Episode Summary: Product Thinking – Episode 261 "AI Implementation in Regulated and High-Trust Industries"

Host: Melissa Perri
Guests: Dr. Mariam Ashuri (IBM WatsonX), Magda de Armbruster (Natural Cycles), Jessica Hall (Just Eat Takeaway)
Release Date: January 28, 2026

Overview

In this episode of Product Thinking, Melissa Perri examines how AI is moving from experimentation into real, production-scale products—particularly in industries where trust, accuracy, and risk management are paramount. Through insights from leaders in healthcare, technology, and consumer services, the podcast explores the organizational, technical, and cultural approaches necessary for implementing AI ethically, safely, and effectively, especially under strict regulation and high user expectations.

Key Discussion Points & Insights

1. Understanding AI Agents and Their Risks

Guest: Dr. Mariam Ashuri, IBM WatsonX
[01:00 – 05:47]

• What Are AI Agents?
  AI agents are systems capable of reasoning, planning, decision-making, and autonomous action.
• LLMs and Their Capabilities:
  • Generative AI, especially LLMs, evolved from content and code generation to performing "tool calling" or "action calling," integrating automation across businesses and legacy systems.
• Limits of LLMs:
  • LLMs operate probabilistically—predicting the next word rather than applying true logic or reasoning (see quote below).
  • This leads to the risk of "hallucinations": highly confident yet factually incorrect outputs.
• Guardrails and Human Oversight:
  • Implementing agentic guardrails (checks for context-relevance and "faithfulness" to source material) helps reduce risk.
  • Human-in-the-loop systems are essential for high-stakes decisions—especially in regulated or sensitive domains.

Notable Quote:
“There is no reasoning really, there is no logic of thinking behind LLMs. ... Because of this, the model can hallucinate ... you can feel the confidence in the tone. But it's not an accurate information. It's just a collection of words nicely put together.”
— Dr. Mariam Ashuri [03:40]

• Calculated Risk:
  • Risk assessment is essential. The higher the risk in an application, the stricter the oversight and guardrails required.

Notable Quote:
"As a product manager you need to know what is non negotiable and what are the risks that can't be jeopardized and just work around that."
— Dr. Mariam Ashuri [05:35]

2. Embedding Regulation and Privacy in Product Development

Guest: Magda de Armbruster, Head of Product at Natural Cycles
[06:17 – 11:00]

• Integrated QA and Compliance:
  • Quality assurance (QA) and compliance teams are embedded from the start, not as post-development gatekeepers.
  • Cross-functional feedback throughout ideation and design minimizes risks and accelerates delivery.
• Data Privacy in Sensitive Environments:
  • As a paid app, Natural Cycles ensures user data is never sold; users have total control, with options like “Go Anonymous Mode” ensuring even the company cannot identify individuals.
  • Data cannot be linked to users—even under legal pressure—addressing rising concerns over governmental access and reproductive privacy.

Notable Quote:
“Our users' data is in their hands... Should women want to go anonymous, they can. And this means that not even us cannot identify who they are. So in an unlikely ... event of a subpoena ... we will not be able to give the users' data because we don't know who the data belongs to.”
— Magda de Armbruster [08:23]

• Regulation as a Framework for Innovation:
  • Strict processes are seen as enablers, not blockers. Well-defined quality management systems streamline new feature development and foster a culture of innovation.

Notable Quote:
“Working together with the regulatory team ... is not a blocker. It just provides us with a framework that enables the innovation.”
— Magda de Armbruster [10:39]

3. Responsible AI Investments and Long-term Capabilities

Guest: Jessica Hall, Chief Product Officer at Just Eat Takeaway
[11:37 – 19:31]

• Cost vs. Impact:
  • AI implementation—especially with LLMs—can be resource-intensive (“training LLMs and running them can be really expensive” [11:41]).
  • Evaluate whether costs match the commercial and customer value, and consider cheaper, simpler alternatives where feasible.

Notable Quote:
“Sometimes with the AI solutions, we can be guilty of over complicating things or over engineering it. And actually simplicity is the answer.”
— Jessica Hall [12:29]

• Simplicity First:
  • Challenge teams to focus on the true problem and the simplest viable solution, avoiding complexity for its own sake.
• Sustaining Governance and Upskilling:
  • AI deployment isn’t just about deploying technical solutions; it means creating departmental capabilities, ongoing training, and supporting future needs.
  • Robust data governance policies, transparency with users (e.g., opt-in for chat data analysis), and building cross-functional oversight teams are essential.
• Bias, Inaccuracies, and Team Diversity:
  • Diverse, cross-functional teams are crucial for detecting and addressing bias.
  • Transparency and rapid, accurate response times help reduce both hallucinations and customer dissatisfaction.

Notable Quote:
“We are building a department that is fit for the future... It isn't just about adopting the technology, it's about creating the entire ecosystem of your department that's able to take it forward.”
— Jessica Hall [15:01]

Notable Quotes & Memorable Moments

• On LLMs Hallucinating:
  “It's just a collection of words nicely put together.”
  — Dr. Mariam Ashuri [03:49]

• On Data Privacy Amid Legal and Political Pressures:
  “We will not be able to give the users' data because we don't know who the data belongs to.”
  — Magda de Armbruster [08:41]

• On Simplicity:
  “Simplicity is the answer. So asking yourself, like, is this the simple answer? Is this the right thing to do?”
  — Jessica Hall [12:29]

• On Responsible AI Adoption:
  “Not getting dazzled by the exciting potential, but actually asking like, what is the problem we're solving and does this really solve that problem?”
  — Jessica Hall [13:20]

• On Diversity and Bias:
  “It is very important to have diverse teams building solutions...especially for AI.”
  — Jessica Hall [17:21]

Important Segment Timestamps

• [01:00 – 05:47] – Dr. Mariam Ashuri breaks down the mechanisms and pitfalls of AI agents.
• [06:17 – 11:00] – Magda de Armbruster discusses healthcare product development, data privacy, and regulatory process.
• [11:37 – 19:31] – Jessica Hall on cost, governance, and building sustainable AI product organizations.

Takeaways for Product Leaders

• AI systems require carefully designed safeguards and oversight—risk rises with the stakes.
• Regulatory and compliance functions, when integrated early, clarify requirements and support innovation.
• Data privacy, especially in high-trust industries, must be architected into products—not just layered on.
• The true ROI of AI investments must consider both obvious tech costs and unseen organizational factors (training, governance, maintenance).
• Simplicity, judgment, and a focus on genuine user/customer problems anchor successful AI adoption.
• Transparent policies and diverse teams are vital for mitigating bias, hallucinations, and other AI-specific risks.

This episode offers product leaders actionable insights for implementing AI confidently and responsibly, balancing innovation with diligence in even the most regulated, high-stakes environments.