Clinical AI Notes

I began with a deep dive into the competitive landscape, analyzing leading players in the space like Abridge, Notable, and Suki. These companies were leveraging voice-based AI to capture and structure clinical conversations, but their footprint was largely US-focused.

• In Europe, and specifically within the NHS ecosystem, there was a noticeable gap: no dominant player was structuring conversational data at scale.
• This presented an opportunity to be first to market in the EU with a solution already validated by competitors in the US.
• Furthermore, NHS Trusts were actively seeking ways to reduce administrative burden—a direct alignment with what AI-driven documentation could solve.

Strategic Decision: Build a solution that captures unstructured clinical conversations and transforms them into structured notes, addressing a real pain point while positioning us as market leaders in Europe.

One of our most significant assets was our existing network of 130,000 clinicians across 200+ hospital systems. This represented a treasure trove of communication data—real-world clinical conversations occurring daily on the platform.

• If we could secure Data Processing Agreements (DPAs) with Trusts, we could mine this data to identify the most valuable clinical pathways for automation and training.
• This would allow us to pseudo-anonymize data and use it for synthetic model training, creating a feedback loop to improve our AI models.
• Unlike our US-based competitors, we could train models on real European clinical interactions, giving us a localized advantage.

Strategic Decision: Leverage existing communication flows to build ambient listening and post-call transcription, allowing for seamless documentation without disrupting clinician workflows.

We had three clear business goals and this product aligned perfectly with all three:

• Increase Engagement: Clinicians could effortlessly turn conversations into clinical notes and share them, increasing their interaction with the platform due to the network effect.
• Expand Reach: Trusts saw the value in streamlined documentation, making them more willing to convert from free to paid and historical data showed once a Trust paid for Pando and made it part of it's standard operating procedures, user acquisition increased 10 fold.
• Capture Valuable Data: Every note created was structured, indexed, and primed for pharma real world evidence insights and specific AI-driven care pathways that we identified early on (advice & guidance and referrals between primary care and visual specialists and emergency medicine to various specialists)

Strategic Decision: Prioritize specific hands-on Trust rollouts first to validate clinical value, followed by Pharma partnerships once the structured data pipeline was proven.

• Original Plan: Real-time transcription of audio and call conversations.
• Pivot: Shifted to post-call transcription through ambient listening, focusing on capturing key moments after conversations concluded.
• Why: Doctors wanted more control over what was captured, and real-time processing was cost-prohibitive.
• Business Goal: Increase Engagement – By making it easy for doctors to capture key clinical moments without friction.
• Result: We leveraged existing call and voice data flows already happening in Pando, creating structured notes effortlessly without user disruption.

• Original Plan: Launch with fine-tuned Llama and various Amazon Bedrock models for higher accuracy.
• Pivot: Opted to deploy a foundational model with narrow prompting instead of fine-tuning.
• Why: Fine-tuning required substantial investment and time; we needed to validate the concept first.
• Business Goal: Increase Valuable Data Capture – Launch quickly, capture structured notes, and validate user demand before scaling.
• Result: We rapidly tested the core functionality, gathered real-world data, and proved the need for structured conversation capture.

• Original Plan: Parallel development targeting both Trusts and Pharma.
• Pivot: Focused initially on Trusts, using structured notes to reduce admin time and increase adoption.
• Why: Trusts had existing relationships with Pando, making it easier to deploy and validate the solution.
• Business Goal: Expand Reach – Move from a free product to a paying model by solving admin challenges for Trusts.
• Result: Demonstrated direct value to Trusts, setting the stage for future pharma monetization by capturing high-value clinical data.

• Original Plan: Build a full-scale MVP with live MDT handoffs and automated referrals.
• Pivot: Reduced scope to core conversational capture and post-call summaries.
• Why: Resource constraints and the need for rapid testing to attract investors.
• Business Goal: Increase Engagement & Valuable Data Capture – Validate the concept with the smallest shippable product.
• Result: Shipped a prototype that investors could use directly, leading to deeper conversations around funding and rollout.

• Original Plan: Monetize through direct payments from Trusts after rollout.
• Pivot: Introduced 5 free notes per user during the pilot, with Trust-level authorization required for further use.
• Why: Reduce friction for initial adoption and demonstrate value before asking Trusts to commit.
• Business Goal: Expand Reach – Pilot success would prove the concept and unlock larger Trust contracts.
• Result: Created a low-risk entry for Trusts, enabling us to scale the sales process more effectively.

• Original Plan: Use existing data streams for model training.
• Pivot: Shifted to a DPA-first approach—ensuring data processors could only train models with Trust consent.
• Why: GDPR regulations made it necessary to get explicit agreements to access clinical data.
• Business Goal: Capture Valuable Data – Create a compliant pathway to leverage conversational data for AI training and pharma insights.
• Result: Positioned us as data-compliant and privacy-focused, giving us leverage in Trust negotiations.

• Original Plan: Deploy the prototype across Pando's current infrastructure.
• Pivot: Realized the legacy system couldn't handle real-time conversational AI; we paused for a complete rewrite.
• Why: Scalability and latency issues required a stronger foundation to support AI-driven documentation.
• Business Goal: Increase Engagement & Valuable Data Capture – Ensure the infrastructure could scale with demand.
• Result: Though we paused full deployment, the prototype was instrumental in investment pitches and user testing to refine our vision.

• Original Plan: Leverage existing Pando communication data for model training.
• Pivot: Opted for pseudo-anonymization and synthetic data generation to comply with Trust agreements.
• Why: NHS Trusts are the data controllers, and we needed explicit permission to access data.
• Business Goal: Capture Valuable Data – Build a sustainable pipeline for model training without legal exposure.
• Result: Set the stage for ethically compliant data usage, allowing us to train models on real clinical interactions.