e.g., “Show me tickets in the last 90 days related to electrical issues.” 

These capabilities will allow for faster diagnosis and prioritization. The next milestone will be to recommend resolutions based on corrective actions used in the past for similar incidents. Additional groups at PG&E are looking forward to applying the reference application for myriad use cases.

With over 400 brands spanning four main divisions, our client is one of the top consumer goods companies globally. Recently, the company has made significant investments in ecommerce to develop closer relationships with shoppers. Leaders were interested in the effect on profitability of removing one pack size from an assortment on total revenue and profitability, without using A/B testing. The client also wanted an interactive, dynamic visualization that would allow its teams to simulate different assortment scenarios on their own.

Slalom data scientists conducted interviews to understand the business need and current state of analytics. We developed a use case for one of the client’s top brand product lines across two ecommerce platforms. This brand accounts for 50% of the company’s revenue. We then built a custom simulation model to simulate the transfer of sales within a product line across pack size assortments—and an interactive dashboard for the client to drive its own simulations and analysis.

Our client was able to quickly and iteratively determine which pack sizes were most profitable and remove others from distribution. The company continues to use the simulation dashboard to optimize its packaging and product mix.

A global music corporation uses sponsored playlists on streaming services to forge better connections with customers. The client wanted to use data-driven methods to update playlists with optimal content.

Slalom data scientists conducted requirements workshops with business stakeholders to frame the business questions and then built a recommendation engine using collaborative filtering and cosine similarity algorithms to suggest the next 40 songs that should be placed on the playlist.

We built a production-ready recommendation engine with a front-end GUI that can suggest the “next best track” for hundreds of playlists based on genre or mood. This model performed five times better than the established benchmark and has decreased the manual work needed to curate playlist content from hours to minutes.

Our client, a global music corporation, wanted to use its digital data to better understand customer behavior. The company engaged Slalom to build a customer segmentation model based on ten billion digital transactions.

Slalom data scientists conducted workshops with business stakeholders to understand the business questions they wanted to answer. We then used k-means clustering to segment customer into 18 clusters based on behavioral attributes, content preferences, and demographics. Finally, we provided a detailed analysis of each cluster, including key statistics and business strategy recommendations.

We delivered a production-ready model that clustered customers into segments like “High Value Streamers,” “Night Owls and Weekend Warriors,” and “Hip Hop Crossover Fans” based on more than 45 variables. This model dramatically changed the depth and precision with which our client looked at consumer data. It's now being used to optimize release marketing, improve playlist targeting, improve pricing strategies, and increase conversion rates.