How to Implement a Design Sprint Approach to AI
The business potential of artificial intelligence (AI) is growing daily, especially as the technology moves beyond prediction and analysis toward true, autonomous decision-making and unlocks unprecedented levels of insights, knowledge, and innovation.
However, while “most executives know that artificial intelligence has the power to change almost everything about the way they do business and could contribute up to $15.7 trillion to the global economy by 2030,” asserted PricewaterhouseCooper in a recent report, “what many business leaders don’t know is how to deploy AI, not just in a pilot here or there, but throughout the organization, where it can create maximum value.”
It is imperative for organizations just starting out with AI to understand the complete picture — how to go from idea to production deployment. That’s why pursuing a one-week AI design sprint can provide a clear view into what is possible, which will facilitate setting expectations on timeline, features, and quality.
A one-week design sprint will help introduce your entire team to modern AI technologies and methodologies. Be sure to tackle a real business challenge and achieve tangible results. This will serve as a proof-of-concept for all your AI innovations.
Here’s a closer look at how an initial design sprint should play out. The specific steps for each phase should be adjusted based on your particular needs.
Day One: Discovery. Following a pre-kickoff, gather the AI team for this structured conversation, which will create a path for the week. Start with the long-term goal, map out the challenge, ask your domain experts to share their knowledge, and pick a sprint goal — an ambitious yet manageable piece of the problem that can be solved this week. The mapped-out challenge should include business requirements, expected outcomes, data sources, deployment scenarios, potential opportunities, and risks.
Day Two: Assessment. Domain experts should provide initial feedback based on the discovery phase. They will report on the data and make solution recommendations. They will discuss building a strategy roadmap for the project based on your specific business case. Any changes to your infrastructure should be highlighted during this step.
Day Three: Design. Prototype several solutions to the sprint goal and identify a candidate solution that seems to be the best fit. Data engineers should prepare the data. Data scientists should discuss and select appropriate features and machine learning algorithms. Machine learning engineers should design, build, and perform preliminary tests on your prototype neural network. Time permitting, start the iterative process of design and discovery on the data and the neural network model.
Day Four: Implementation. Complete the design process and begin training and testing your AI model until it reaches the desired accuracy threshold. Build a pipeline that will put your model into a suitable environment for testing and feedback from additional stakeholders. Domain experts should offer guidance on assessing machine learning predictions and putting discovered insights into action.
Day Five: Operation and feedback. This may include UI development, technical documentation, and hands-on knowledge transfer between development and operations teams to ensure they can operate the solution. It is important to discuss production deployment options — for example, a dark launch or integration with a prototypical business application. During this step, elicit feedback from key stakeholders, covering the baseline architecture and model.
This five-day program relies on a couple of extra steps to ensure a successful experience:
Day Zero is the aforementioned pre-kickoff. This involves getting enough of the preliminary information, data, stakeholders, and systems ready for Day One. It includes preliminary discovery and assessment with a small set of domain experts who are capable of delivering the key artifacts for this phase. These artifacts serve as input for the team attending Day One.
Day Six is a post-mortem. It is a longer follow-up to Day Five. The key output from this final step is the set of lessons learned during the sprint week. Examples: What was learned during the design sprint? What are the implications for the preliminary plan that was created for the broader program? What action is recommended and presented to project sponsors? What worked well? What didn’t? These lessons serve as input into the AI program and contribute to creating a virtuous cycle.
With any new major new technology, sometimes just starting is the hardest part. By following the design spring philosophy, organizations can start their AI journeys on the right foot.
Feature image via Pixabay.