Saturation score. How do we know how many interviews to run?
Determine your interview target for achieving topic saturation using our efficient approach, leveraging the historical wisdom of research pioneers. This method ensures deep insights with theoretical sampling at its core.
Right now you need to define the number of interviews you wish to run.
What we describe in this post is a method whereby our agents will run the necessary interviews to achieve topic saturation. It’s a much more cost effective solution and speaks to the core of what we’re building.
A bit of history
Foundational Contributions:
It's the swinging '60s, and Barney Glaser and Anselm Strauss are about to drop a research bomb with their book, "The Discovery of Grounded Theory." Imagine them as the dynamic duo of qualitative research, turning the tables on the stuffy old ways and saying, "Hey, let's make theories straight from the data!"
The '90s Revolution:
Fast forward to the '90s, and Janice M. Morse steps onto the scene with her masterpiece, "Critical Issues in Qualitative Research Methods," a treasure map guiding researchers on how to pick participants and decide on the size of their study group with style and substance. Janice was all about keeping things legit and making sure everyone played by the rules of good research.
21st Century Shake-Up:
As we zoom into the 21st century, enter Kathy Charmaz, wielding her "Constructing Grounded Theory". Kathy's all about building theories from the ground up, using real data and constantly revisiting her ideas. It's like she's saying, "Let's keep digging and refining until we strike gold." She brought a fresh, constructivist vibe to the party, making sure everyone understood that creating theories is a journey, not just a one-stop destination.
Theoretical sampling drives our approach
Theoretical sampling is a strategy in qualitative research, particularly in grounded theory, where the selection of participants, cases, or data to analyze is driven by the developing theory.
When we say that analysis is "driven by the developing theory" in the context of qualitative research, especially within grounded theory methodology, it means that the process of data analysis is closely integrated with and informed by the evolving theoretical insights that emerge from the data itself. In other words, it’s an ongoing conversation, as it should be.
Unlike probability sampling in quantitative research, which aims for generalisability through random selection, theoretical sampling focuses on deepening the emerging theory's conceptual development.
In order to fulfil our vision of Continuous Insight, we need to aim towards saturation and branch out until we hit saturation again. Imagine agents having multiple conversations that will end when a saturation threshold is met.
Getting to a saturation number
Yes, you can use OpenAI's API (or any other LLM) to analyze interview transcripts and estimate a saturation score, although the process will be somewhat indirect. The LLMs’s don't directly compute a saturation score as it's a specific application not built into the model. However, it’s fairly straightforward to design a workflow that leverages the model's capabilities in text analysis, summarization, and topic extraction to approximate a saturation score.
Here's how we do it (this is a simplified version):
Step 1: Preprocess and Summarize Interviews
First, our interview transcripts don’t need to be clean and organized. You might want to summarize each interview to focus on key points, especially if they are lengthy, but that’s it. We do this.
Step 2: Extract Topics from Each Interview
Next, we need to identify and list the main topics or themes from each interview summary.
Simple prompt for topic extraction:
"List the main topics or themes discussed in this interview summary:"
[Summarized Interview]
Step 3: Aggregate Topics and Identify New Ones
Aggregate the topics from each interview and keep track of new topics that emerge with each subsequent interview. This step might be more manual or semi-automated, depending on your setup. You’re basically creating your codebook, as they say in qualitative research. This codebook is the basis for this algorithm.
Step 4: Calculate Saturation Score
The saturation score can be approximated by monitoring the rate at which new topics emerge as you process more interviews. A simple approach is to calculate the percentage of new topics in each subsequent interview compared to the total topics identified so far.
Step 5: Automate the Process
To automate steps 2 and 3 we created a script that sends each interview (or summary) to the API with the prompt for topic extraction. Then, the script can compare the topics from each interview to those from previous interviews to track new topics.
Example Python Script Outline:
import openai
openai.api_key = 'your_api_key_here'
def extract_topics(interview):
response = openai.Completion.create(
engine="input your engine here",
prompt=f"List the main topics or themes discussed in this interview summary:\\\\n{interview}",
temperature=0.5,
max_tokens=100,
top_p=1.0,
frequency_penalty=0.0,
presence_penalty=0.0
)
return response.choices[0].text.strip().split('\\\\n')
# Assume `interview_summaries` is a list of your interview summaries
all_topics = set()
new_topics_per_interview = []
for summary in interview_summaries:
topics = extract_topics(summary)
new_topics = set(topics) - all_topics
new_topics_per_interview.append(len(new_topics))
all_topics.update(new_topics)
# Calculate saturation score as needed, e.g., by analyzing the trend in `new_topics_per_interview`
Step 6: Interpret the Saturation Score
A decreasing trend in the number of new topics per interview indicates approaching saturation. When the number of new topics drops to zero or near-zero over several interviews, you've likely reached saturation.
The additional logic involves creating a function to calculate the saturation score. This function takes as input the list of new topics per interview and outputs a score representing the degree of saturation. We define saturation as the point when less than 5% of the topics are new. Here is an example of what this function in Python:
def calculate_saturation(new_topics_per_interview):
total_topics = sum(new_topics_per_interview)
saturation_points = [new_topics / total_topics for new_topics in new_topics_per_interview]
saturation_score = 0
for point in saturation_points:
if point < 0.05:
saturation_score += 1
return saturation_score / len(saturation_points)
This function calculates the proportion of new topics in each interview and checks if it's less than 5%. The saturation score is then the proportion of interviews that meet this criteria.
Releated Articles
More articles for you
The Lie We Tell Ourselves About Customer Research
Most research asks what people say. The problem is people don't do what they say. This piece breaks down the gap between stated and revealed preference — and why behavioral modeling, not better interviews, is how you close it.
Two ways to run research with Synthetic Users and why the difference matters
Iris, what is the difference of using agents to accelerate research.
Synthetic Users vs digital twins
You don’t need a twin for “a parent in rural Ohio who shops weekly at Walmart, prefers fragrance-free, and has a toddler with eczema.” You sample a parent profile with relevant traits and constraints, add retail and dermatology context, and generate behaviors consistent with both.
Two major papers. One shared direction.
LLM-powered Synthetic Users have crossed from concept to validated method. This proves they can predict human behavior accurately, letting teams run fast, low-cost behavioral experiments without replacing real participants.
Gartner says we lead. That's kind of them.
Gartner’s latest report on AI-powered synthetic user research cites Synthetic Users as a leader.
Introducing Shuffle v2
Shuffle v2 is a feature that intelligently shuffles between multiple large language models via a routing agent to produce more realistic, diverse Synthetic Users with better organic parity.
Chain-of-feeling
Synthetic Users use a “chain-of-feeling” approach—combining emotional states with OCEAN personality traits—to produce more human-like, realistic user responses and yield richer UX insights.
Generative Agent Simulations of 1,000 People
A paper that thoroughly executes a parity study between Synthetic and Organic users.
21 Peer reviewed papers that support the Synthetic Users thesis
Here is a compilation of all the papers that help make a case for Synthetic Users.
Why we shuffle between models — to ensure both parity and diversity!
Synthetic Users balances aligned and unaligned models to maintain diversity and authenticity in simulated users while ensuring ethical standards and user expectations are met.
Latest press articles for Synthetic Users
Synthetic Users and AI are transforming research methodologies, offering innovative, cost-effective alternatives to traditional human subject studies.
Comparison studies. The opportunity lies in the deviation.
When we compare different studies, especially looking at what synthetic (artificial interviews) and organic (real-world interviews) data tell us, we often find they mostly talk about the same things but there's also a bit where they don't match up. This gap is super interesting because it's like finding hidden treasure in what we thought we knew versus what we might have missed.
How we deal with bias
Harnessing the power of AI in our Synthetic Users, we strive for a balance between reflecting reality and ethical responsibility, ensuring diversity and fairness while maintaining realism.
The transition to Continuous Insight
The transition towards Continuous Insight™ aligns research activities more closely with the dynamic needs of the business and ensures that product development is continuously informed by up-to-date user insights.
The Art of the Vibes Engine
Large language models (LLMs) like GPT-4 serve as powerful "vibes engines," empathizing with diverse groups and generating contextually relevant content. Their applications span market research, customer support, user experience design, and mental health support, offering invaluable insights and personalized experiences. While not infallible sources of truth, LLMs enable creativity, personalization, and connection within the realm of human language.
There is a faster and more accurate way to do research. Use Synthetic Users.
How Synthetic Users is changing the research process.
The wisdom of the silicon crowd
In the light of an ancient parable, we explore a new paper that dives into how ensembles of large language models match the prediction accuracy of human crowds. It reveals that combining machine predictions with human insights leads to the most robust forecasting results.
Three research papers that helped us build ❤️ Synthetic Users
For the sceptics amongst us who need more tangible research in order to engage with this brave new world. Full disclosure: we are part of the sceptics.
What is RAG and why it’s important for Synthetic Research
Ahead of our RAG launch we explain Retrieval-Augmented Generation (RAG) and how it enhances Synthetic Users by providing increased realism, contextual depth, and adaptive learning, with profound implications for market research, user experience testing, training, education, and innovative product development.
Synthetic Users system architecture (the simplified version).
Foundation models underpin Synthetic Users with advanced capabilities, enhanced by synthetic data and RAG layers for realism and business alignment, all within a collaborative multi-agent framework for richer interactions.
Saturation score. How do we know how many interviews to run?
Determine your interview target for achieving topic saturation using our efficient approach, leveraging the historical wisdom of research pioneers. This method ensures deep insights with theoretical sampling at its core.
How Synthetic Users are gaining depth
Synthetic Users are evolving to address criticism about their generalist nature by incorporating representative data sets and personal narratives.
How we compare interviews to ensure we improve our Synthetic Organic Parity — 85 to 92%
How do we know we are right? How do we know our Synthetic Users are as real as organic users? We compare.
Synthetic Users: Merging Qualitative and Quantitative Research, in seconds.
At Synthetic users we are blurring the lines between qualitative and quantitative research. Here's how we are going about this transformative approach.