Help Me Decide

In partnership with UX researcher Shenoah Plewes, we collaboratively set out to understand the problem domain, current competitive landscape, and create proof of concept prototypes to develop learnings.

Develop a systematic approach to delivering relevant insights to merchants about their business. Understand what makes insights useful, through proof of concept testing, theory research, and practical information gathering.

As this was a completely new unexplored area, it was clear that the first step was to understand the basics, before an entire platform could be built.

From interviews and research, we identified that an insight is: not obvious, helpful in decision making, and comes from a place of being deeply informed. An insight becomes truly useful when it can provide the right information, at the right time.

A good decision is one where: you have sufficient information, understand your options, use that information to make a decision, and can learn from that decision. Insights are very well positioned to solve the step in understanding and evaluating your options.

We identified two categories of decisions: simple, and complex. Simple decisions have a right and wrong option there is a binary (yes/no) choice. Complex decisions require weighing multiple factors, to come to a decision, and is fuzzy, or grey, the right or wrong choice may not be clear until you receive feedback on the outcome of that decision. Sometimes you may never truly know, but we set out to investigate those where the outcomes could inform whether it was a good or bad decision.

An example of a complex decision: I need to decide what new phone to buy: Apple iPhone, Google Pixel, or Samsung Galaxy, or a host of other options. I might ask questions like: How much does it cost? How much storage does it have? What are the specs for the camera? What does it look like? What is my level of trust in the brand? All these factors play some part in my decision to buy one phone over another. Based on researching Multi-Criteria Decision Making Analysis (MCDA), and Analytical Hierarchy Process, we learned that when these factors are appropriately weighted together, they can help to determine the best option for you.

Method: Conduct a broad investigation into: web-technology insights, as well as far reaching domains with practical applications.

1. Establish Data Foundations to Leverage: Insights rely on large swaths of structured data in order to be useful. Google for example is able to provide useful insights because of its data platform investment over many years.

2. Know your Limits: Insights can either be hands-on: where they take the action for the user, or hands-off: where the user is always in control, and the insight serves to assist with useful information. The level of control the user retains depends on the products objectives. For example Airbus and Boeing, have completely different philosophies related to pilot control. Airbus has a philosophy where automation should be leveraged wherever possible, where as Boeing has a philosophy the pilot always retains final authority.

3. Timely: Insights have the highest chance of being useful when placed alongside decision making areas of the software product. For example: I want to run a marketing campaign to boost the traffic coming to my website. I have an important decision to make: should I optimize for traffic quality, traffic quantity, or balance both? Presenting users with helpful information where and when they need to make that decision can make for more informed decisions.

Based on our research, and heavily inspired by SAE’s levels of self-driving car automation, we identified eight levels of insight and automation.

The usefulness of level 7, and level 8, are deeply dependent on the goals of the product. For example, self-driving cars could make use of levels 7 and 8, when drivers aren't in the mood to drive, and the system can do the hard work for them. In other contexts, where the task is fulfilling and meaningful, automation may remove tasks that could be enjoyable to complete. It's important to remember that automation serves the end goals of the user, not the other way around.