Qlik builds grounded enterprise AI assistant on Amazon Bedrock
Qlik has detailed how it built Qlik Answers, a natural-language assistant that returns sourced, checked answers to enterprise questions, using Amazon Bedrock as its model layer. According to an AWS blog post, the tool has been generally available since February 2026 and serves customers across 11 Regions.
- Qlik Answers has been generally available since February 2026 and serves customers across 11 Regions.
- It uses Amazon Bedrock via Qlik's own LLM gateway, with OpenSearch Service for retrieval and SageMaker AI as an in-Region fallback.
- Bedrock Guardrails' contextual grounding check validates each generated answer against its sources.
- Qlik's Discovery Agent has surfaced more than 100,000 discoveries since launch, according to the post.
- Lintech International reports 75 percent faster response times after indexing more than 17,000 documents.
The problem Qlik set out to solve
According to the AWS post, employees are not short on data but on a fast way to ask a real question of it and trust the answer. Analysts spend hours searching documents, dashboards and institutional memory. Organizations that turn to generative AI often hit the same obstacles: answers without sources, no way to verify results, and no path to deploy in a regulated environment.
Qlik, which describes itself as a global leader in data integration, data quality, analytics and AI, built Qlik Answers for its more than 40,000 customers worldwide. It identified three problems to solve at once:
- Specialized reasoning without slowdown. Questions range from quick lookups to structured analytics, documents, glossary definitions or automations. A single general-purpose assistant tends to get slower and less accurate as capabilities are added.
- Data sovereignty across Regions. Customers in Europe, Asia Pacific and the Americas have different data residency expectations. One global deployment would not satisfy them, while 11 separate Regional builds would be hard to maintain.
- Capacity forecasting. Token consumption and model availability had to be planned 3 to 6 months before major launches.
A layered architecture instead of one large assistant
Qlik built the system around clear boundaries. An entry layer inside Qlik Cloud stays stable, so backend capabilities can change without altering the user experience. A lightweight routing layer reads the message and conversation context and decides where the request goes. The answer layer then chooses between a fast path for simple requests and a more deliberate path that splits a question into sub-questions and draws on multiple sources.
A shared swarm runtime defines how specialist agents, tools, state and human-in-the-loop steps work, so new specialists can be added without each inventing its own orchestration. Structured data questions go to a conversational analytics path built for analytics rather than general text generation. Unstructured document indexing and retrieval runs on Amazon OpenSearch Service.
The model access layer connects through Qlik's own LLM gateway to Amazon Bedrock for chat, streaming, embeddings and reranking. If a model a customer relies on is not yet available in-Region through Bedrock, Qlik hosts it on Amazon SageMaker AI and later moves the workload back to Bedrock.
Grounding as a checked control
The answer layer pulls from structured app metadata, knowledge base content, glossary definitions, automation context and document content. Qlik also runs a grounding-validation check that compares each generated answer with the source content it drew from, using the contextual grounding capability in Amazon Bedrock Guardrails. According to the post, this turns grounding from a retrieval step into a checked production control.
Guardrails are applied to every request and response, with content filtering for prompt injection, personally identifiable information, secrets and denied topics. For unstructured questions, retrieved content is carried forward so the final answer includes citations. Capability controls and tenant-specific options let Qlik switch features on selectively while the architecture stays consistent.
Why Bedrock, and what customers report
Qlik cites three reasons for choosing Bedrock. Multi-model flexibility lets it assign the best-suited model to each agent's task. Cross-Region inference lets it serve all 11 Regions while keeping data resident where compliance requires. Capacity reservation options and the Regional deployment model give headroom to scale, and services such as Amazon Bedrock AgentCore could later take on more operational overhead for specific workloads. Because Qlik goes through its own gateway, product teams can change model choice by task without rewriting the surrounding application.
For capacity, the team built a model projecting token consumption by feature and Region, then checked actual usage against it after each rollout.
The post reports adoption figures since general availability in February 2026. Qlik's Discovery Agent, an anomaly and outlier detection agent, has surfaced more than 100,000 discoveries, and the majority of Qlik Cloud accounts with agentic tools turned on are actively using them. Named customers include:
- Bystronic, which deployed an AI chatbot in 15 minutes.
- Lintech International, which indexed more than 17,000 technical documents and, according to the post, sped up response times by 75 percent, giving business managers back as much as 7 hours a week.
- TouchPoint Support Services, which serves 650 healthcare sites and uses Qlik Answers to give compliance-aligned guidance to 15,000 staff.
What it means for enterprise AI buyers
The case shows one approach to the trust and compliance barriers that stall enterprise generative AI projects: route requests to specialists, validate answers against sources, and keep the model layer swappable. For organizations in regulated sectors, the combination of citations, guardrails and Region-aware deployment is the part the source highlights as making production use possible.