What we learnt in completing our Innovate UK Smart grant

Over the past 12 months Bendi has been delivering a project funded by Innovate UK, in partnership with Professor Chee Yew Wong, Chair of Supply Chain Management at Leeds University Business School. Delivering this innovation grant (Bendi’s second!) has been a huge learning process for us. As the project reaches conclusion, we wanted to reflect on some of the choices we have made and the challenges we have faced whilst making sure we maximised this funding. 

We were awarded the grant by Innovate UK to develop a tool to help fashion and textile companies better understand the sustainability risks in their supply chains. This is a topic familiar to Bendi. However, there were a number of choices we had to make early on to place the project on the path to success. Below we share just a small selection.


Boiling the ocean (aka identifying the starting point)
Given that grant funding was only for 12 months, we knew we could not execute fully on all the possible sustainability risks facing fashion supply chains. At the beginning of the project we spoke to a number of industry experts and with their guidance made the decision to focus on a subset of risks. We reached consensus to focus the project on sourcing risks relating to infringements of labour standards and Human Rights - which, unfortunately, are issues of high concern in textile sourcing. 


Same same but different
A significant early problem we faced was around the formatting of supply chain company names. This is a common data science problem. A human can easily tell that IBM, International Business Machines and IBM UK LTD are all the same company. However, this is not always so easy for a computer. Cleaning and disambiguating our large data sets of supplier names, and efficiently accounting for company aliases has been an incredibly important component to ensure accuracy of Bendi’s output.


Parlez-vous 日本語? 
Fashion is a global business. Producing a garment often involves numerous processing stages in factories which frequently span multiple continents. Just as the companies that make up fashion’s supply base are located in countries around the world that speak numerous languages and write in a multitude of scripts, so too Bendi’s systems have had to adapt to local environments. Given the complicated nature of labour rights violations, building local context within our team (as well as our tech) has been a huge achievement during this project.


In spite of these challenges, the outputs we’ve achieved during this project have been significant.

1. Supply chain visibility: mapping trade connections in supply chains 
   By analysing global trade data, Bendi has managed to connect brands with their indirect suppliers, resulting in a 7x increase in visibility throughout their supply chains. This process is significantly quicker than traditional traceability exercises. 
   
   However, this progress hasn't been without its challenges, particularly in terms of trade data quality and availability. In one of the pilots we completed, we encountered discrepancies between different jurisdictions and variations in the level of data availability between different countries. We were able to correct this by expanding our providers of raw data to increase our coverage.
   
   

2. Supply chain risks: identifying significant events
   The programmatic identification of supply chain risks at scale has been a joint effort between Bendi’s data science and engineering teams. Once identified, the conceptualisation of those risks has been a major research focus of this project. 
   
   In our research methodology, our definition of risk is taken from the dictionary. Simply, this is defined as “the possibility of something bad happening”. This is based on the degree of likelihood of a given event materialising, and the probability that said event will occur. Defining what qualifies as bad and assessing the degree of severity have been central to Bendi’s categorisation of events.
   
   To help assess these, we have developed frameworks that analyse event severity, risk categorisation, and source credibility. These frameworks serve as key tools for our research team to structure and classify our findings. 
   
   The frameworks have built off the substantial existing body of literature in this space. Bendi categorises labour risks according to the Fundamental Conventions of the International Labour Organisation. To assess impact and salience, our severity framework has been adapted from the work of human rights due diligence assessments. 
   
   

Finding needles in the digital haystack
Automating processes at scale is always a sticky point for any start-up. You’ve done the small scale proof of concept and everything worked perfectly - but can you replicate it if you 10x or 100x the data ingestion? 

For us, a pivotal aspect for our work has been evolving the sophistication of the filters we use to make sense of the large data sets we work with. Our filters can take various forms - NLP, topic detection, entity recognition, sentiment analysis, amongst others. These functionalities enable us to sift through substantial volumes of information efficiently, and are key to unlocking the technology we have developed at scale.


Beyond the grant
This article has just been a taste of what we’ve learnt. It’s been fun to reflect on some of the challenges and the creative problem solving that we have had to undertake as a team. We’d be happy to talk in detail about any of these - just get in touch.

We are concluding the project having built out Bendi’s product and having completed a number of extremely informative projects with leaders in responsible fashion. This has been a year of the most substantial learnings for us as a company. We are excited to share more shortly on what’s coming next.