Pixel Assist
How Pixel Agency are deploying Pixel Assist, an AI knowledge assistant, on a client's website to surface product knowledge in real time for their customers, with a rollout that proved the system out on a subset of users before scaling.
An AI knowledge assistant for a Pixel Agency client, rolling out the careful way.
The client
The client is a category leader in the Australian manufacturing industry, with a long history of market-leading products and a deep well of technical knowledge to support them. They have an ongoing partnership with Pixel Agency across discovery, design, and development work, and are always looking for ways to use technology to make their customers' lives easier.
The client has a large and diverse audience. They come to the client's website to find technical information that helps them make decisions about which products to use for specific applications. The client is committed to providing that information, but the depth and breadth of it can make it hard for users to find what they need quickly.
We're not naming the client here because they see Pixel Assist as a genuine differentiator from their competition. They're planning to surprise the market with this release in the new financial year, and want to maintain that competitive advantage until launch.
The problem
The depth of information on the client's website is part of why their customers come to it. It's also part of what gets in their way. Users trying to find the right product for a specific application have to hunt across HTML pages, PDF documents, and images to piece an answer together.
Slow answers cost the client in two ways: support requests pile up where users can't self-serve, and a buyer who can't quickly find the right product is more likely to drift to a competitor whose information is easier to surface. With product specifications increasingly driven by performance, compliance, and time pressure, frictionless decision-making isn't a nice-to-have.
The approach
We brought Pixel Assist to the client as a new addition to the ongoing partnership. Pixel Assist is our managed AI knowledge assistant: a RAG-based system that learns from a client's existing knowledge, with no fine-tuning required, and answers user questions in real time in the client's tone of voice. We started with discovery, working with the client to map the knowledge that would feed the system, the kinds of questions users actually need answered (product fit, install methods, application suitability), and the brand voice the assistant would need to use so every answer felt unmistakably the client's.
The solution
Pixel Assist was deployed on the client's website with a familiar chat interface, but the deployment itself was where the care went in. Rather than turn the assistant on for every visitor at once, we used feature flagging to expose Pixel Assist to just 2% of users. That gave us a controlled environment to validate the system in the wild without putting the wider user base at risk if anything needed adjusting. The assistant answers user questions in real time, suggests the client's products that fit specific needs, and is built to support the various customer types in real time, helping them find what they need quickly.
The results
The 2% trial validated what the system could do on the three dimensions the client cared about:
A measured fall
in customer support requests from users who reached the assistant
Questions answered to a high quality
with the client's tone of voice preserved across responses
No incidents
on the wider user base during the test period, thanks to the feature-flagged exposure
A proven system
ready to scale, with confidence on both quality and brand fit
AI doesn't have to be all or nothing. The smart way in is the careful way.