Designing for Automation
Automation is happening all around us in big and small ways. Tesla is working toward production lines staffed entirely with robots (albeit with some delays). Amazon is already using robots to autonomously move and stack bins in warehouses. Waymo is building driverless cars and creating new paradigms for how humans will interact with them. And just a few weeks ago we witnessed Google Duplex call a salon and make an appointment on behalf of a user.
These innovations have dominated headlines, but many lesser-known automation experiences already permeate our lives. Financial management websites are using artificial intelligence (AI) and machine learning to automatically build their clients’ portfolios and maintain diversification. Credit card companies use chatbots to notify their customers of potential fraud and allow them to resolve issues with a simple text message. Wellness applications autonomously monitor health data and offer advice via both human and AI-driven coaches.
With all of these new technologies come complex design, communication, and trust issues. In this post we’ll explore some of these issues and suggest ways designers can successfully navigate the uncharted automation landscape. We’ll include high level design process recommendations and three key design principles that can be applied to nearly any automation solution.
Automate to streamline workflows, or eliminate them altogether
If you’re unsure what aspects of your product can benefit from automation, consider conducting generative user research. Generative research such as ethnography or contextual inquiry helps designers understand their users’ goals and challenges which may then spark ideation around automation. For example, through shadowing sessions you may observe users doing mundane, repetitive tasks which take time away from more productive, intellectually stimulating work. Automation may even be able to remove those tasks from the user’s workflow. User research can also uncover ways automation may be used to prevent problems. Take a cardiologist who oversees hundreds of patients but often doesn’t see them until something goes wrong. With an automated remote monitoring solution, the clinician could be alerted before patients cross dangerous thresholds, enabling the clinician to help the patient through less invasive, more preventative measures.
Don’t automate the things humans do best
Nearly every product can benefit from automation, but there should be careful considerations around when not to automate. For example, applications like Sketch will delight their users when boring, repetitive tasks like batch processing are automated. On the other hand, creative users might be frustrated if more fulfilling tasks were automated such as color and typography selection. Privacy, safety, and the sensitivity of the subject matter must also be factored into deciding whether or not automation is appropriate. Let’s say a computer analyzes lab test results. Automatically notifying the patient and telling them that they need to follow-up with their doctor is possible but highly discouraged. Consider that this information could greatly upset the patient, and it would therefore be beneficial if a human delivered the information and remained present to put the patient at ease. Designers should carefully weigh all of the possible pros and cons before recommending an automation solution.
Start fresh to design the best solution
One easy trap to avoid when it comes to automation is simply replicating existing functionality in a more efficient way. While this design strategy may improve the status quo, it may not fully harness the power of automation. Designers should take a step back and focus on what users are trying to achieve, not how they are currently achieving their goals. For example, at Flexport, our operations team often receives documents from clients and must share many other types of documents with service providers and government agencies. Our automation solution for this problem could simply involve offering better upload or document management tools. Instead we have shifted our mental model from a document-driven one towards a data-driven one. We are completely reconceptualizing how we gather data from our clients and aiming to remove explicit document transfer from user workflows, which makes automation easier and is vastly more efficient for all parties.
To build trust, incorporate users throughout the design process
Automation often involves removing user-facing controls and tools. Even though the visible aspects of the user experience may be reduced, the underlying user experience isn’t any less complex. The smaller, nuanced interactions which comprise an automated system can make or break the user experience, so it’s critical to get the details right, particularly around feedback timing and messaging. Designers may also encounter user concerns that go beyond usability. Prospective users may be scared of automation simply because it’s new, unfamiliar, different. The potential for such issues underscores the importance of early user research to understand user needs, preferences, and concerns. Armed with these insights, designers are more likely to create innovative solutions that go beyond making existing tasks more efficient. Equally important, user research can help designers establish trust with users, which is critical for successful adoption.
Automation design principles
Designers can apply the following principles to nearly any automation design solution:
1. Display system status. Without sufficient understanding of what an automated system will do, what it’s currently doing, or what’s been done, users may not trust the system. Explore ways to educate users on what the system will do, and communicate its progress along the way. Once the system has finished a given task, indicate who completed the task (human or computer), precisely when the task was done, and your confidence level in the outcome. Finding just the right amount of feedback will depend on the users and the problem you are trying solve.
2. Support human-computer collaboration. Some automated experiences will reliably get the desired outcome 100% of the time but others may need minor modifications. Enabling users to fine-tune the system will improve outcomes and can also be an opportunity to improve any underlying machine learning. While this can be helpful for training machine learning systems, it’s important to make sure users aren’t overriding automated systems so frequently that they revert back to manual operation. Human-computer collaboration is also important when some parts of a process are more suited to a computer whereas others are more appropriate for a human. For example, a computer can quickly process large amounts of data, but the human may be more adept at completing the final analysis when the results are inconclusive and need additional review.
3. Allow for human intervention. When system status and human-collaboration are insufficient to produce an optimal outcome, the software should allow users to stop the process and manually intervene if they see something is wrong. For example, Waymo allows humans to take control of their vehicle because of a system failure or a traffic, weather, or road situation that requires human intervention. And 2001: A Space Odyssey fans will remember HAL, the sentient computer, who malfunctions and needs to be completely shut down. At the same time, designers should consider how some skills could become obsolete as automation becomes more pervasive. Phantom Auto is tackling this issue by enabling humans to remotely operate autonomous vehicles when the vehicles encounter scenarios they cannot handle on their own.
Designing for automation at Flexport
Flexport is a creating an OS (Operating System) for global trade. Automation is essential for obvious reasons such as efficiency and scale, but it also enables future innovations. Because of the important role automation plays for us, our designers often have an “automation first” mindset. We want to ensure we aren’t just making antiquated, inefficient processes faster. We want to challenge whether the processes should even exist. We are also mindful that technology alone will not enable us to provide the best service possible. Although we hope to build the most intelligent software for global trade, it will always be complimented with an equally intelligent operations team.
Want to learn more?
Come join us at Flexport’s San Francisco office for the SF Design Week panel Designing for Automation to be held on June 14, 2018 from 6:00 to 8:00 PM. Panelists from Netflix, Verily, Instacart and Flexport will share their perspectives on automation. Read on for more info about the panelists and our esteemed moderator, Cara Silver.
SF Design Week Event Page https://2018.sfdesignweek.org/events/designing-for-automation/
Cara Silver is a UX researcher who strives to find the deeper story. She is currently building the research team at Skype. Before, she led research teams at Verily and frog, and has extensive experience in international research contexts (including China, Afghanistan, Zimbabwe, and USA).
Ghaida Zahran is a Product Designer at Netflix and co-leader of the Oakland chapter of Girl Develop It. She’s a techno-optimist who’s super passionate about how tech and automation can improve people’s lives. She enjoys running, videogames, and exploring California.
Peilun Shan is a UX Manager and Interaction Designer at Alphabet’s Verily where he is working on a diabetes management solution for Type 2 diabetics and previously worked on Project Baseline. Prior to Verily, Peilun led design at 23andMe where he launched a direct-to-consumer genetic test.
Himani Amoli is a Senior Designer at Instacart, working to improve the Shopper experience for Instacart’s vast global shopper network. She co-founded Wedding Party which was acquired by Instacart and designed the first version of Aaptiv, a wellness application.
Zoe Padgett is a Product Designer on Flexport’s Air team who believes in serious user research and playful experimentation. Prior to Flexport, Zoe helped develop the product vision for IBM’s Data Science Experience and was a Design Fellow with UNICEF’s Innovation Lab in San Francisco.
To Ghaida Zahran, Himani Amoli, and many Flexporters for their insightful feedback on this post.