Course Relevance: This caselet will be useful for BBA, PGDM Analytics course
Academic Concepts: In this caselet, students will be introduced to the basic concept of synthetic data and it’s opportunity and challenges
Teaching Note:
This caselet introduces the concept of synthetic data and its growing importance in market research. It explains how advances in AI, machine learning, and generative AI enable researchers to create datasets that mimic real consumer behavior, supporting activities such as data augmentation, synthetic respondents, and scenario simulation. The note encourages learners to critically evaluate both the benefits of synthetic data, including cost savings, scalability, flexibility, and privacy, and its limitations, particularly its inability to generate entirely new consumer insights. The discussion emphasizes that synthetic data should complement, rather than replace, real-world data collection when making marketing and business decisions. It also discussed on opportunity and challenges of using synthetic data.
.
Learning Objectives:
By the end of this caselet, students should be able to:
- Define synthetic data and explain its role in modern market research.
- Differentiate between real data and synthetic data and understand their respective uses.
- Describe technologies such as Machine Learning, GANs, VAEs, and LLMs used for synthetic data generation.
- Identify key applications of synthetic data including data augmentation, synthetic respondents, and scenario simulation.
- Evaluate the benefits of synthetic data in terms of cost reduction, scalability, flexibility, and privacy.
- Assess the limitations of synthetic data, particularly its inability to generate entirely new consumer insights.
- Critically analyze when synthetic data should complement rather than replace primary research.
- Recommend appropriate use cases of synthetic data for market research and business decision-making.
Introduction:
In market research we are using surveys, interviews, focus groups, customers’ purchases and observations to understand consumer behavior. These techniques remain very effective and still relevant today, but these are costly and time-consuming. It may also face some difficulties in reaching some particular audiences.
In recent times with the progress in artificial intelligence (AI), machine learning(ML) and specifically in Generative AI, it is very much possible to create synthetic data, which would help the researcher to develop model, analyze market scenarios and conduct analysis based on synthetic data. Synthetic data is not any random or fake data but it contains behaviour of actual data. It can be used for hypothesis testing or analysing some niche markets without the need to collect large amounts of primary data all the time. Based on the available real data more data can be created which will be like the real data.
One of the important point to check whether artificially generate data contain any valuable information to make business decisions before using the same. Researcher should consider the use of synthetic data in market research as a valuable instrument for conducting research, and not only as a tool to save money.
The usage of synthetic data is increasing day by day due to it’s scalability, flexibility and privacy.
Background of Synthetic Data:
Synthetic data refers to the data on a particular subject/topic which is synthesized with the purpose of emulating certain statistical features, relations or behaviour of real datasets. It is not any random data.
Synthetic data can be created with the help of machine learning models, generative adversarial networks (GANs), variational autoencoders (VAEs) and large language models (LLMs) based on the real dataset feature/behaviour.
According to Gartner’s prediction, AI-generated synthetic data will eventually capture vast majority share of data by 2030. Synthetic data can exist in various forms such as in market research.
| Synthetic data generation |
Some of the approaches of generating Synthetic data:
| Data augmentation |
| Synthetic respondents |
| Scenario simulation |
1.Data augmentation: increasing the sample size of the analytical data by using additional observations with the help of available real data set.
2.Synthetic respondents: This can be used to generating consumer responses to questions of preliminary study. It will reduce effort require for survey.
3.Scenario simulation: This can be used to analyzing hypothetical scenarios, What if – analysis like : What would happen to sales if price is changed for a product.
There are limitations of using LLM generated samples in consumer and marketing research which was explored by Sarstedt et al. (2024). The review draws attention to the potential role of synthetic respondents in the preliminary research and emphasizes the contextual dependency of their reliability.
One point researcher should note that with synthetic data, existing knowledge can be used to prolong its validity but it does not provide new evidence of consumer behavior. For new evidence real data is important.
Opportunities:
- Faster and more cost-effective way of doing research
Traditional survey-based research will involve recruitment, distributing questionnaires, providing incentives, field work and data cleaning. These requires investment and it is also time consuming. However usage of Synthetic data can reduce the same. Synthetic data can speed up initial analysis, help create test datasets and enable researchers to test multiple hypotheses before investing in costly primary research.
Example: A consumer goods company might use synthetic data to test out a number of different model and hypothesis, and researcher can avoid conducting a more limited consumer survey.
This will help to reduce research costs. But human validation at crucial points is very important.
- Exploring customer segments which is difficult to reach:
For reaching hard to reach customer segment it may require reach specialised respondents, such as enterprise purchasing managers, rare-product users or customers who have special buying experiences.
- Improving customer segmentation
We can use synthetic data to test customer classification model and clustering algorithms.
Similar approach could be used by retailers, digital platforms and banks to trial segmentation strategies and not use them on real customer data. However final segments can be tested with real world evidence. This will reduce overall time and cost.
- Testing marketing strategies
Marketers can experiment with different pricing models, promotions, product features, distribution channels etc by using synthetic data, which can be validated with real time data during final implementation.
Using synthetic data organization might analyse two or more campaign ideas, determine if there are any shortcomings in the campaign planning, and use them in order of preference to test in the real world.
- Ensuring Data privacy
Due to data privacy organisations may be reluctant to provide customer-level data.In those cases we can use synthetic data. if appropriately developed, can help in analytical experimentation and machine learning model training, with out sharing actual data of employees.
Pls note Safeguards of privacy and disclosure-risk assessment are still required.
Challenges of Synthetic data usage:
- Data accuracy and representativeness
The synthetic data has to represent the properties of the target market, otherwise, it won’t be useful. It should inherit properties of the real survey data points. The problems associated with lack of representativeness may persist in the synthetic data generation if the initial dataset used to generate synthetic data does not represent the target segment example, rural consumers or older consumers, lower income group which might be needed for particular research or to train particular machine learning model.
It may create an illusion of certainty. Although the bigger synthetic dataset is statistically valid, it is wrong to assume that the larger synthetic sample size equals the bigger market evidence, it may lack real time data properties.
- Bias and distorted consumer preferences
The patterns learned by the AI are consistent with the assumptions of the model itself or data which is used to train the model. If there is any inherent bias in training data or in algorithm itself then synthetic data can be biased as well. Therefore, there can be some stereotypes communicated and trending consumer behaviour can be overlooked.
Example: A fashion store may be looking to test advertisements of a new line of clothes. If the machine learning model is trained mainly on the purchasing history of a particular type of data set, then there can be no flexibility and only very specific preferences can be recognized. Training data set needs to be diverse and should have representation for all possible segment.
- Privacy and data protection
Though it is advantage of using synthetic data because it allows minimizing exposure to sensitive consumer data. Instead of sharing the identifiable records for any consumer, the researchers would be able to use the synthetic datasets. However, it is necessary to remember that the synthetic data is not always anonymous and risk-free because synthetic data generation technique is based on closely resemble rare records from the source dataset, then sensitive information can be revealed, which may have cause adverse effect to organization.
Point to be noted, privacy of data depends on the type of synthetic generation, the initial input data, who gets access to the data, and how the data was protected.
US Census Bureau emphasizes the same in its research of the synthetic microdata (Stanley & Totty, 2024).
- Validation of synthetic insights
The synthetic dataset can be a good reflection of average characteristics of the customers. It may not, however, to reflect actual relationships between factors of the real dataset. Testing it with the real user data is vital.
Example: It will be a good representation of the overall brand awareness level but will distort the connection between brand loyalty, price sensitivity and purchase intent.
- Over-reliance on Synthetic data
Over dependent on Synthetic data may not be good idea. The synthetic respondents can give realistic explanations, preferences and feedbacks. But survey is not necessarily an authentic human opinion. If an organization completely replaces interviews and surveys with AI-generated replies, then there is a risk of mistaking plausibility of the model for consumer reality, which will affect decision making process later.
Conclusion
Synthetic data is becoming a important tool within market research, as it enables speeding up experimentation process and reducing over cost of the project. This definitely offers new possibilities for analysis and segmentation and solving problems related to data accessibility.
Key advantage of this tool it creates big volumes of data mimicking real data to test any particular model or hypothesis. It also helps researchers to ask better questions, and make right decisions.
On disadvantage part synthetic data might lead to biases, distort consumers’ preferences and/or fail to reflect changes in the market. Consumer insights cannot be confused with realistically looking outcomes. Biased synthetic data may leads to wrong management decision and failure of any process or project.
One of the questions users of synthetic data should answer – Are we using synthetic data together with traditional methods of research and validate it independently?
Companies, which will find the right balance between these tools, will improve their market research capabilities through accelerated and flexible research process without compromising its quality. Hybrid approach which use actual data to generate and validate synthetic data should be encouraged.
Finally, synthetic data must be used to ask better questions and leverage evidence for market researchers and not to substitute consumers whose behaviour we need to understand.
References
- B2B International. (2025). The role of synthetic data in B2B market research. https://www.b2binternational.com/wp-content/uploads/2025/04/The-Role-of-Synthetic-Data-in-B2B-Market-Research-B2B-International.pdf
- Sarstedt, M., Adler, S. J., Rau, L., & Schmitt, B. (2024). Using large language models to generate silicon samples in consumer and marketing research: Challenges, opportunities, and guidelines. Psychology & Marketing, 41(6), 1254–1270. https://doi.org/10.1002/mar.21982
- Stanley, J., & Totty, E. (2024). Synthetic data and social science research: Accuracy assessments and practical considerations from the SIPP Synthetic Beta (Working Paper No. CED-WP-2024-004). U.S. Census Bureau. https://www.census.gov/library/working-papers/2024/adrm/ced-wp-2024-004.html
- Kalidas, A., & Ostler, J. (n.d.). Synthetic data: The real deal? The opportunities and challenges of synthetic data for market research. Kantar. https://www.kantar.com/inspiration/ai/synthetic-data-the-real-deal
- https://www.gartner.com/en/newsroom/press-releases/2025-11-10-gartner-survey-finds-artificial-intelligence-will-touch-all-information-technology-work-by-2030
