Best practices for creating and managing assistants

Plan governance of data sources

Successful assistants depend on governance of their data sources. Ensure owners of content used by the assistant are responsible for maintaining their content and keeping it up to date.

Curate information sources

Curate your data sources to be as specific to the use case as possible. Do not add unrelated knowledge bases to assistants. Qlik Answers works best when it is finding specific information and providing a summary. It is not as effective when there are a lot of potential answers and you are expecting an aggregation or different answer each time. Minimize repetitive content and try to include single versions of possible answers.

Qlik Answers does not generate its own rankings or aggregations, but it can present them if such summaries already exist in your data sources. If you think your users need that information, add them to the sources used by your assistants.

Qlik Answers does not select or prioritize information sources by age or filtering, just semantic relevance. An old document in a knowledge base is treated just as relevant as a new document. Update your source content and knowledge bases as newer information becomes available. Dynamic filters in data connections can help exclude older content that may no longer be relevant or accurate.

Make specific and targeted assistants

Create assistants that are unique and targeted to specific purposes. Your end users need to know that they have single points to go to for information on a specific subject. Identify your use cases, identify the right people to build and maintain the content, and the right people to use the content. Do not create multiple assistants that overlap in the questions they can answer.

Create conversation starters for users. These can cover commonly asked questions for your users. They can also help guide users as to both the kinds of questions the assistant can answer and how users can ask their questions. For example, instead of just having a question that directly asks what something can do, such as What can the TRIM() function do?, instead show how users can ask a question that describes the problem they are trying to solve, such as When I read text line by line, what function should I use to remove extra spaces?.

Similarly, ensuring your assistant has a helpful description and welcome message can also help users orientate themselves to the kinds of questions your assistants can answer.

Review questions

Provide your subject matter experts with access to assistants to review answers. Subject matter experts can help assess if the assistant is meeting user needs and if the data used by the assistant is serving the right purpose. If not, update your data sources to better meet user needs.

Reviewing your users' answers is key for content improvement. Answers are unique each time, with no specific caching, so review similar questions to see how similar the answers are.

Validate content in knowledge bases

When creating knowledge bases, you should validate their contents. Assign the new knowledge base to an assistant as the only knowledge source. Then, start asking questions to which you know the answer and that should be answerable using the data sources in the knowledge base. For example, in a Qlik Cloud knowledge base, you might ask:

• Give me an introduction on Qlik Cloud and list the top 10 features

• What is Qlik-embed for and how do I set it up?

• Create a list of 10 keywords glossary for AutoML

Validate each response to confirm that your data sources have been properly added and indexed by the knowledge base.

Keep your knowledge bases specific in scope

When creating knowledge bases, limit the scope of data sources within a single knowledge base to a specific category or domain.

A narrow knowledge base scope keeps the information sources focused on the specific subject, which in turn helps assistants answer questions on the subject more easily. With specific knowledge bases, it is also easier to tailor the exact information available in assistants. It also makes it easier for curators to manage and validate content from a knowledge base when it has a specific and defined scope.

Additionally, when you have specific knowledge bases, if it easier to control user access to specific categories or domains of information. By putting each knowledge base in a separate space, you can tailor user access to these sources using separate permissions in each space. Even if an assistant has all possible knowledge bases, users will only get answers using knowledge bases that their space permissions allow them to access. This allows a single assistant to offer different users different answers based on what information it is appropriate for them to access.