Using AI to Solve Tedious Things About Sampling

Quick insights into 2 of our AI products and our technical approach!

AI is everywhere, and we have gradually integrated it across our products to enhance their capabilities. Our mantra is “Solve the tedious things first!”

This blog attempts to share how we are approaching building and implementing AI-powered features to invent the future of FMCG customer engagement and sampling.

Product 1: Planning Agent

Product sampling is an omnichannel activity, you can distribute your samples at offline retail stores, via quick or e-commerce platforms or via promoters. FreeStand has over 50 sampling destinations with over 1Million+ unique setup combinations.

To help a marketer navigate this quickly, we have created a “Planning AI. " Its primary focus is to help a marketer set up an entire sampling campaign with hundreds of thousands of samples simply by using natural language.

This alone has been a huge bottleneck for marketers at FMCG companies of all sizes. A well-planned product sampling campaign can be ten times more successful.

Below is a screenshot of what it looks like:

FreeStand Planner AI Agent in Action

Product 2: Validation Agent

India is a complex country with a huge socio-economic divide. Brands often give away product samples to their customers for free; it would be helpful if you could guarantee that the samples reach only the most desirable customers.

FreeStand’s “Validation agent” attempts to improve customer and data quality across all product sampling workflows.

The validation agent has several capabilities, including suggesting the purchasing capacity for a given address and facilitating warehousing and inventory operations.

FreeStand Validation AI Agent at Work

The technology behind the builds:

We are leveraging OpenAI’s Assistants API, tool calling, and custom functions to develop our Planner agent.

The Planner agent uses OpenAI’s API to interpret user inputs and execute complex workflows. Tool calling enables seamless integration with external functions customised to perform specific tasks. These custom functions enable the agent to handle specialised queries and access on-demand customer data, ensuring highly accurate and contextually relevant results.

For the Validation agent, we are utilising vector embeddings in combination with a PostgreSQL database enhanced by the pg-vector extension. By integrating OpenAI’s embedding model with the vector database, we efficiently store, search, and retrieve high-dimensional embeddings, enabling a robust similarity matching and validation process.

Conclusion

It is still early days for integrating AI capabilities into our offering, but we are thrilled to see the benefits. From a technical perspective, we feel that we should restructure our backend databases to be more optimised for the additional data load that AI tends to bring.

Select customers with whom we have shared our offerings have been great design partners during our development process, and we thank them wholeheartedly.

We are excited to share more about what we build in the future. Cheers!