Was this email forwarded to you? Get Behavior Analytics​

---

AI is your biggest competitor.

Whenever you use ChatGPT, CoPilot, Cloude.ai, or other intelligent agents, you teach AI how to “steal” your intellectual property.

Another way to think about AI is as your partner. Every relationship has a give-and-take and a good enough reason for you to stay. One reason for store optimizers is the tools that generate synthetic data.

Because,

A significant challenge to being an experimentation practitioner in physical retail is data.

Specifically,

The low volume of data and the complexity of human behaviors in physical retail prevent you from declaring test success based on statistical significance.

Thanks to Ton Wesseling, the Experimentation Guru, who pointed it out in the first minute of our conversation in Amsterdam (the day before Covid closed international travel in February 2020😂).

There are different ways to solve that data challenge. Today, I’d like to introduce you to Synthetic Data.

🤔What's *Synthetic Data*?

Synthetic data refers to artificially generated data.

It is often generated from a sample of real-life data without the specifics of actual personal information.

The most typical usage of synthetic data is training AI systems.

The obvious application of synthetic data is in market research.

However, there was little enthusiasm from approximately 200 market research professionals. (Source: Cint.ai)

That said,

Using synthetic data in physical retail has distinct benefits for zero-party research.

😀Behavior Analytics of *Synthetic Data*

Synthetic data is used in various ways in retail, including improving personalized recommendations, monitoring the supply chain, and enhancing demand analytics.

In physical retail, you use synthetic data to:

• Simulate Customer Behaviors: simulation of in-store shopping patterns, product engagement, and impact of marketing campaigns.

• Sales Forecasting Accuracy: Predict future trends and consumer demand to help retailers optimize in-store stock levels, product placements, promotions, and pricing.

• Store Layout Optimization: Test different store layouts, display fixtures, and product placements to see how they impact the shopping experience.

• Security & Fraud Detection: Train systems that detect fraudulent activities and security risks for effective store operation.

• Employee Performance Analysis: Creating data to evaluate and improve employee performance, customer service, and overall store efficiency.

Let’s dig deeper.

Advanced Store Layout Optimization Using Synthetic Data

Courtesy of Preplexity.ai, below are three highlighted examples of deploying synthetic data in technology systems for physical retail :

1. Customer Flow Simulation

Detailed simulations of in-store customer movements.

• Heat mapping: Generating synthetic heat maps to visualize high-traffic areas
• Path analysis: Simulating customer journeys through the store
• Dwell time prediction: Estimating how long customers spend in different sections

For example,

Sephora has used advanced analytics to optimize its store layouts, which has resulted in a 15% increase in customer engagement with products.

2. Product Placement Optimization

Synthetic data is being used to test product placement strategies:

• Cross-selling opportunities: Simulating placement of complementary products
• Impulse buy optimization: Testing locations for high-margin impulse items
• Category management: Optimizing the arrangement of product categories

For example,

Walmart has reported a 3% increase in sales for specific categories after using data-driven layout optimization.

3. Personalization at Scale

Retailers are using synthetic data to create personalized in-store experiences:

• Digital signage optimization: Testing placement and content of digital displays aligned with consumer behaviors.
• AR/VR integration: Simulating customer interactions in Digital Twins simulations.
• Mobile app integration: Optimizing in-store app experiences based on layout

For example,

After optimizing its in-store integration, Macy's has seen a 60% increase in customer engagement with its AR furniture placement app.

Limitations

At the same time, you must consider the limitations of current models.

1. Scene Complexity: Physical retail environments are highly dynamic and complex. Current synthetic data models often struggle to accurately capture the full range of in-store scenarios, including:

• Fluctuation in lighting conditions throughout the day
• Complex product arrangements on shelves
• Occlusions where products partially obscure one another

2. Customer Behavior Simulation: Accurately modeling customer interactions within a store is challenging. Synthetic data models may not fully capture the following:

• Realistic customer movement patterns
• Product interaction behaviors
• Group shopping dynamics

3. Significant costs: Implementing synthetic data solutions can have significant cost implications.

• Development: Depending on complexity, custom AI solutions can cost $20,000 to over $500,000.
• Integrations: Integrating synthetic data solutions into existing systems can be expensive and time-consuming.
• Training and support: Costs include employee training and ongoing system support, updates, and performance monitoring.

There is a lot of potential:

​The ROI from AI projects using synthetic data is 5.9% on average, with top performers reaching up to 13%.​

The challenge in applying synthetic data solutions to physical retail environments is the model dependence caused by the nuances of in-store operations and customer behaviors.

🤩*Synthetic Data* for Store Optimizers

Synthetic data allows Store Optimizers to test changes with minimal disruption to the actual store operations, providing a low-risk way to innovate and optimize store performance.

Where should you start?

Start Small

Generating synthetic datasets helps design tests with small samples (Technically, non-random quasi-experiments).

For example,

Here’s a scene from a chaotic apparel store.

An in-store experiment needs to account for the context of the local environment, such as the purchase funnel, traffic flow, product positioning, and shopping behaviors.

For simplicity's sake,

Say you want to identify the value of having a full-length mirror installed in a specific location in the department. 🧐

You can narrow the test to the single metric of Stay Time, which is how long shoppers stood in front of the mirror within the virtual mirror zone.

I assumed:

• The ‘Optimal Stay Time Mirror Zone’ is 45 to 180 seconds.
• The number of ‘Visitors in Area Zone’ is between 5 and 47 visitors.
• There is a correlation between the number of people who stood in front of the mirror for an optimal stay time and the area's sales conversion rate.

Sample Dataset for Mirror

The following sample of synthetic dataset illustrates the relationship between the three key metrics:

• Mirror_Capture_Rate: The percentage of people in the Area Zone Passing with Optimal Stay Time in front of the mirror.
• Optimal_Stay_Visitors: The number of visitors who stayed in front of the mirror for the optimal time
• Impact_8, Sales_Impact_7, and Impact_5 represent the number of purchases assuming 8, 7, and 5 purchases, respectively, for every 10 visitors with optimal stay time.

The synthetic data provides a starting point for further investigation and potential strategies to optimize the store layout and operations even when dealing with lower foot traffic.

For example,

Positive Correlation: Mirror Capture Rate and Sales Conversion have a clear positive correlation. As the Mirror Capture Rate increases, we generally see a corresponding increase in Sales Conversion. This suggests that improving the mirror's effectiveness in capturing customer attention could lead to higher sales.

Variability in Effectiveness: The scatter plot shows considerable variability in Sales Conversion for similar Mirror Capture Rates. This indicates that other factors influence sales conversion while the mirror is generally effective. These could include product appeal, pricing, or overall store layout.

High-Impact Outliers: Some data points show exceptionally high Sales Conversion rates despite average Mirror Capture Rates. These outliers suggest that there might be specific conditions or days when the mirror's effectiveness is significantly amplified, possibly due to factors like promotions, weekend traffic, or seasonal trends.

🤣🤣🤣To get the retailer to play, you calulate the impact on sales.

Adding the Average Order Value of $23 to the sample dataset indicates that improving the relationship between shoppers trying out clothes using the mirror and purchasing the items from 5:10 to a 8:10 ratio will increase sales from $432 to $692 (60%).

WOW

Synthetic data helps store optimizers better understand in-store staff and consumer behaviors and empowers the store.

Lean in,

Ronny

​

PS. If you’re a serious Store Optimizer, join me and other top experts at the conference Experimentation Island.

Save $250 with my discount code RONNY-250.