Fashion marketplaces provide a hub for the circular economy, allowing clothing to avoid sitting unused in a cupboard or ending in landfill.

Key to the utility of the marketplace is discoverability - how do you find something you might want to buy, when the inventory spans everything from niche 1980s football shirts to designer handbags. The search engines for such websites therefore become a critical path.

In this work our focus was on improving the quality of search results for our client. We started by training better machine learning models to rank the search results. We then addressed the data science process itself, and through automation of workflows and upskilling of the team we were able to more than double the number of A/B experiments data scientists released in a year.

Overall our impact was measured in lifting purchases from search by over 10% and unlocking over £10 million in annual GMV.

The manufacturing industry is a complex space, with many moving parts and legacy processes that have been shaped over decades of experience across the industry.

Enter AI, which promises to be a disruptive tech, and one whose adoption and growth is skyrocketing arguably faster than even the internet did at its inception.

Companies in this space therefore need to have an answer to the question of how they’re going to leverage AI, at the risk of being left behind not just by competitors but their own staff, as usage of things like ChatGPT becomes ubiquitous.

In this engagement we worked to shape the company’s AI strategy, working with stakeholders across the business on a cohesive plan, focusing on the highest leverage and lowest risk items first. We additionally led the execution phase, launching generative AI platforms for text and image workflows.

Key to our approach was a strong focus on security and privacy, enabling our client to safeguard their sensitive IP whilst allowing access to staff across the business to the latest developments in AI for accelerating knowledge-work.

Being able to prove the provenance of high value goods between multiple suppliers in a chain in a reliable and trustworthy way is a difficult challenge.

In this engagement we worked with a client with a legacy traceability platform, in which the data science and software workflow was tightly coupled. The platform was being used in production globally with multiple high value clients, and so it was critical to not disrupt core business operations, but data scientists were struggling to improve matching accuracy without doing so.

Our first step was to implement best practices for turning machine learning models into production - raising the bar on CI/CD, local testing and development, and upskilling the team in Python best practices.

We then re-architected the platform in a systematic way without disrupting production, to enable data scientists to safely iterate on their models without needing to worry about the more engineering bits taking place in Kubernetes/Docker/SQLAlchemy etc.

The simpler architecture helped us address performances issues, such that the system was able to scale better under peak load and processing time was reduced by 95%.