AI Customer Behaviour Analysis for Retail
Understand Customer Behaviour Across Every Retail Touch Point
Imagine the salesperson already knows what you will buy
Imagine the salesperson knows in advance what the customer is going to buy, and can also see stock across all locations. The conversion rate is then at least 20% higher than it was without AI-driven predictive analytics.
Retail Clouds, with its deep roots in R&D, developed and trained an AI model for customer behaviour analysis.
We integrated the behavioural results at every touch point where customers engage. This matters most on an omnichannel software platform, where sales happen through multiple channels.
One model, every touch point
At the store
The salesperson uses an Android or iOS tablet to get real-time predictions about the customer's behaviour, and uses them to convert the sale. AI-Powered Customer Behaviour Intelligence, Embedded into Retail POS.
On the eCommerce site
When the same customer uses eCommerce site, the site also gets inputs from the behavioural data. It decides what to display next and how to guide the customer to complete the transaction.
On the phone
The same applies when the customer uses a phone to make online purchases.
How we built it: four stages
Data extraction, cleaning and loading
Customer history from all sales channels sits in a MySQL database. We used SQL and Pandas for this layer.
Customer embedding and training
We used PyTorch for customer behaviour events and embedding, with a Transformer for the customer embedding. The model was trained on data we generated for fashion customers over a period of time.
Predictions
From the embedded vectors we built the predictions below. A single Transformer handles them, instead of multiple neural networks such as MLPs.
Integration with sales channels
We created a RESTful API so the predictions are available wherever they are needed.
What the model predicts
- Purchase prediction
- Customer segmentation
- Churn prediction
- Product recommendation
- Next-best-offer prediction
- Customer lifetime value
- Personalized campaigns
Modern Retail Software must have the capability to capture the customer's behavior and give the predictions. How you know that the AI model is well developed and trained? Ask for Evaluation license and validate the performance of the AI Model.
Under the hood: PyTorch
PyTorch turns raw customer events into tensors, feeds them through the embedding model, and trains it by repeating a loop: predict, measure the error, adjust the weights.
Under the hood: the Transformer
A Transformer reads a customer's events as a sequence. Self-attention lets every event look at the others, so a late purchase can be understood in light of what the customer browsed earlier. The result is one customer vector, and a single model reads it to produce every prediction.
Use cases
Once the customer behaviour model is trained and ready, it can be used across sales channels. Here are some of the ways we apply it.
Outlet & POS
The salesperson can see customer behaviour and predictions. This improves the customer experience and the conversion ratio.
Campaign management
We can pick the right campaigns based on customer segmentation and behaviour. This brings customers back and grows the top line.
Online & customer journey
We can show the right products at the right price, so the customer can decide to buy.
See it in action
Watch the model predict customer behaviour across store, web and mobile.