How Qlik built grounded, enterprise-scale AI with Amazon Bedrock
Amazon Web Services Sunil Yerkola
Qlik built an AI answer tool on Amazon Bedrock that gives sourced replies, not just guesses. It’s aimed at regulated companies that can’t afford black-box nonsense.
Based on reporting by Amazon Web Services, Sunil Yerkola — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Qlik’s pitch here is simple: employees don’t need more data, they need a fast way to ask it something useful and get an answer they can trust. That’s the gap Qlik Answers is trying to close, and it’s why the company built the system around grounded responses instead of a chat box that sounds confident and makes things up.
The scale matters. Qlik says the tool is now in daily use, with general availability in February 2026. Since then, its Discovery Agent has surfaced more than 100,000 discoveries for customers, and the majority of Qlik Cloud accounts with agentic tools switched on are actively using them. This isn’t a lab demo. It’s shipping into places where people are expected to rely on the result.
To get there, Qlik split the product into separate layers rather than stuffing everything into one giant assistant. A routing layer decides where a question should go. An answer layer picks between a quick response and a more deliberate path that can break a question into pieces and pull from multiple sources. Specialist agents, conversational analytics, retrieval, and the model access layer all do different jobs, which keeps the system from slowing down every time someone adds another capability.
The retrieval side runs on Amazon OpenSearch Service, while the model access layer reaches Amazon Bedrock through Qlik’s own LLM gateway. That layer also applies Amazon Bedrock Guardrails on every request and response, including content filtering for prompt injection, PII, secrets, denied topics, and a grounding-validation check on generated answers. If a model isn’t yet available in a customer’s Region through Bedrock, Qlik uses Amazon SageMaker AI until the Bedrock option shows up in that Region.
The company had to solve sovereignty and capacity at the same time, which is where Bedrock’s multi-Region setup comes in. Qlik says it serves customers across Europe, Asia Pacific, and the Americas, and it needed a deployment model that could respect data residency expectations without turning into 11 separate systems to maintain. It also forecasts model capacity 3–6 months ahead of major launches, then checks real usage against those projections after rollout. That’s not glamorous, but it’s how you keep a product like this from collapsing under its own popularity.
The customer examples show why the architecture matters. Bystronic deployed an AI chatbot in 15 minutes. Lintech International indexed more than 17,000 technical documents and cut manual research time while speeding responses by 75 percent, giving business managers back as much as 7 hours a week. TouchPoint Support Services is using it to help 15,000 staff across 650 healthcare sites. The pattern is obvious: the winner here isn’t the flashiest model, it’s the system that can stay boring, safe, and useful in regulated places.
My take — AI-written commentary, not fact-checked reporting
This is the right way to do enterprise AI: keep the model behind a gate, force it to show its work, and don’t pretend one giant assistant should handle everything. The industry loves “autonomous” agents until someone asks where the answer came from, then suddenly everyone discovers the joy of guardrails. Qlik gets credit for building for regulated customers instead of for demo-day applause.
Read more about this at: Amazon Web Services