Uber's journey with Automation and ML in revolutionizing operations

In an age marked by the digital renaissance, businesses are increasingly turning to automation and machine learning (ML) to revolutionize their operations. The convergence of automation and machine learning technologies is driving a transformative shift across industries, as businesses harness their combined power to enhance operational efficiency, scalability, and innovation.

In an interview, Anand Janardhanan, Director, Program Management - GSS at Uber shares Insights about their own automation and ML odyssey, offering a roadmap to steer through challenges, capitalize on opportunities, and drive a transformative revolution that shapes industries and enhances lives. He offers a glimpse into the future, where generative AI and large language models paint a canvas of boundless possibilities, elevating Uber's services to new dimensions.

Excerpts:

Can you provide an overview of Uber's Global Scaled Solutions team and its role in revolutionizing operations with automation and ML expertise?

Global Scaled Solutions (or GSS) is a function within Uber that brings operational and technical expertise to help scale various parts of our business. We deliver on a range of organisation-wide programs - e.g., end to end sanity testing of our products; large-scale data annotation services for ML development, operations support for varied functions like risk, menu digitization, account management; localization and translation for our product and marketing teams; tech stack migrations; data and analytics services.

Clearly with such a scaled and diverse program portfolio, our intent is to go in with a tech-first mindset - which means using the best tools, techniques, data to run these at the best quality and cost outcomes. Hence automation and ML is a big part of our focus for the scope that we manage, and we regularly go through the cycle of design-build-deploy for AI and ML solutions in our work.

How has the implementation of automation and machine learning technologies impacted Uber's operational efficiency and scalability on a global level?

Across Uber, the impact of automation and machine learning has been very significant. For instance, we’ve enabled automated ML solutions to validate and process the documents of new drivers and merchants, which allows them to onboard to the platform and start earning quickly. For Uber Eats, physical menus/catalogs of restaurants and grocery merchants are converted into structured data through ML-powered tech (such as Optical Character Recognition) that eliminate manual effort and lower operating costs. The augmentation of the menu content using custom ML models means that consumers can access item details such as name, description, price, calorie information, and add-on variants to make their choices, hence driving an enhanced customer experience.

Safety is a non-negotiable goal for us on the Uber platform, and we use automation and ML to enable this as well. As an example, account-sharing for driver-partners is not permitted, and we periodically ask our drivers to upload a selfie - these images are assessed at scale using image recognition tech to check for digitally tampered selfies or picture-in-picture, and low-confidence matches are reviewed and actioned.

These are some illustrative examples, and over time we periodically identify and solve for new use cases for ML technologies to drive efficiency, cost and safety outcomes for Uber.

What were the major challenges faced by the Global Scaled Solutions team during the adoption of automation and ML, and how were they overcome?

With any new and innovative tech, we can expect that there will be some challenges or ambiguity around how it can be effectively harnessed. For ML automation, especially with emerging LLM tech, one aspect is the trade-off between using plug-and-play solutions which are readily available, vs. building custom solutions - here we not only need to assess comparative functionality, but also be very mindful of what data, if any, will these ready-made solutions ask for. So we get into the details of what versions we can use, and what we need to build at our end, in order to do the right thing from a data privacy perspective.

Another challenge - let’s say we deploy an AI solution to interact with our customers or earner partners in a support use case. In any such scenario we need to be absolutely sure of the quality of these interactions, and that there will be no negative impact to the user’s experience on the app. In GSS we use several approaches to address this - e.g., ensuring rigor in the model training and testing to drive the right quality, put in a human ‘in the loop’ along with the AI to handle cases with low algorithmic confidence. We also go through a stagewise ramp up of the deployment to iteratively assess and confirm its performance. We build with heart, and care deeply about the impact of our product with our user community.

Uber operates in diverse markets worldwide. How does the Global Scaled Solutions team tailor automation and ML solutions to meet the specific needs and regulatory requirements of different regions?

When we identify a potential ML use case, even if it’s initially from a single region, one of the initial evaluations is to assess whether it’s a globally applicable functionality or a region-specific one. Alternatively, if a global-level solution exists, can we put a region-specific wrapper around it to account for additional regulatory and other requirements, while still getting speed-to-market. It’s a question of our product experts seeing the forest and the trees, so to speak.

And when we build, we solve similarly for re-use, scale and maintain-ability vs. building something bespoke, in the tech design and architecture phase. So today, even if we build something that’s specific to one region, it’s done so in a modular way, and using consistent global-first principles.

With the continuous advancements in AI and ML technologies, how does the team ensure that Uber stays at the forefront of innovation in this area?

As a leading tech company, we have always made the right investments in specialized talent and skills in AI and ML, and this is true today for GSS, too. Our ML Automation team in GSS works closely with other AI practitioners within Uber, and are plugged into communities and interest groups that track and adopt the latest innovations in the market. We’ve incubated an ‘Innovation Lab’ in GSS whose purpose is to be forward-looking - identify emerging tech trends, try them out in proof of concepts, and proactively re-imagine what tomorrow looks like.

GSS also actively participates in Uber-wide programs focused on ideation and innovation. ‘UberML’ is an annual global conference where we conduct presentations, panel sessions and workshops for participants all over the world. We recently held a program called ‘My Innovation Time’, which was a grassroots ideation campaign for our India tech teams. Through the year, we periodically tap into such opportunities to engage, learn and play with cutting-edge tech, it’s a very exciting vibe for our people!

What advice would you give to other companies looking to implement automation and ML strategies on a global scale based on the experiences and lessons learned at Uber?

“Eyes on the sky, and feet on the ground,” is my bit here. Rather than thinking incrementally, one should be bold in thinking about how AI, ML can totally reimagine how something works. This is when we can truly unlock the transformative power of these technologies. At the same time, one should continue to care deeply about aspects of data security, customer experience, so that we always do the right thing.

What excites you the most about the future of Uber's automation and ML initiatives, and what can we expect to see from the company in the years to come?

Not surprisingly, the excitement now is all about generative AI and large language models (LLM). As we have seen so far with existing ML capabilities over the last few years, these new solutions now also open up enormous possibilities for innovation. For instance, how we can create a more magical user experience for our community using AI and LLM-led capabilities in the context of user support, personalization at scale, language experience and localization. How will emerging ML improvements help enhance a typical ride or delivery in future - with better routing for improved ride quality, with more relevant and personalized menus and grocery selections. And broadly in the industry, can advancements in AI re-shape the evolution of autonomous vehicles which Uber can leverage? We’re very excited about these prospects, and look forward to riding this next wave for the future in GSS!