Another example of using synthetic data in product development? Building a robot.
“Mobile engineering has been getting a lot better these days,” said Agustin Huerta, SVP of Digital Innovation at software development company Globant. There is a virtual environment like Nvidia Omniverse, where simulated robots interact with simulated objects, create a large amount of training data, and start the robot's ability to navigate space and process products.
“And when you're talking about computer vision data to train an autonomous driving solution, you need synthetic data. There's no other way to do that,” he says. “If not, you'll need to crash the car.”
5. Explore new markets without past data
Another use case for synthetic data is when a company has a product but wants to sell it in a new market. Companies can model how consumers behave, what they like, and how they respond to new products and services, says Thota. You can also use simulated data to help refine your features and marketing strategies.
“Banks considering entering new regions can use synthetic data to simulate the economic situation in the region, spending habits and how people adopt their financial products,” he adds.
Anandrao, an AI professor at Carnegie Mellon University, once worked with a ride-sharing company to expand into new markets. However, because the conditions are geographically different, using the same strategy everywhere would not have been very effective.
“In New York City, you need a five- to ten minute turnaround,” says Rao. “I say eight minutes, but they're not tolerant of misconceptions, as they take 12 minutes to get the car. But in Ann Arbor, Michigan, if you're a few minutes late, they can live with it.”
In other words, optimization strategies are necessary for differentiation, and synthetic data helped to improve these strategies.
“Ten cities had more than 200,000 market scenarios,” he adds. This gave executives real insight into how to adapt to new markets.
6. Building digital twins
Historically, digital twins have been used to model jet engines, assist businesses in predictive maintenance, and to design and manage factories and other complex physical facilities. Today, the definition of digital twins is expanding to include software systems, business workflows, and even people.
According to Tom Edwards, American consumer AI leader at EY, businesses simulate customers, behavior, shopping trips, buying patterns and how to respond to specific promotions. This is done by creating a synthetic customer profile. “It helps us understand how different demographics respond to different product positioning,” he says. “And what we're coming up is better demand forecasting and better targeting.”
And he sees companies using synthetic personas instead of focus groups.
“You can create hundreds of personas and test different messaging,” he says. “The synthetic data allows you to fill in the details of your psychographic.”
These simulated personas can also be used to improve e-commerce personalization.
“I can run millions of different combinations. When it's time to shop, I can quickly match you based on one of these pre-configured personas built on synthetic data,” he adds. “I know you better than traditional algorithms because traditional algorithms have already advanced millions of potential paths.”
The business value here could be millions of dollars as it unlocks ways to seamlessly align with consumers and deliver recommended products they have never seen before. Companies can also create digital twins for their employees.
“Internally, one of the things we see is our staffing and skills,” said Nick Kramer, leader in applied solutions at management consulting firm SSA & Company.
“There's historical data about consultants and unreliable data about skills and abilities,” he says. “But we have a wealth of project data. Then there's a clay chunk, so to speak, and we're experimenting with different ways of synthesizing the data.”
A synthetic persona could be a person, a project role, or a specific title, he says. These are combined into a simulated project team, which creates an perspective on how staff will look, how to balance skills and tools, and how to optimize for results, speed, revenue, and margins.
7. Preparing Agent AI
As AI evolves, there are opportunities to use synthetic data. For example, this year is about Agent AI.
According to a Cloudera survey in April, 96% of corporate IT leaders plan to expand their use of AI agents in the next 12 months. And 57% say they already have AI agents implemented, but the only biggest barrier is data privacy, with 53% saying they are slower adoption. However, when it comes to training AI agents, it's not just about maintaining privacy.
“Synthetic data is a great way to accelerate learning for these agents and map complex scenarios,” says EY's Edwards. It can also be used to allow agents to handle everything thrown.
“If we can run millions of different scenarios based on complex interactions, it can be an invaluable tool,” he says. “It will be a fundamental aspect of how agents are deployed within an organization.”
Reality Check: Risks of Relying on Synthetic Data
There is also the risk of exaggerating the synthetic data. Panetta has its limits, as he discovered when he tried to create a composite image of a person wearing a face mask.
“If abused, we risk comparable to an overfitting problem that will result in a very repetitive output,” says Gordon Van Huizen, SVP of AI platform company Mendix. “The next step is to provide prompts outside the training data, which can result in random or strange results, as the system is difficult to interpret the new patterns.”
However, there is a way to deal with this. Companies can create more diverse datasets, blend synthetic and real data, and add noise to the data to create outliers.
“But the key to leveraging synthetic data is to always include human verification protocols whenever possible,” he says.
