In the realm of start-ups, likening the pursuit of product-market fit to the ultimate challenge, or “final boss,” is a fitting analogy. In this context, product discovery parallels pursuing the main quest in open-world video games like Breath of the Wild, Grand Theft Auto, or Skyrim, where players can freely explore vast virtual landscapes at their discretion. Just as players must maintain focus on the main quest and resist the allure of side missions to ultimately confront the final adversary, our journey in product discovery similarly requires unwavering focus to reach our ultimate goal – achieving product-market fit.
Popular product discovery methods apply different processes but share one common trait: learning, rapidly testing assumptions, and iterating on ideas. These product discovery methods are optimised for B2C domains where access to users is plentiful, prototyping is straightforward, and release cadences are frequent. In our B2B domain, access to users (pathologists) takes time; prototyping is more complex, and release cycles are longer.
Franklin.ai’s product team has adopted practices that work effectively within these constraints by:
- Investing in domain awareness and education for the team
- Clearly defining and ruthlessly prioritising questions we want to answer
- Aligning prototypes and protocols to specific questions
- Accessibility and transparency of ideas, learnings and decisions
Investing in domain awareness and education for the team
Tutorials are ubiquitous in video games because they teach players the rules and controls. On the surface, rules and controls represent constraints. However, it is an understanding of these constraints that allows players to express themselves in creative ways. Similarly, understanding the ‘rules’ and ‘controls’ associated with histopathology allows creative exploration of solutions.
Franklin.ai is exceptionally fortunate to have access to some of the world’s most pre-eminent pathologists. Their profound knowledge has played a pivotal role in providing expert clinical guidance to our clinical AI efforts and in simplifying the comprehension of histopathology for our teams. Clinical Associate Professor Fiona Maclean, renowned uropathologist, delivers lectures to our team enabling us to grasp the complexities of the domain and envision the enhancements that software and AI can bring to the field of pathology.
We also encourage team members to shadow pathologists. Nothing can replace the learning via osmosis associated with being immersed in the same environment as your users. We make this as non-intrusive as possible for pathologists by observing their work from a distance and only asking questions between cases to avoid interrupting their workflow.
This helps us understand five key areas:
- Physical environment of work (i.e., laboratory or similar)
- Information they use to achieve their goals and where they get it
- Current workflows and workarounds
- Tools they use, and how they use them
- Types of interactions they have with their colleagues
These foundational learnings have helped spark our collective imagination for how software and AI can help pathologists become more effective and efficient.
Clearly defining and ruthlessly prioritising questions we want answers to
The non-linear, sandbox nature of open-world games can feel overwhelming because too many choices exist. Similarly, product discovery in its early stages feels overwhelming due to the seemingly endless options teams can pursue. Product discovery team members will inevitably have thoughts about the direction of the product. These can take the form of ideas, hypotheses or assumptions. Whilst these words aren’t synonyms, product teams often use them interchangeably. This causes issues because of the implicit questions behind them:
- Is this a good or bad idea?
- Is this hypothesis supported or not supported?
- Is this assumption true or false?
The loaded nature of these words can cause friction and misunderstandings within teams. An approach that has helped cut through these issues is simply using well-articulated questions. As Erica Hall so eloquently puts it in her book “Just Enough Research”:
Research is simply systematic inquiry.
Product discovery is also a process of systematic inquiry. Some questions we’ve posed along the way have included:
- How should we handle situations where pathologists disagree with the AI’s findings?
- In what ways can we assist pathologists in confirming the accuracy of the AI’s conclusions?
- Is it preferable for pathologists to review the AI’s findings before or after examining the data (slides) independently?
- Should we convey the AI’s level of confidence in its findings, and if affirmative, what is the best approach to do so?
Due to a multitude of inquiries and a restricted amount of time available from pathologists, we’ve had to ruthlessly prioritise which questions need to be answered now, and which we can come back to later. We typically prioritise which questions we want answers to based on:
- The range of potential answers (how many significantly different answers are there to this question)
- The impact of a given answer downstream (e.g. workflow questions have far-reaching consequences so are more likely to be prioritised over an instance of content understandability, which is relatively self-contained)
Healthy product discovery debates don’t just centre around ideas, hypotheses or assumptions. They also involve identifying the right questions to ask at any given point in time. These questions become a lexicon and allow us to (relatively) objectively make progress as answers are systematically generated over time.
Aligning prototypes and protocols to specific questions
The early stages of product discovery typically require us to ask and answer questions about concepts and flows. At this point in time, uncertainty is highest, and it makes the most sense to seek answers by prototyping concepts that cover a breadth of ideas. These prototypes will have implicit answers to ‘big ticket’ questions relating to workflow, information hierarchy and key features. Breadth allows us to have “compare and contrast” based discussions with participants (pathologists). These are especially valuable in a novel area like medical AI, where users don’t have an entrenched mental model of how AI should work. Once we reach an informed conviction about the desirability and usability of these ‘big ticket’ items, we can progress to answering questions relating to more detailed design considerations, such as screen layouts and supporting features.
Accessibility and transparency of ideas, learnings and decisions
If product discovery is the main quest, then there are the 4 key milestones or missions to achieve:
- Defining our learning objectives and methods
- Creating the protocol and prototypes to gather the data
- Identifying themes and patterns in the data
- Agreeing on next steps
The product team regularly communicates progress against these milestones with the rest of the organisation. Our “communication stack” is made up of:
- Slack for asynchronous updates
- Recorded videos posted to Slack where voice-overs from members of our team provide additional context
- Figma where our prototypes and concepts live
- Dovetail as our research repository
This approach enables our colleagues to access a comprehensive range of information, spanning from concise updates in bullet-point format to in-depth examinations of design files and ‘raw’ research findings. As a result, we maintain strong alignment despite being geographically dispersed across Australia. Most significantly, it bolsters our confidence and allows us to progress with product discovery with greater confidence.
Although we’re yet to complete the main quest, our product discovery journey is well underway, with multiple missions done.
As we continue our product discovery journey, much like dedicated gamers, we remain committed to maintaining focus on our main quest: achieving the ultimate product-market fit. In this pursuit, our product discovery strategies and continuous communication serve as our trusty tools and companions, guiding us ever closer to our ultimate goal.
Lan Huang is a UX researcher and designer. He works with Franklin.ai’s product team to understand the relationship between pathologists, software and AI.
If the work we’re doing sounds interesting to you, we’re always on the lookout for new talent to join us. Check out our latest career opportunities here or read more about the key challenges facing the field of pathology, and how AI can help.