How to implement Machine Learning projects beyond algorithms?
A Supreme Audit Institution from Latin America aimed to transform the audit process based on incorporation of state of the art technologies in order to increase transparency as well as control over government projects and their bidding process. The project was financed by the InterAmerican Development Bank (IADB), one of our long-term clients. Our objective was to find solutions to their audit’s needs through the application of Machine Learning algorithms and transform their work processes as well as their organizational culture to be able to carry out these type of projects.
We decided to begin the project with a visit to the customer with the purpose of collecting all the necessary information to address our objectives. It helped us understand how mature they were in the use of data, which were their weak points, and in which areas they needed help from our consultancy.
This project meant for us many learnings on how to work with our clients. One of them was the importance of understanding the client’s current position in terms of technology, teams, processes, and organizational culture and their needs so as to define our approach and where to focus our efforts. As well as in the case of this institution, in every organization, for Machine Learning projects to work, it is essential that not only do operational units involved adapt but also that there is upper management buy-in and that the push reaches every vertical of the organization. This implies a substantial cultural shift for any organization and it is key for a successful implementation. From the people needed for each stage of these type of projects, to their infrastructure, workflow, data mining, model development, and data analytics, among others, they all had to be part of this same process. As we progressed through the project, we learnt it is not simple and done in a short period of time because it depends on many different areas as well as people, and requires an evolution of the company’s organizational culture towards data.
We achieved a positive workflow with them through in situ visits and workshops, especially when designing the technological infrastructure model so that they could progressively implement Machine Learning projects in the future. The quality exchange in our communication made it possible to accomplish outstanding results: after our visits and workshops they started to replicate the workshop modality to start with other Machine Learning initiatives progressively, major involvement from them was achieved. We guided them on defining the problem to solve, based on their data we recommended the use of unsupervised models, within these we suggested the most efficient one for their problem, isolation forest, sent them the algorithms’ codes and then discussed their feedback together. Their participation was truly proactive and we were able to actively collaborate with them on continuous improvements to the model and algorithm. Through this they showed they have qualified professionals for this kind of data focused project and needed order as well as organization. In other words, their accomplished human resources needed to make a change and adapt to these type of projects, that call for time, achievable deadlines, putting a team together, and transforming the original workflow.
One of the main challenges we encountered was building trust with them. This was key for them to share with us all the data needed for Data Mining. Considering the power this public entity has, and the sensitivity of the data it collects, data privacy was a major concern. Listening to them and understanding their situation along with their needs was much more important than collecting data and running algorithms. They needed guidance and support from an experienced team in Machine Learning projects that transmitted the way to proceed and proposed the tools needed to the organization as a whole.
Looking back at the project we see clearly how soft skills are as important as hard ones for Machine Learning projects. Our challenges related to navigating the teams and processes were as key to the success of the project as providing the right tools and algorithms to solve the proposed requirements. It is paramount to understand the client’s current maturity in data management, identify which are the points that need more work — in terms of technology, teams, processes, and culture — as well as to build the proper trust in order to become a part of their team that would defend their own interests. These are our key takeaways from this project.
This project left us multiple learnings, but what did it leave them? They continued with Evaluation and Deployment, since we carried out this project with Cross-industry standard process for data mining, known as CRISP-DM, which includes different stages that we covered until Modeling because of time constraints. Frequently, prototypes are made and processes are followed up to a point and fail to be finished. Fortunately, not only did we see real will in them to continue with this project but also to go on with Machine Learning Initiatives in the future.
We help clients grow their businesses by making creativity, flexibility, and talent converge into software development and UI/UX design. We are passionate and committed to take on challenges that can drive your business forward.
We combine our experience in Ecommerce, mobile, and web across various industries such as retail, wholesale, healthcare, government, and entertainment. This way, we work closely with you as a team in order to collaborate towards successful implementations. See more at www.onetree.com