Qlik has documented the architecture behind Qlik Answers, its enterprise AI service built on Amazon Bedrock, with the aim of giving employees sourced responses from both structured analytics and unstructured company content. The design affects Qlik Cloud users by routing each question to an appropriate data and reasoning path instead of relying on a single general-purpose assistant.
According to the AWS Machine Learning Blog case study, Qlik Answers became generally available in February 2026 for a customer base of more than 40,000 organizations. Qlik says its Discovery Agent, a separate anomaly- and outlier-detection agent, has surfaced more than 100,000 discoveries since launch, while a majority of Qlik Cloud accounts with agentic tools enabled are actively using them.
Separating routing from answer generation
Qlik’s central architectural choice is to divide the service into layers with distinct responsibilities. A stable conversational entry point in Qlik Cloud feeds a lightweight routing layer, which reads the user’s message and conversation context to select a destination. The router is intended to make a fast classification decision rather than solve the underlying request.
Once routed, the answer layer can choose a fast path for simpler requests or a more deliberate path that breaks a question into sub-questions and gathers material from multiple sources. Behind it, a shared specialist-agent runtime manages tools, state and human-in-the-loop steps. Qlik says this common runtime lets it add specialist agents without creating separate orchestration systems for each one.
The service also distinguishes structured and unstructured work. Questions about analytics are sent to an app-aware conversational analytics path, while documents and knowledge-base content are indexed and searched through Amazon OpenSearch Service. Answers can draw on app metadata, retrieved content, glossary definitions, automation context and documents used for summarization.
Grounding and controls sit at the model gateway
Qlik routes chat, streaming, embeddings and reranking through its own LLM gateway before accessing Amazon Bedrock. The gateway applies Amazon Bedrock Guardrails to every request and response, including filters for prompt injection, personally identifiable information, secrets and denied topics.
For document and knowledge-base questions, Qlik says it carries question-specific retrieved material into the final response so users can receive citations. It also uses Bedrock Guardrails’ contextual grounding capability to compare generated answers against source content. That is a meaningful distinction in the implementation: retrieval provides context, while the separate validation step is intended to check whether the resulting answer is grounded in it.
Capability controls operate at the tenant level, allowing Qlik to selectively enable functions such as glossary lookup, automations and document summarization. The company says this permits the available orchestration path to be rebuilt for a particular request without changing the default experience across all customers.
Regional deployment and capacity planning
Qlik says Amazon Bedrock cross-Region inference allows it to serve 11 Regions while retaining data where residency requirements apply. Its stated challenge was avoiding both a single global deployment that could conflict with sovereignty expectations and 11 independently maintained Regional builds.
If a customer-dependent model is not yet available through Bedrock in a required Region, Qlik uses Amazon SageMaker AI as an in-Region fallback and moves the workload back to Bedrock after Regional availability catches up. Its LLM gateway also decouples application logic from a particular model, allowing teams to alter model selection by task without rewriting the surrounding application.
The company forecasts token consumption and model availability by feature and Region three to six months before major launches, then compares forecasts with actual usage after rollout. Qlik presents that process as a capacity-management practice for adoption growth, rather than as a benchmark of model performance.
Reported deployments and remaining caveats
Qlik cites customer results that illustrate the intended use cases. Lintech International indexed more than 17,000 technical documents and reported response times 75% faster, returning up to seven hours a week to business managers. Bystronic deployed an AI chatbot in 15 minutes for sourced access to real-time operations data. TouchPoint Support Services uses Qlik Answers to provide compliance-aligned guidance to 15,000 staff across 650 healthcare sites.
These figures are reported in an AWS-published implementation case study and customer materials, not an independent evaluation. The source describes the system’s filters and grounding-validation controls but does not present measured answer accuracy, safety effectiveness, latency, cost or comparative model results. Qlik is evaluating Amazon Bedrock AgentCore for selected workloads and says it is building a systematic framework for evaluating model performance. Source: AWS Machine Learning Blog
Definition. Qlik Answers is Qlik’s enterprise AI service that routes questions across structured analytics and unstructured company content to generate sourced responses.
| Question type | Qlik Answers path |
|---|---|
| Structured analytics questions | App-aware conversational analytics path |
| Documents and knowledge-base questions | Amazon OpenSearch Service indexing and search with retrieved material for citations |
| Simpler requests | Fast answer path |
| Complex requests | Deliberate path that breaks questions into sub-questions and gathers material from multiple sources |
Key takeaways
- A lightweight router selects the appropriate data and reasoning path using the user message and conversation context.
- Analytics questions use an app-aware conversational path, while document and knowledge-base questions use Amazon OpenSearch Service.
- Qlik's LLM gateway applies Amazon Bedrock Guardrails to requests and responses, including prompt-injection, PII, secret and denied-topic filters.
- Retrieved source material supports citations, while contextual grounding is used to validate answers against that material.
- Cross-Region inference serves 11 Regions, with Amazon SageMaker AI used as an in-Region fallback when needed.
- The reported customer results are not an independent evaluation and do not include measured accuracy, safety, latency, cost or model comparisons.
FAQ
How does Qlik Answers route enterprise questions?
A lightweight routing layer reads the user's message and conversation context, then directs it to the appropriate analytics, document or reasoning path.
How does Qlik Answers handle grounded responses?
For document and knowledge-base questions, Qlik carries retrieved material into the final response for citations and uses contextual grounding to compare the generated answer against source content.
What safety controls does Qlik apply?
Qlik routes chat, streaming, embeddings and reranking through its LLM gateway, which applies Amazon Bedrock Guardrails to every request and response.
What are the limits of the reported results?
The cited case study describes customer results and system controls, but it does not provide measured answer accuracy, safety effectiveness, latency, cost or comparative model results.
