Consumer Products Company: An AI Assistant Trained on Its Own Data

Our client

A consumer products company wanted employees across marketing, sales, and support to use AI for everyday work: drafting content, answering common questions, preparing investor summaries. Off-the-shelf ChatGPT could handle some of this, but not the tasks that depended on real company knowledge.

Challenge

Generic ChatGPT often invented answers when it needed internal information it didn’t have, or required so much manual effort gathering source material that it stopped saving anyone time. On top of that, leadership was wary of feeding sensitive company data into a public AI tool. Together, this made staff cautious about using AI at all, undercutting the reason they adopted it in the first place.

Solution

There were three primary goals for our solution:

1. Allow staff to use an LLM to create documents without requiring significant work gathering the information needed.
2. Ensure that documents and answers generated by the LLM do not include incorrect data.
3. Address any security concerns around data leakage.

We built a private AI assistant that automatically pulls the right internal information before answering a question, instead of relying on the AI’s general knowledge alone. That kept every answer grounded in the company’s own documentation rather than invented, while keeping the entire system on a private, secured endpoint so no company data left the building. We started by studying the actual questions employees asked, then built the system to tell the difference between a quick factual lookup and a request that needed real judgment, like an investor summary, and respond accordingly.

Results

An AI Assistant Employees Could Actually Trust.

Staff could safely use the new assistant to answer internal questions, draft documents, and support customer inquiries, without the invented answers or security concerns that made the original approach unusable. It also meaningfully lightened the workload for the sales and marketing teams.

Technical Details

Built on Microsoft Azure using retrieval augmented generation, a private Azure OpenAI endpoint, an Azure Function to classify and route queries, and an Azure Cognitive Search index of the client’s internal documentation with semantic ranking enabled. Delivered through a React-based web interface, with the full pipeline stored and deployed via Azure DevOps.

What’s Next

Automating the currently manual process of indexing new documentation, and integrating the assistant directly into existing internal tools, including their marketing email platform, so employees no longer have to copy information back and forth by hand.

Want Your Team Using AI Without the Guesswork?

If your staff are hesitant to trust AI with real company knowledge, we can help you build something grounded enough to trust.