Snapshot

Good Research has been working with a large transportation Client on a variety of projects for several years. As they have grown over the years in terms of reach, scale, and reputation, we have been able to help them with their privacy and security needs. As one of the largest commercial telematics vendors in the world, this Client has an excellent leadership team that prioritizes privacy. During our time together, we’ve built strong individual relationships with their engineers, lawyers, and company leaders. 

Landscape

The Client provides a suite of products that provides extensive analytics for managing transportation fleets. Launched over 20 years ago as a small family business, they now have over three million customers worldwide. The wealth of telematics data has enabled fleet managers unprecedented transparency into operations, allowing them to optimize routes and schedules, cut costs, increase efficiency, and manage driver safety at a level of detail that has  never before been possible. 

However, the amount of data and intimate nature of the information collected on location and routes poses challenging questions around privacy and data responsibility. To add to this complexity, geolocation is notoriously difficult to anonymize, and in some cases, common methods of perturbing geolocation datasets can actually impact safety and reliability. Despite these challenges, we do not see privacy and usability of data as tradeoffs, but rather essential teammates. By embracing a privacy-first approach to data, we believe that companies can use telematics to provide even greater societal benefits while respecting individuals’ data.

Motivation

Our relationship began in 2019, when the Client was considering releasing a dataset for a hackathon and needed help de-identifying the data. They hired a firm that provides data anonymization services and privacy risk assessments. The external firm engaged Good Research as an independent third party to run a motivated intruder attack to assess the likelihood of re-identification.

Our analysis uncovered a series of risks that had not been considered or even expected. The Client’s head Privacy & Data Governance was impressed not just with the work but also with the way our teams worked together. 

The Client then invited Good Research to their 2020 annual conference where we met some “freaking awesome, super knowledgeable” people and learned more about their data infrastructure. It was clear the Client was interested in developing a comprehensive privacy practice to be used across the entire company. The following describes one recent engagement, a training Good Research conducted in 2022. 

Approach

Today, when working with human data, data scientists and engineers need more than a basic understanding of privacy; however, many do not have the experience or training in privacy or privacy engineering. The Client had hired a mid-career data scientist who did not have experience in privacy engineering but had an interest in developing the skills and knowledge, not just for her job, but also for professional growth. As part of their responsibilities, the data scientist would be running tests to de-aggregate and re-identify the company’s aggregated and de-identified data and find opportunities for automation. Good Research was brought in to teach them how to design and conduct the tests as well as to help them learn the tactics, tools, and processes to scope and address privacy problems.

We created a customized curriculum based on their experience and the Client’s needs, and identified four learning objectives:

• Learn various methods a data scientist can use to scope and address a privacy problem
• Develop a deeper awareness of what privacy engineering is and how it works in practice
• Evaluate different techniques to protect data (i.e. lower the risk of re-identification)
• Identify connections between data science thinking and privacy engineering thinking

Our two four-hour sessions were broken into four sections.

Privacy by Design – In Practice

This section offered a high level understanding of current guidelines and regulations. The goal was to understand the relationship between policies (what is required by law), promises (what the company commits to doing), and practices (what the company actually does). We showed the roles a privacy engineer plays in each step of the data life cycle and introduced core concepts in privacy, such as anonymization, de-identification, identifiers, and personal information.

Protecting Data

In this section, we moved into privacy engineering techniques and methods. This included teaching different approaches in aggregation, encryption-management mechanisms, and potential implementations to achieve differential privacy (such as noise blurring). We also covered user-based and role-based access controls. 

Testing & Validation

Knowing the core concepts and techniques was a good foundation to then move to testing and validation. We used case studies to demonstrate the importance of testing, specifically when data releases have gone wrong. We presented a structured approach to talking about and implementing re-identification risk. We looked at threats to de-identified data and our approach to conducting motivated intruder exercises. 

Privacy Tech, Online Tracking, and other topics

Here, we chose topics that the data scientist would be able to apply immediately to their work. We provided a snapshot of vendors who offer privacy tech solutions and how to assess their products and services. We also included a section on what’s really going on with web and mobile app tracking, using real world examples, and left a set of recommended privacy tools 

Value

At the conclusion of our most recent engagement, a senior member of their team remarked that there is no one out there with the breadth and depth of expertise that Good Research has. They were struck that we were able to talk policies with regulators and compliance teams, as well as talk technical with their data scientists and engineers. We were able to get some hands-on experience working with challenging telematics data and continue our quest to bring more privacy engineering skills to more technical teams. 

Conclusion

This Client has been a leader in applying a risk-based approach to privacy. Whether creating processes, such as in actual code and data pipelines, or building multi-disciplinary teams who need to be privacy fluent, Good Research has been able to help systematize how the Client thinks about privacy engineering. As the company has grown and changed, we have grown and changed. Together, we have made decisions and built technology that is intrinsically valuable to human life and economic value.