Design sprints are the hands-down best way of getting started and making early and rapid progress on a messy, upfront, ambiguous design problem.
- Design sprints are in the goldilocks zone for time and attention. People can carve off a week for a sprint if they know they’ll get a high fidelity prototype that gives real data on the most important design choices and save them weeks of unproductive meetings and arguments.
- Design sprints do something clever: they give a team permission to spend valuable time with a single focus, working “alone together” on the most important problem. They reduce debate, eliminate the bias of the person who has the most to say and increases the diversity of solutions. Even though brainstorming is fun, sprints have daylighted how brainstorming is useless; a placebo for team building. Sprints solve real problems by getting the right minds on the problem, for the right amount of time and in the right way.
- With a “decider” always present, a sprint creates an audience for the solution and an accountability for getting unstuck, making decisions and getting on with the project in a way that no other technique has been shown to do. Forget “executive buy-in,” a sprint modernizes organizational accountability in the age of agile.
These three reasons make sprinting an ideal way to get cross-functional involvement, to understand an opportunity and to accelerate progress with a complex and emerging technology.
We chose the Google design sprint methodology as our starting point and customized this popular technique to make it work even better for AI, specifically non-technical people who want to design AI that works for humans.
What makes an AI Design Sprint different?
We have added new steps to our sprints to make them even better for AI.
- More time on figuring out the problem to solve. AI is so powerful that it can make it possible to solve problems previously considered unsolvable. We apply techniques that help teams mine the corporate memory and combine potential ideas with new AI solutions, revealing creative ways to solve persistent problems or entirely new opportunities which have been made possible. The power of AI means that people need to consider how best to use AI. Just because it can be done, doesn’t mean it should.
- Steps to help you figure out if AI is the right solution. For almost any problem there is a low tech solution. Our sprints make it super clear where AI is worth it. AI needs to address the problem in a unique way. We take teams through the process of data evaluation, the mental models and expectations of users and how AI would interact with these.
- Specifics of AI design. We spend time on AI specific design choices. Teams make decisions on key trade-offs such as precision versus recall; is it more important to include all of the right answers even if it means letting wrong answers slip through or is it better to minimize the number of wrong answers at the cost of leaving out some of the right ones? Teams plan and design for co-learning and adaptation. The sprint uncovers assumptions and mental models so that designers avoid users manipulating an output according to their own imaginary rules, seeing conspiracy theories and imagining other intelligences.
- AI prototyping. Perhaps the most challenging – but also the most rewarding – part of a sprint is the prototyping phase. AI design brings particular challenges so teams work to identify and test the most important aspect of an AI system. People respond differently when they believe they are working with an autonomous system and our sprint prototyping methods capture the human considerations that are the most important for business leaders to know and understand well before any AI system is built.
- Planning for AI in the wild. The most valuable AI systems evolve over time in tandem with users’ mental models. In the sprint, design teams answer the question “how do we create a virtuous circle for the user and the machine?” Once the system is in the wild, new inputs will, by definition, result in behaviors that can’t be predicted. Sprints anticipate this for what to do when things don’t go as planned.
At the end of an AI Design Sprint the design team will have a broad portfolio of hand-crafted examples of what the AI is to produce, a set of policy considerations for senior managers to worry about, a high-fidelity prototype, a roadmap for data collection, a strong set of labels to start training models and a framework for designing large scale labelling protocols.