🎉 AI Engineers: Join our webinar on Prompt Engineering for AI Agents. Register here >>

May 28, 2024 - last updated
GenAI Rollout Blueprint

AI Rollout Blueprint: Start Small (3/5)

Niv Hertz
Niv Hertz
7 min read Jan 02, 2024

In our ongoing series on the AI Rollout Blueprint, the focus now shifts to the pragmatic “Start Small” step. The ‘Start Small’ step stands out as a crucial phase for organizations looking to implement Artificial Intelligence (AI) in a strategic and manageable manner. 

This article explores the complexities of initiating AI rollouts by pinpointing specific, small-scale challenges within organizations. The aim is to effectively address these challenges within the bounds of current technological capabilities and other limitations. The outcome of this process is a list of targeted, achievable use cases for AI implementation in the short term.

Visit the first two installments in this series to catch up on how we got here: “Start Here”“Think Big”.

The objective of the “Start Small” step

The “Start Small” approach allows organizations to identify achievable use cases for AI implementation in the short term, ensuring that the deployment of AI is strategic, manageable, and aligned with the organization’s resources and technological capabilities. 

This step is essential for organizations to navigate the complexities of AI product launch and maximize the potential of AI in driving business growth and innovation.

Evolution of GenAI

Artificial intelligence has undergone a significant journey, progressing from dedicated copilots to fully autonomous agents. This evolutionary process will help businesses incorporate AI into their operations, providing a gradual transition from assisting AI to operating independently, all while retaining human oversight.

Evolution of GenAI

1. AI Copilot

AI typically starts its journey as specialized LLM-based copilots, serving as assistants for humans. Notable examples include ChatGPT and RAG chatbots tailored for customer support. While these copilots excel at specific tasks and offer valuable assistance, they still can’t handle intricate decision-making processes or function independently.

2. AI Agent with Human-in-the-Loop (HITL)

The copilot evolves into an agent capable of handling a broader range of tasks, with a crucial distinction—human oversight for every decision made by the AI. This stage is a vital intermediary step, allowing organizations to gain confidence and proficiency in AI implementation within a controlled environment before progressing to fully autonomous AI agents.

Responsible AI with Humans in The Loop

Image source

For instance, AI-powered customer support agents like SiteGPT seamlessly integrate AI technology with human input, enhancing customer interactions with a personalized and efficient touch. The HITL approach is implemented to guarantee the AI consistently delivers accurate and fitting responses to customer queries. 

With a human in the loop, the system is overseen, and adjustments are made to the AI’s decision-making process as needed, ensuring a harmonious blend of technological efficiency and human discernment in healthcare customer support.

3. Fully autonomous AI agents

In the final stage of evolution, AI achieves the capacity to operate independently, making real-time decisions without direct human input. Nevertheless, human involvement remains essential. Instead of approving each decision individually, humans oversee the entire system, balancing AI autonomy and human oversight. This approach ensures the AI operates safely and efficiently.

A practical instance of Fully Autonomous AI Agents in healthcare involves deploying AI-powered autonomous agents to enhance patient outcomes. These agents operate independently, making decisions on behalf of healthcare professionals. Their capabilities encompass diverse tasks, including disease diagnosis, treatment recommendations, and patient condition monitoring.

AI evolution means new risk

While the “Think Big” step in our series of the AI Rollout Blueprint may have initially led organizations to consider fully autonomous AI agents for their industry, the importance of starting with AI copilots becomes evident. This evolutionary approach allows organizations to build confidence and expertise in AI implementation, mitigating risks such as data privacy concerns, hallucinations, and security vulnerabilities and ensuring a more controlled and strategic integration of AI technologies. 

By comprehending this evolution, organizations can adopt a phased and pragmatic approach to AI implementation, ultimately maximizing the potential of AI in driving business growth.

Case Study: Navigating AI implementation in healthcare

Illustrating the evolution from vision to practical implementation in the healthcare sector exemplifies the strategic approach to “starting small” with AI.


The envisioned scenario involves a generative AI agent equipped with a comprehensive medical database, delving into a patient’s genetic profile, medical history, and symptoms to propose personalized treatment plans. While ambitious, the journey towards this vision is wisely charted through incremental steps.

In a practical application within healthcare, a tangible instance of the envisioned scenario emerges by creating a GenAI tool. This tool draws insights from an extensive dataset comprising over 200,000 Accident & Emergency (A&E) visits to a bustling London teaching hospital. By incorporating data such as age, test results, and other relevant factors, the tool predicts the likelihood of a person being admitted to the hospital. 

Furthermore, the utilization of GenAI in healthcare extends to identifying individuals with elevated risks of specific conditions, disease diagnosis, and tailoring treatments.

Taking baby steps

Rather than leaping straight to an “Autonomous AI Agent,” the healthcare industry begins with deploying an “AI Assistant.” This initial phase focuses on creating a Q&A assistant for doctors, providing immediate value in aiding medical professionals with information retrieval and basic decision support.

An example is the launch of AI-powered symptom checkers like Symptoma,” allowing users to input symptoms for a step-by-step health issue diagnosis, accompanied by health tips and relevant articles. 

Moreover, these virtual assistants can be tailored to assist doctors in handling the daily operations of their clinics, providing instant value by aiding medical professionals with information retrieval and rudimentary decision support.

Data sensitivity

Recognizing the sensitivity of medical information, the industry adopts a cautious approach. Instead of accessing patients’ genetic profiles or detailed medical histories, the focus is on utilizing public research and clinical trials. This ensures compliance with privacy regulations and establishes a foundation of trust in AI technologies.

For instance, AI-powered Radiology tools analyze medical images to identify abnormalities and aid radiologists in diagnosing conditions. Siemens Healthineers has introduced the AI-Rad Companion3, a sophisticated solution aiding radiologists in the diagnostic process. This intelligent tool automates routine tasks. 

This enables clinicians to dedicate their attention to more intricate cases demanding human interpretation. The innovation streamlines workflows, enhancing efficiency and effectively utilizing radiologists’ expertise in handling complex medical scenarios.

AI-powered Radiology tool

Image source

Human-centric AI Rollout

Shifting away from the prospect of “Customer-facing AI,” the healthcare sector opts for a more human-centric approach with “Employee-facing AI.” Rather than directly suggesting personalized treatment plans to patients, the initial step involves incorporating a human in the loop, ensuring that medical professionals maintain a pivotal role in decision-making.

A great example of a patient-focused strategy in healthcare is witnessed in the application of AI-powered chatbots for mental health support. For instance, Woebot, an AI chatbot, employs natural language processing and principles of cognitive behavioral therapy to extend emotional support and guidance to individuals facing mental health challenges.

By engaging users in conversations, monitoring emotional states, and providing coping strategies, Woebot enhances accessibility and affordability in mental health care. This example shows the healthcare sector’s commitment to a patient-centric implementation of AI, prioritizing responsible and ethical integration to enhance patient care and well-being.

This healthcare case study emphasizes the importance of aligning AI implementation with practical considerations. By starting with manageable steps and prioritizing data sensitivity and human involvement, organizations can lay a robust foundation for integrating AI in healthcare, ensuring efficiency and ethical practices.

Control your AI with Aporia

When building a RAG chatbot, numerous risks exist, such as negative customer experiences due to AI hallucinations, brand damage due to biased language or toxicity, and compliance violations. Aporia is an innovative solution that helps mitigate many risks while accelerating your AI rollout. 

Aporia goes beyond basic firewalls, supporting your needs for secure AI rollout – proactively mitigating hallucinations and blocking prompt injections in real time. Guardrails are seamlessly layered between your LLM and interface, allowing you to control every interaction. This streamlined risk management frees your product and developer teams to focus on new use cases and improving user interaction, building more trust in your AI. 

Ready to take control of your AI? 

Book a demo today to learn more about Guardrails and how it can support your goals.

Green Background

Control All your GenAI Apps in minutes