Canvas for AI Strategy
How to build user-centered AI products
AI is a technological transformation that will generate economic benefits across most industries. But many factors prevent companies from adopting AI. Data is one of the major stoppers. While it is plentiful, many companies do not have a strategic plan to collect the data needed to build successful machine learning use cases. In cases where data has been collected the blocker to adopting AI is the labor-intensive process required to prepare the training data. Labeling data is a human-led activity, and machines need a lot of this training data to learn. Contrary to the common narrative, AI will employ many humans of many skills and backgrounds before it can displace some job activities. Figure Eight, for example, employs human workers across the globe to crowdsource the labeling work. By doing this, clients of Figure Eight can successfully build use cases with AI.
Another major issue hindering the adoption of AI is the technical barrier facing product managers and designers. Product managers are responsible for defining the value proposition of new products. Yet they lack a framework to generate ideas where AI is essential to delivering measurable customer value. I want to propose such framework which has taken shape after months of working with IBM Watson clients, collaborating with some of the smartest minds at IBM, and teaching courses in entrepreneurship at Texas State University.
Start from the beginning with the Job to Be Done.
If you you are new to Jobs To Be Done start with Clay Christensen book. Then read Intercoms book on applying Jobs to Be Done to software products. Like the authors at Intercom, I strongly believe Jobs to Be Done is the best way to design for users. Personas are useful artifacts to align big distributed teams, but in my experience that is the extent of their usefulness. In my experience Personas are deterrents for understanding user motivation, they bundle users with artificial dimensions which in turn produces broken user interfaces. To illustrate the AI strategy framework lets look at a job many office workers have performed at some point in their career: impress our coworkers by organizing a restaurant reservation that can provide a great experience for everyone.
To get started we need to map the progress needed to achieve this job:
1.Make a decision to go to a restaurant
2. Create the list of attendees
3. Invite all attendees
4. Receive confirmation from attendees
5. Finalize list of attendees
6. Find out dietary restrictions
7. Find restaurants matching dietary restrictions and availability
8. Evaluate ratings from qualified options
9. Share final options with a small group of coworkers
10. Make decision
11. Make reservation
Mapping the steps will also help you identify the scope of your product. Very few companies should attempt at solving the end to end of any job. Doing so dilutes the user experience with overwhelming features and options. Companies are not strong at solving every problem and competition becomes meaner once products enter territories where others are specialists in solving the job. Scoping is particularly important when using AI. A well-defined problem is a core to building useful systems that generate business value. For our exercise, we will focus on steps 7 and 8. These steps are an arbitrary choice in this case, but if this was a real use case the first step would be to identify where the user pain is acuter. It is also a good idea to identify where strong incumbents are not dominating the market, unless I wanted to compete directly with them. Finally, given that machine learning will be our core technological differentiator, is important to reflect on the strengths of the team and the potential data we can acquire.
Identify the AI Value
Artificial Intelligence is not just another technology stack. When put to use in the right use case and for the right pain-points, AI can dramatically change the job. When machines are collaborating with humans to get things done, their value should not be measured on computational advancement alone. It should be measured by how much AI can help humans achieve greater things.
The next step in the framework is to find the desired outcome for Artificial Intelligence. Using a Value Map, each activity can be outlined to identify the steps and data required to achieve it. Identifying the data is the holy grail of this exercise. Product managers must acquire unique datasets to help differentiate their offerings from competitors. These datasets are essential to identifying what the AI will know and don’t know from the get-go.
Using our job example the first activity is to find restaurants matching the dietary criteria. Humans, in this case, would need to perform the following steps to acquire the information:
1. Search for online listicles from food experts
2. Use yelp to filter options
3. Scan the menu for red flags
4. Call the restaurant to speak with a person about availability and dietary restrictions.
AI can also look at a variety of listicles and narrow down options by reading restaurant menus and yelp’s information on restaurants. The system can then look at the availability using services such as OpenTable, and narrow down options with the best available candidates for the group.
By leveraging AI, users can be confident they are selecting restaurants with the best options for their attendees. The data we have identified here are online listicles, yelp, menus, and OpenTable data.
Design the collaboration between humans and machines
To-be journeys are standard tools for product design teams. With to-be journeys, designers and product managers can articulate the desired experience for their users. In this adapted to-be journey, we can map the experience for users in partnership with the AI. The key element here is to identify opportunities where the machine learning algorithms can continuously learn. This is the major difference between building programmatic applications and AI applications. In the latter, the system is always evolving, it can learn from new data fed into the system. For example, listicles change daily, as well as online reviews. AI also has the opportunity to learn from user behavior. For example, if we choose one restaurant over the other one the system can interpret the action as input that determines user preferences. The next time a similar set of requirements are given by the user, the AI can provide faster, and more accurate suggestions.
End with a project statement
Articulation is the most important skill for any product manager. A project sheet helps you and your team understand the reasons why AI is an indispensable part of your solution. The Project Sheet will highlight the key elements of the project. The problem statement, how AI adds value, the data needed to train a useful machine learning algorithms, and finally the measure of success for the AI.
Note that when defining the scope in the project sheet, I suggest to articulate the final story with a method proposed by the Intercom team: Job Stories. In this format, instead of identifying the user, we give priority to the context.
[Context] __________ [Need] __________ [Outcome] ________
When I need to make reservations for large group of people, I want to select the best option based on all variables so I can impress my friends and coworkers
I look forward to workshopping this framework at The Next Web conference and later this year with educators at the DEL conference. If you have suggestions for improvement, let me know.
Im sharing the templates including the Sketch file if you want to make changes to it.
