As a global telecommunications supplier, with coverage across more than 70 countries, Vodafone were faced with an enviable, but difficult challenge. With such a high volume of sales, across so many regions, comes the need to ensure product descriptions, stock keeping units (SKUs) and product variations can be tracked and managed effectively. Vodafone were operating at a point where there were inconsistencies between product types in one region against the same products in another.

They needed a solution that could effectively, and accurately, identify product types by description alone, so that they could be managed under one closely managed and centralised data point.

Vodafone has been working with Firemind for a number of years now, with some of our earliest work revolutionising their invoice processing prowess. From the back of that case study’s effectiveness, came the need for a more expansive project of a similar nature, that could help solve the challenge of inconsistent product lines of mobile devices, routers and a host of other telecom units.

The project began by undertaking a thorough understanding of the SKU IDs in play at Vodafone. We had to assess the current state of product lines, descriptions in use, as well as the commercial ID properties we’d need to map towards.

This was to be a project with extensive use of Amazon Web Services (AWS), to enable effective management of the data, as well as allowing for uploaded file types to undergo machine learning (ML) assisted processes.

Uploaded files were sent into an Amazon S3 bucket, an object storage service that allows for unparalleled data storage for any use case. This step then triggered an AWS Lambda function, enabling a serverless, fully event-driven code processor, used for most application services.

From the Lambda function begins a set of three unique AWS Glue crawlers, each specific to the data type in question. We enabled one for general SKUs, another to check against Commercial IDs in use by Vodafone, and finally, a Full Hierarchy Crawler. We use Glue as it is a serverless data integration service that makes it much easier to discover, prepare, move, and integrate data from multiple sources for analytics, ML, and application development.

Glue was specifically useful for this particular project as we could enable multiple ‘transforms’. Transforms allow for ML integration, by finding similar data records, then teaching further Glue processes how to identify matching (and non-matching) records. Refining and becoming more accurate after every use.

Following on from this ML part of the process, data would go through a series of Lambda functions, Amazon EventBridges and then into a new S3 bucket, where the matching records were correctly categorised and ready to collate by data owners within Vodafone.

Time is often of the essence when dealing with solutions for global, active companies, like Vodafone. We worked quickly to assess the challenge, plan the AWS architecture, build, test and deploy the solution. Solving a specific, but harmful challenge, within a single business week.

By using AWS Glue Transforms, Vodafone were setting themselves up for succesful reproduction, of a workflow that could be beneficial to many areas of their business. They’d already seen the benefits of intelligent document processing (IDP), in a past project with Firemind. Now, armed with two unique ML processes, they can better prepare for further scalable, resilient and smart workflows, using the AWS cloud.